WO2013108686A1 - 情報処理装置および方法、並びにプログラム - Google Patents
情報処理装置および方法、並びにプログラム Download PDFInfo
- Publication number
- WO2013108686A1 WO2013108686A1 PCT/JP2013/050209 JP2013050209W WO2013108686A1 WO 2013108686 A1 WO2013108686 A1 WO 2013108686A1 JP 2013050209 W JP2013050209 W JP 2013050209W WO 2013108686 A1 WO2013108686 A1 WO 2013108686A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- person
- information
- image
- likelihood
- moving body
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/215—Motion-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/292—Multi-camera tracking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
Definitions
- the present technology relates to an information processing apparatus, method, and program, and in particular, an information processing apparatus, method, and program that can search and track a person who wants to search and track with high accuracy from images captured by a plurality of cameras. About.
- a system for tracking moving objects in a plurality of camera images and a system for searching are proposed.
- a method of tracking a person in a single camera image and linking it with a plurality of cameras has been proposed (see Patent Document 1).
- the foreground area in the captured image is divided into blocks, similar image search is performed in block units, and whether or not the same person is determined by the matching result Has been proposed (see Non-Patent Document 1).
- JP 2006-245795 A Human tracking report using similar image retrieval in sparse distributed camera environment, vol. 110, no. 330, PRMU2010-130, pp. 25-30, 12December 2010.
- Non-Patent Document 1 in order to determine whether or not all foreground areas divided into blocks are similar, as the shooting time increases, the processing time increases and the accuracy increases. There was a risk of lowering.
- the present technology has been made in view of such a situation.
- images that are not suitable for search are deleted, and a user searches for search processing results. It is possible to improve the accuracy of the person search and tracking process by repeating the process of correcting the search process after identifying an appropriate image as the target image.
- the information processing device captures an image, detects a moving object, extracts a moving object image including the detected moving object image, and based on the moving object image, spatial position coordinates of the moving object
- a plurality of imaging units that output the moving body information including the moving body image, the spatial position coordinates of the moving body, and the imaging time when the image is captured, and the moving body information including the moving body image of the moving body to be searched
- a moving object image likelihood calculating unit that calculates a moving object image likelihood that is a likelihood of a moving object image included in moving object information other than the search contrast moving object information with respect to a moving object image of certain search target moving object information; and the moving object image likelihood calculation It is determined whether each of the moving image likelihoods calculated by the unit is higher than a predetermined threshold value, and the moving image information of the moving image likelihood higher than the predetermined threshold value is determined as a moving image of the search target moving object information.
- a moving body image threshold determination unit that searches as moving body information that is moving body information including a moving body image of the same moving body as the body, and a search result that stores the moving body information searched as the moving body information searched as the search result moving body information by the moving body image threshold determination unit
- An operation input unit that receives input of information, and of moving body information other than the fixed moving body information that is the moving body information in which the confirmed information is input among the moving body information stored as the search result moving body information in the result moving body information storage unit.
- a spatio-temporal likelihood for calculating a spatio-temporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time for the definite moving body information to which the definite information is input It is determined whether each of the spatiotemporal likelihood calculated by the calculation unit and the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and moving body information having a spatiotemporal likelihood lower than the predetermined threshold is obtained. And a spatio-temporal likelihood threshold determination unit to be deleted from the search result moving body information storage unit.
- the spatiotemporal likelihood threshold determination unit After determining whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold by the spatiotemporal likelihood threshold determination unit, Of the moving body information stored as the search result moving body information in the result moving body information storage unit, based on the moving body image, the user inputs confirmation information for newly confirming that it is the search target moving body information.
- the spatio-temporal likelihood calculation unit the fixed information of the moving body information other than the fixed moving body information among the moving body information stored as the search result moving body information in the result moving body information storage unit is newly added.
- a spatiotemporal likelihood composed of likelihoods based on the spatial position coordinates and the imaging time is newly calculated for the input confirmed moving body information, and the spatiotemporal likelihood threshold determination unit calculates the spatiotemporal likelihood. It is determined whether or not each of the spatiotemporal likelihood newly calculated by the above is lower than a predetermined threshold, and moving body information having a spatiotemporal likelihood lower than the predetermined threshold is obtained from the search result moving body information storage unit.
- the operation input unit, the spatiotemporal likelihood calculation unit, and the spatiotemporal likelihood threshold determination unit cause the same processing to be repeated each time new confirmation information is input by the operation input unit. Can be.
- the moving body information may further include an ID for identifying any of the plurality of imaging units that captured the moving body image included, and includes a moving body image of the moving body to be searched.
- ID for identifying a plurality of image capturing units that have captured respective moving body images from among the search target moving body information that is information and the confirmed moving body information for which the input of the deterministic information that determines that the search target moving body information is accepted BTF calculation unit for calculating BTF (Brightness Transfer Function) for correcting the color change between the imaging units based on the two different moving body images, and the result moving body information storage unit stores the search result moving body information.
- BTF processing that applies BTF to a moving body image of moving body information including a moving body image captured by an imaging unit having an ID for which the BTF is required among moving body information other than the determined moving body information.
- a BTF moving body image for calculating a BTF moving body image likelihood including a likelihood based on the moving body image of the moving body information including the moving body image subjected to the BTF by the BTF processing section with respect to the moving body image of the fixed moving body information
- An image likelihood calculating unit and a BTF moving image threshold determining unit that determines whether each of the BTF moving image likelihood calculated by the BTF moving image likelihood calculating unit is lower than a predetermined threshold.
- the spatiotemporal likelihood threshold value determination unit determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold value. If not lower than the predetermined threshold, the BTF moving image threshold determination unit determines whether each BTF moving image likelihood calculated by the BTF moving image likelihood calculation unit is lower than a predetermined threshold. Whether or not When the BTF moving image likelihood is lower than a predetermined threshold, moving information including moving images whose BTF moving image likelihood is lower than the predetermined threshold is deleted from the search result moving information storage unit. Can do.
- the moving image likelihood calculation unit includes a moving image of moving object information that is moving object information including moving object images of moving objects to be searched, and a moving image included in moving object information other than the search contrast moving object information. On the basis of the moving body image included in the image, a similarity indicating how similar each moving body is can be calculated as the moving body image likelihood.
- the spatio-temporal likelihood calculation unit calculates the distance between the spatial position coordinates of moving body information other than the fixed moving body information and the fixed moving body information to which the fixed information has been input as an average human moving speed.
- the spatiotemporal likelihood can be calculated from the relationship between the required time when moving and the time between imaging times.
- An information processing method includes imaging an image, detecting a moving object, extracting a moving object image including the detected moving object image, and spatial position coordinates of the moving object based on the moving object image.
- an information processing method of an information processing apparatus including a plurality of imaging units that output moving object information including the moving object image, the spatial position coordinates of the moving object, and the imaging time when the image is captured
- a moving image likelihood calculation that calculates a moving image likelihood that is a likelihood of a moving image included in moving object information other than the search contrast moving object information with respect to a moving image of search target moving body information that is moving object information including a moving object moving image.
- a moving object image threshold determination process for searching for information as search result moving body information that is a moving body information including a moving body image of the same moving body as a moving body image of the moving body image of the search target moving body information;
- the moving body information stored as the search result moving body information in the search result moving body information storage process in the search result moving body information storage process for storing the moving body information searched as information, based on the moving body image, by the user,
- the confirmation information is input.
- the spatial position coordinates and imaging of the moving body information other than the fixed moving body information which is moving body information with respect to the fixed moving body information to which the fixed information is input Whether the spatio-temporal likelihood calculation process for calculating the spatio-temporal likelihood composed of the likelihood based on the time and the spatio-temporal likelihood calculated by the spatio-temporal likelihood calculation process are lower than a predetermined threshold value And a spatiotemporal likelihood threshold determination process for deleting moving body information having a spatiotemporal likelihood lower than the predetermined threshold from the moving body information stored by the search result moving body information storage process.
- the program captures an image, detects a moving object, extracts a moving object image including the detected moving object image, and detects a spatial position coordinate of the moving object based on the moving object image.
- a moving object that calculates a moving object image likelihood that is a likelihood of a moving object image included in moving object information other than the search contrast moving object information with respect to a moving object image of search object moving object information that is moving object information including a moving object image of a moving object to be searched It is determined whether each of the image likelihood calculation step and the moving image likelihood calculated by the processing of the moving image likelihood calculation step is higher than a predetermined threshold.
- Moving body image threshold determination for searching moving body information having a moving body image likelihood higher than the predetermined threshold as moving body information including moving body images of the same moving body as the moving body image of the moving body image of the search target moving body information.
- a search result moving object information storing step for storing the moving object information searched as the search result moving object information by the processing of the moving object image threshold determination step, and a search result moving object information stored by the process of the result moving object information storing step.
- the spatio-temporal likelihood calculation step comprising the likelihood based on the spatial position coordinates and the imaging time with respect to the definite moving body information to which the deterministic information is input, and the spatio-temporal likelihood calculation step of body information It is determined whether or not each of the calculated spatiotemporal likelihoods is lower than a predetermined threshold, and moving body information having a spatiotemporal likelihood lower than the predetermined threshold is stored by the processing of the search result moving body information storage step. And causing the computer to execute a spatiotemporal likelihood threshold determination step to be deleted from the moving object information.
- the information processing apparatus captures an image, detects a person, extracts a person image including the detected person image, and based on the person image, the spatial position coordinates of the person
- a plurality of imaging units that output personal information including the person image, the spatial position coordinates of the person, and the imaging time at which the image was captured, and person information including a person image of the person to be searched
- a person image likelihood calculating unit that calculates a person image likelihood that is a likelihood of a person image included in person information other than the search reference person information with respect to a person image of a certain search target person information; and the person image likelihood calculation It is determined whether each of the person image likelihoods calculated by the unit is higher than a predetermined threshold, and the person information having a person image likelihood higher than the predetermined threshold is used as the person image of the search target person information.
- a search result person information that is person information including person images of the same person as the object, and a search result that stores the person information searched as the search result person information by the person image threshold determination part Confirmation that the user information is determined by the user based on the person image, out of the person information stored as the search result person information in the person information storage unit and the person information storage unit.
- An operation input unit that accepts input of information, and personal information other than the confirmed person information that is the person information to which the confirmed information is input among the person information stored as the search result person information in the result person information storage unit.
- a spatio-temporal likelihood for calculating a spatio-temporal likelihood composed of a likelihood based on the spatial position coordinates and the imaging time for the confirmed person information to which the confirmed information is input It is determined whether each of the spatiotemporal likelihood calculated by the calculation unit and the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and the person information of the spatiotemporal likelihood lower than the predetermined threshold is obtained. And a spatio-temporal likelihood threshold determination unit to be deleted from the search result person information storage unit.
- the spatiotemporal likelihood threshold determination unit After determining whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold by the spatiotemporal likelihood threshold determination unit, Of the person information stored as the search result person information in the result person information storage unit, based on the person image, the user inputs confirmation information for newly confirming that the person information is to be searched.
- the confirmed information of the person information other than the confirmed person information among the person information stored as the search result person information in the result person information storage unit is newly added.
- a spatiotemporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time is newly calculated with respect to the input confirmed person information, and the spatiotemporal likelihood threshold determination unit calculates the spatiotemporal likelihood.
- each of the spatiotemporal likelihood newly calculated by the above is lower than a predetermined threshold, and the person information of the spatiotemporal likelihood lower than the predetermined threshold is obtained from the search result person information storage unit.
- the operation input unit, the spatiotemporal likelihood calculation unit, and the spatiotemporal likelihood threshold determination unit are deleted, and the same processing is repeated each time new confirmation information is input by the operation input unit. Can be made.
- the person information may further include an ID for identifying any of the plurality of imaging units that captured the included person image, and includes a person image of the person to be searched ID for identifying a plurality of image capturing units that have captured respective person images among search target person information that is information and confirmed person information for which input of confirmation information for confirming the search target person information is accepted BTF calculation unit for calculating BTF (Brightness Transfer Function) for correcting the color change between the imaging units based on the two person images having different values, and stored as search result person information in the result person information storage unit Among the personal information other than the confirmed personal information, the BTF processing for applying the BTF to the personal image including the personal image captured by the imaging unit having the ID for which the BTF is required.
- BTF Bitness Transfer Function
- BTF person image likelihood including a likelihood based on the person image of person information including the person image including the person image subjected to BTF by the BTF processing unit with respect to the person image of the confirmed person information
- An image likelihood calculating unit and a BTF person image threshold determining unit that determines whether each of the BTF person image likelihood calculated by the BTF person image likelihood calculating unit is lower than a predetermined threshold.
- the spatiotemporal likelihood threshold value determination unit determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold value. If not lower than the predetermined threshold, the BTF person image threshold determination unit determines whether each of the BTF person image likelihoods calculated by the BTF person image likelihood calculation unit is lower than a predetermined threshold. Whether or not When the BTF person image likelihood is lower than a predetermined threshold, the person information including the person image having the BTF person image likelihood lower than the predetermined threshold is deleted from the search result person information storage unit. Can do.
- Search target person information that is person information including a person image of the person to be searched, and confirmed person information for which input of confirmation information for confirming that it is the search target person information is received as the search target person.
- the same person information holding unit holding the same person information as the person information, and the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculating unit is lower than a predetermined threshold or the BTF person image
- the person information including the person image lower than the predetermined threshold is stored, and the search target person is the other person who holds the other person information that is the person information of the other person.
- a unique feature search unit that selects a unique feature for searching for the person to be searched by learning, and the unique image of the person image included in the person information other than the search reference person information with respect to the person image of the search target person information; Whether or not each of the unique feature likelihood calculating unit that calculates the unique feature likelihood that is the likelihood based on the feature and the unique feature likelihood calculated by the unique feature likelihood calculating unit is lower than a predetermined threshold value
- a unique feature likelihood threshold value determination unit that deletes personal information having a characteristic feature likelihood lower than the predetermined threshold from the search result person information storage unit, and The spatiotemporal likelihood threshold determination unit determines whether each of the spatiotemporal likelihoods newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and lower than the predetermined threshold In this case, the BTF person image threshold determination unit determines whether each of the BTF person image likelihoods calculated
- the other person information holding unit includes the predetermined threshold Personal information with lower unique feature likelihood can be held as other person information that is other person's person information
- the same person information is obtained by learning based on the person image in the person information held in the same person information holding unit and the person image in the person information held in the other person information holding unit.
- Person information having a feature quantity that increases the likelihood of the person image in the person information held by the holding unit and the person image of the person to be searched and held by the other person information holding unit The feature quantity that reduces the likelihood of the person image in FIG. 5 and the person image of the search target person can be selected as the unique feature.
- the unique feature likelihood calculating unit includes a person image of search target person information that is person information including a person image of a person to be searched and a person image included in person information other than the search reference person information.
- the similarity indicating how similar each person is based on the unique feature of the person image included in the image can be calculated as the unique feature likelihood.
- the person image likelihood calculating unit includes a person image of search target person information, which is person information including a person image of a person to be searched, and a person image included in person information other than the search reference person information. On the basis of the person image included in the image, a similarity indicating how similar each person is can be calculated as the person image likelihood.
- the spatio-temporal likelihood calculation unit calculates the distance between the spatial position coordinates between the person information other than the confirmed person information and the confirmed person information to which the confirmed information is input, at an average human moving speed.
- the spatiotemporal likelihood can be calculated from the relationship between the required time when moving and the time between imaging times.
- An information processing method captures an image, detects a person, extracts a person image including the detected person image, and based on the person image, the spatial position coordinates of the person
- An information processing method of an information processing apparatus including a plurality of imaging units that output personal information including the person image, the spatial position coordinates of the person, and the imaging time at which the image was captured,
- a person image likelihood that calculates a person image likelihood that is a likelihood of a person image included in person information other than the search reference person information with respect to a person image of search target person information that is person information including a person image of the person to be It is determined whether each of the degree calculation process and the person image likelihood calculated by the person image likelihood calculation process is higher than a predetermined threshold, and a person having a person image likelihood higher than the predetermined threshold
- a person image threshold determination process for searching for information as search result person information that is a person information including a person image of the same person as the person image of the search target person information;
- the search result person information storage process for
- the spatio-temporal likelihood calculation process for calculating the spatio-temporal likelihood comprising the likelihood based on the imaging time and the spatio-temporal likelihood calculated by the spatio-temporal likelihood calculation process are lower than a predetermined threshold value.
- a spatiotemporal likelihood threshold determination process for deleting personal information having a spatiotemporal likelihood lower than the predetermined threshold and deleting the personal information stored by the search result personal information storage process are lower than a predetermined threshold value.
- the program according to the second aspect of the present technology captures an image, detects a person, extracts a person image including the detected person image, and detects a spatial position coordinate of the person based on the person image. And a program that is executed by a computer that controls an information processing apparatus including a plurality of imaging units that output person information including the person image, the spatial position coordinates of the person, and the imaging time at which the image was captured,
- Person image threshold determination for searching person information having a person image likelihood higher than the predetermined threshold as search result person information that is person information including a person image of the same person as the person image of the search target person information
- a search result person information storage step for storing the person information searched as search result person information by the process of the person image threshold determination step, and a search result person information stored by the process of the result person information storage step.
- Search result by the process of the operation input step for accepting input of confirmation information for confirming that it is the search target person information based on the person image, and the result person information storage step based on the person image Of the person information stored as the person information, the confirmed person information that is the person information to which the confirmed information is input.
- a spatio-temporal likelihood calculating step for calculating a spatiotemporal likelihood composed of likelihoods based on the spatial position coordinates and the imaging time with respect to the confirmed person information for which the confirmed information is input, It is determined whether or not each of the spatiotemporal likelihoods calculated by the processing of the likelihood calculating step is lower than a predetermined threshold, and the person information having a spatiotemporal likelihood lower than the predetermined threshold is obtained as the search result person
- a computer is caused to execute processing including a spatiotemporal likelihood threshold determination step to be deleted from the information storage unit.
- an image is captured, a moving object is detected, a moving object image including the detected moving object image is extracted, and a spatial position coordinate of the moving object is detected based on the moving object image.
- the moving object information including the moving object image, the spatial position coordinates of the moving object, and the imaging time when the image was captured is output, and the moving object of the search object moving object information that is moving object information including the moving object image of the moving object to be searched
- a moving image likelihood which is a likelihood of a moving image included in moving object information other than the search contrast moving object information, is calculated for each image, and whether or not each of the calculated moving image likelihoods is higher than a predetermined threshold value
- the moving body information having a moving body image likelihood higher than the predetermined threshold is moving body information including a moving body image of the same moving body as the moving body image of the moving body image of the search target moving body information.
- the moving object information searched as information and the search result moving body information is stored, and the moving object information stored as the search result moving body information is the search target moving body information by the user based on the moving body image.
- the spatio-temporal likelihood composed of the likelihood based on the spatial position coordinates and the imaging time is calculated with respect to the determined moving body information inputted, and whether or not each of the calculated spatio-temporal likelihood is lower than a predetermined threshold value Is determined, and moving object information having a spatiotemporal likelihood lower than the predetermined threshold is deleted from moving object information stored as the search result moving object information.
- an image is captured, a person is detected, a person image including the detected person image is extracted, and a spatial position coordinate of the person is detected based on the person image.
- the person information including the person image, the spatial position coordinates of the person, and the imaging time when the image was captured is output, and the person in the search target person information is the person information including the person image of the person to be searched
- a person image likelihood that is a likelihood of a person image included in person information other than the search reference person information for the image is calculated, and whether each of the calculated person image likelihoods is higher than a predetermined threshold value
- the person information having the person image likelihood higher than the predetermined threshold is person information including person images of the same person as the person image of the person information to be searched
- the person information searched as information and the search result person information is stored, and among the person information stored as the search result person information, based on the person image, the user information
- the confirmation information of person information other than the confirmed person information which is the person information to which the confirmation information is input is received from the
- the spatiotemporal likelihood composed of the likelihood based on the spatial position coordinates and the imaging time is calculated with respect to the confirmed person information inputted, and whether or not each of the calculated spatiotemporal likelihood is lower than a predetermined threshold value Is determined, and the person information having a spatiotemporal likelihood lower than the predetermined threshold is deleted.
- the information processing apparatus of the present technology may be an independent apparatus or a block that performs information processing.
- FIG. 11 is a diagram illustrating a configuration example of a general-purpose personal computer.
- First embodiment an example using moving object detection
- Second embodiment an example using person detection
- FIG. 1 shows a configuration example of an embodiment of a monitoring system to which the present technology is applied.
- the monitoring system 1 in FIG. 1 captures a plurality of areas in a monitoring area that requires search and tracking of a person, searches for a person in the monitoring area based on the captured image, and a movement path of the person. Is to track.
- the monitoring system 1 includes monitoring cameras 11-1 to 11-n, a person search tracking server 12, and a network 13.
- Each of the monitoring cameras 11-1 to 11-n is installed in n locations in the monitoring area covered by the monitoring system 1, and images of each monitoring area are captured to detect a moving object in the captured images. Then, an image of the detected moving object is extracted. Then, the monitoring cameras 11-1 to 11-n supply the moving object information including the moving image of the extracted moving object to the person search tracking server 12 via the network 13 including the Internet, the public line, or the dedicated line. .
- the moving body information is information including an imaging time, a foreground image of the moving body image, a world coordinate of the moving body obtained from the moving body image, and an ID for identifying the surveillance camera 11 in addition to the moving body image.
- the world coordinates are coordinate information for specifying the position of the moving object in the space.
- the coordinates are latitude and longitude coordinate information for specifying the position on the earth.
- the monitoring cameras 11-1 to 11-n when it is not necessary to particularly distinguish each of the monitoring cameras 11-1 to 11-n, they will be simply referred to as the monitoring camera 11, and the other configurations will be referred to in the same manner.
- the person search tracking server 12 acquires the moving object information supplied from the monitoring camera 11 and filters out images inappropriate for the search based on the moving object image and the foreground image of the moving image included in the moving object information. Then, moving body information including a moving body image suitable for the search is held. In addition, the person search / tracking server 12 executes the moving object information matching process based on the person information designated by the user from the held moving object information, and extracts the moving object information of the designated person. . Furthermore, the person search / tracking server 12 generates and displays a display image of the search / tracking result including the movement path of the designated person based on the moving body information extracted by the moving body information matching process.
- the person search / tracking server 12 accepts input of confirmation information for confirming that the part of the moving body information as the search tracking result is that of the person specified by the user, and the confirmation information is accepted.
- the search tracking result is corrected and updated by executing matching correction processing using the moving object information.
- the person search tracking server 12 displays the search tracking result by the moving body information matching process or the matching correction process in this way, receives the input of the user's confirmation information, and repeats the matching correction process to thereby obtain the search tracking result. Improve the accuracy.
- the surveillance cameras 11 are arranged at a plurality of positions in the surveillance area, and are arranged so as to cover the entire area of the surveillance area as a whole by combining images captured by the surveillance cameras 11-1 to 11-n. Is done.
- the monitoring camera 11 includes an imaging unit 31, a moving object detection unit 32, a foreground image extraction unit 33, a camera ID storage unit 34, an imaging position coordinate calculation unit 35, an imaging time detection unit 36, and a moving object information output unit 37.
- the imaging unit 31 includes a CCD (Charge Coupled Devices), a CMOS (Complementary Metal Oxide Semiconductor), and the like, and continuously captures images in a range that can be captured from the position where the monitoring camera 11 is installed.
- the imaging unit 31 captures a moving image composed of images that can be handled as still images continuously in time series. Therefore, the image captured by the imaging unit 31 can be processed as a single image in frame or field units, and can also be processed as a moving image by continuously reproducing them. is there. In the following description, it is assumed that images consisting of still images are continuously captured. Therefore, hereinafter, the image indicates a still image.
- the moving object detection unit 32 detects, for each of the images captured by the imaging unit 31, a region where the moving object is captured in the image based on the relationship with the images that follow in time series, and includes a region including the region where the moving object exists.
- a shape image is extracted as a moving object image.
- the foreground image extraction unit 33 extracts a foreground image composed of binary pixels of the foreground region and other regions from the moving object image extracted by the moving object detection unit 32.
- the foreground image is, for example, an image in which the foreground area is displayed in white and the other areas are displayed in black.
- the binary value may be a value representing a color other than this, or may be an image in which the relationship between white and black is reversed.
- the area where the moving object is imaged often constitutes the foreground area, so the foreground image is divided into the area where the moving object is displayed and the other areas of the moving object image, respectively.
- the pixel value is extracted as an divided image.
- the camera ID storage unit 34 stores camera IDs for identifying each of the monitoring cameras 11-1 to 11-n, and is read when the moving body information is generated in the moving body information output unit 37.
- the imaging position coordinate calculation unit 35 calculates the world coordinates of the object detected as a moving object based on the information of the image captured by the imaging unit 31. That is, the imaging position coordinate calculation unit 35 recognizes its installation position by, for example, a GPS (Global Positioning System) (not shown) and recognizes a deviation angle with respect to the imaging direction with respect to the geomagnetism, and The distance to the object detected as the moving object is obtained from the size and position of the moving object, and the world coordinates are calculated based on the information.
- GPS Global Positioning System
- the imaging time detection unit 36 has a built-in function for generating time information such as a real time clock (not shown), and uses the time at the timing when each image captured by the imaging unit 31 is captured as the imaging time. To detect.
- time information such as a real time clock (not shown)
- the moving body information output unit 37 obtains each of the moving body images extracted from the image captured by the imaging unit 31 based on the corresponding foreground image, camera ID, world coordinates of the moving body, and imaging time information.
- the moving body information is generated and output to the person search tracking server 12 via the network 13.
- the person search tracking server 12 includes a moving body information acquisition unit 51, a moving body information filtering processing unit 52, a moving body information holding unit 53, a moving body information likelihood calculation processing unit 54, a moving body information likelihood threshold determination unit 55, a result storage unit 56, and a display.
- the image generating unit 57, the display unit 58, the matching correction processing unit 59, and the operation input unit 60 are configured.
- the moving body information acquisition unit 51 acquires the moving body information supplied from the monitoring camera 11 via the network 13 and temporarily stores the moving body information, and supplies the moving body information to the moving body information filtering processing unit 52.
- the moving body information filtering processing unit 52 filters the moving body information supplied from the moving body information acquisition unit 51 according to a predetermined condition, extracts only the moving body information suitable for searching for a person, and the moving body information holding unit 53, the moving body information which is not suitable is discarded.
- the moving object information filtering processing unit 52 includes a detection frame size determination unit 71, a foreground bias determination unit 72, a foreground / background correlation determination unit 73, an edge length determination unit 74, and a multi-person determination unit 75.
- the detection frame size determination unit 71 determines whether the moving body image is suitable for person search based on whether the frame size of the moving body image is larger than a predetermined size, and determines that the moving body information is not suitable for person search. If this happens, the moving body information is discarded.
- This predetermined size may be the minimum size required for the person search empirically.
- the detection frame size determination unit 71 estimates the height of a person detected as a moving object based on the frame size of the moving object image, and searches for a person with the estimated height with an optimal frame size. You may make it determine whether it exists. That is, for a tall person, a moving image with a small frame size has a low resolution, and is therefore considered inappropriate for person search. On the other hand, in the case of a person having a short height, even if the frame size of the moving object image is small, the resolution may not be lowered as the tall person is. Therefore, such a case is regarded as appropriate for the person search.
- the foreground bias determination unit 72 calculates a ratio of the foreground area of the foreground image in the moving object image, and when the ratio is higher than a predetermined threshold, the moving object information including the moving image is selected as a moving object suitable for person search. Discard it as non-information. In other words, since only a moving body image in which only an image of the foreground area is captured is not suitable for searching for a person, moving body information including such a moving body image is not suitable for searching for a person. Discarded as being.
- the foreground / background correlation determination unit 73 obtains the correlation between the foreground area and the background area of the moving body image based on the moving body image and the foreground image, and when the correlation is high, that is, the person to be originally searched should be captured. If the foreground area is an image that is almost the same as the background area, the moving object information is discarded. That is, in such a case, since there is a high possibility that a moving object, that is, a person is not captured in the region regarded as the foreground region, it is regarded as unsuitable for person search, and the foreground / background correlation determination unit 73 is considered. Discards moving object information including such a moving object image.
- the edge length determination unit 74 generates an edge image that is a boundary with the background image based on the foreground image.
- the moving object information including the moving object image is used for person search. Is deemed inappropriate and is discarded. That is, if a person who is a moving object is a foreground image that is correctly extracted as a foreground, the moving object image is regarded as an area in which the person is accurately imaged.
- the foreground region is often extracted in a spot shape unrelated to the shape of the person, and as a result, the length of the edge that is the boundary between the foreground region and the background region is extracted. Therefore, the edge length determination unit 74 regards moving object information including such a moving object image as not suitable for person search and discards it.
- the multi-person determination unit 75 generates a waveform with the horizontal pixel position of the foreground image as the horizontal axis and the integrated value of the pixel values arranged in the vertical direction as the vertical axis, and the waveform has a plurality of maximum values. It is determined whether or not it is suitable for person search depending on whether or not it exists, and moving body information not suitable for person search is discarded. That is, the waveform obtained using the horizontal pixel position of the foreground image as the horizontal axis and the integrated value of the pixel values arranged in the vertical direction as the vertical axis is obtained when there is one person captured in the moving object image. Is considered to be a waveform having a maximum value of 1 and a convex shape of only 1. However, when a plurality of persons are imaged, a maximum value is generated for the number of persons in the waveform. In such a case, it is regarded as not suitable for person search.
- the moving body information holding unit 53 holds moving body information suitable for person search by the moving body information filtering processing unit 52, and supplies the moving body information likelihood calculation processing unit 54 and the display image generation unit 57 as necessary.
- the moving body information likelihood calculation processing unit 54 is the reference moving body information that is a search target including the moving body image of the moving body information specified as the search target among the moving body images included in the moving body information held in the moving body information holding unit 53. With respect to the moving image of the other moving body information, the moving body image likelihood is calculated for each moving body information and supplied to the moving body information likelihood threshold determination unit 55.
- the moving body information likelihood threshold determination unit 55 determines whether the moving body image likelihood obtained based on the moving body image calculated by the moving body information likelihood calculation processing unit 54 is higher than the threshold, and the moving body image likelihood. Is stored in the result storage unit 56. That is, the moving body information likelihood threshold determination unit 55 performs moving body image matching processing based on the moving body image likelihood of other moving body information with respect to the reference moving body information to be searched, and the moving body image having a high moving body image likelihood. The moving body information including is extracted as a matching based on the moving body image. Then, the moving body information likelihood threshold determination unit 55 stores the extracted moving body information in the result storage unit 56 as a matching result with the reference moving body information that is the search target.
- the display image generation unit 57 displays the search tracking result of the person who is the search target based on the moving body information stored in the result storage unit 56 as a match with the reference moving body information that identifies the person who is the search target.
- a display image to be generated is generated.
- the display image generation unit 57 displays the generated display image on a display unit 58 formed of an LCD (Liquid Crystal Display), an organic EL (Electro Luminescence), or the like.
- the display unit 58 displays a display image and accepts an operation input for the display image by using the operation input unit 60.
- the operation input unit 60 generates an operation signal corresponding to the operation content on the display image as a touch panel or a pointer, and supplies the operation signal to the moving object information likelihood calculation processing unit 54 and the matching correction processing unit 59.
- the matching input processing unit 59 determines that the user is a moving body image of a person to be searched based on the moving body image displayed in the search tracking result displayed on the display unit 58, the operation input unit 60 The matching correction process is executed based on the confirmation information that is input by being operated.
- the matching correction processing unit 59 executes the matching correction processing again when the confirmation information is input again even after the matching correction processing is executed, and repeats the matching correction every time the determination information is input. Execute the process.
- the matching correction processing unit 59 includes an operation input recognition unit 91, a BTF calculation unit 92, a spatiotemporal likelihood calculation unit 93, a spatiotemporal likelihood threshold determination unit 94, a BTF image processing unit 95, and a BTF image likelihood.
- a calculation unit 96 and a BTF image likelihood threshold determination unit 97 are provided.
- the operation input recognizing unit 91 recognizes, based on the operation signal of the operation input unit 60, that confirmed information has been input for the selected moving body information among the search tracking results displayed on the display unit 58.
- the BTF calculation unit 92 compares the camera IDs of the moving body image specified as the search target and the moving body image searched by the moving body image matching process or the matching correction process. Colors constituting images generated due to individual differences of the monitoring cameras 11, environmental differences, or the like based on pixel signals between moving body images or between moving body images designated as search targets and moving body images to which confirmation information is input BTF (Brightness Transfer Function) that corrects changes in In other words, even if the actual color is the same color, the images captured by the different monitoring cameras 11 have different colors depending on the individual differences and environmental differences. However, when the moving body image specified as the search target and the actually searched moving body image are matched by the user and the confirmation information is input, the same in both moving body images. It is determined that the areas constituting the part are the same color. Therefore, the BTF calculating unit 92 calculates a BTF for correcting the color of the moving body image for which matching is confirmed based on the reference moving body image from these relationships.
- BTF Bitness Transfer Function
- the spatiotemporal likelihood calculation unit 93 obtains a moving distance from the difference between the world coordinates included in the moving body information to which the confirmation information is input and the world coordinates included in other moving body information stored in the result storage unit 56.
- the spatiotemporal likelihood of the moving object information is calculated from the average required time required for the moving distance based on the average moving speed of the human and the time between the imaging times of the moving object information.
- the spatiotemporal likelihood corresponds to the ratio of the time between imaging times to the average required time between moving object information
- the spatiotemporal likelihood decreases as the ratio goes away from 1.
- the spatiotemporal likelihood gradually decreases. That is, when the time between imaging times is close to 0, it becomes practically impossible to move as the spatial distance increases. In such a case, the spatiotemporal likelihood becomes extremely small.
- the spatiotemporal likelihood becomes extremely small.
- the spatiotemporal likelihood is 1 in such a case. Although it is lower than the time, it is feasible, so the value is higher than when the time between imaging times is zero.
- the spatiotemporal likelihood threshold determination unit 94 determines whether or not the spatiotemporal likelihood between the moving object information to which the confirmation information is input and the moving object information stored in the other result storage unit 56 is lower than a predetermined threshold value. If the spatiotemporal likelihood is lower than a predetermined threshold, the moving body information is deleted from the result storage unit 56.
- the BTF image processing unit 95 when the spatiotemporal likelihood is not lower than a predetermined threshold, among the moving body information stored in the result storage unit 56 in which the fixed information is not input, the moving body information in which the fixed information is input.
- a BTF between moving object images is obtained between the two, a BTF color conversion process is performed on the moving object image included in the moving object information.
- the BTF image likelihood calculation unit 96 uses the respective pixel signals of the moving object image of the moving object information to which the definite information is input and the moving object image to which the definitive information is not input and subjected to the BTF color conversion process. BTF image likelihood, which is the likelihood between moving body images, is calculated.
- the BTF image likelihood threshold value determination unit 97 compares the BTF image likelihood calculated by the BTF image likelihood calculation unit 96 with a predetermined threshold value, and obtains moving object information including a moving image lower than the predetermined threshold value as the result storage unit 56. Remove more.
- step S1 the imaging unit 31 of the monitoring camera 11 continuously captures images composed of still images or moving images in the monitoring area that can be monitored from the installed position.
- step S ⁇ b> 2 the moving object detection unit 32 detects a moving object by comparing the images captured by the imaging unit 31 with images before and after the image, and generates a rectangular moving object image including a region where the detected moving object exists. To do. For example, when an image picked up by the image pickup unit 31 is fixedly picked up, no change occurs in the background region without movement. On the other hand, since an area in the image where the moving object exists changes on the image due to some movement, the moving object detection unit 32 detects the area where the change occurs as an area where the moving object exists, and the detected moving object A moving body image is generated by trimming a rectangular image including a region in which an image exists.
- step S3 the foreground image extraction unit 33 extracts a foreground image from the moving body image generated by the moving body detection unit 32, and for example, a foreground image composed of binary values in which the foreground area is white and the other areas are black. Is generated.
- the imaging position coordinate calculation unit 35 calculates the imaging direction, angle, and distance from the position and size of the moving object in the moving object image, and further determines the subject from the world coordinates where it is installed. Calculate the world coordinates of a moving object. That is, the imaging position coordinate calculation unit 35 calculates, for example, a coordinate position including a latitude and longitude on the earth of a moving object as world coordinates based on the moving object image.
- step S5 the imaging time detection unit 36 detects the time information at the timing when the image is captured as the imaging time based on the time information generated by a real time clock (not shown).
- step S6 the moving object information output unit 37 reads the camera ID from the camera ID storage unit 34, and generates moving object information by combining the moving object image, the foreground image, the world coordinates, and the imaging time.
- step S7 the moving body information output unit 37 outputs the generated moving body information to the person search and tracking server 12 via the network 13 represented by the Internet.
- an image is captured for each monitoring area in each of the monitoring cameras 11, a moving object in the captured image is detected, and a moving object image is extracted. Then, together with the moving object image, moving object information including the foreground image of the moving object image, the world coordinates of the moving object in the moving object image, and the information of the imaging time when the image was captured is generated and supplied to the person search and tracking server 12. .
- Moving object information filtering process Next, moving object information filtering processing by the person search and tracking server 12 will be described with reference to the flowchart of FIG.
- step S21 the moving body information acquisition unit 51 acquires and stores the moving body information sequentially supplied from the monitoring server 11 via the network 13.
- step S22 the moving body information filtering processing unit 52 sets any of the unprocessed moving body information among the moving body information stored in the moving body information acquisition unit 51 as the target moving body information to be processed.
- step S23 the detection frame size determination unit 71 extracts a moving object image included in the attention moving object information.
- step S24 the detection frame size determination unit 71 determines whether or not the frame size of the moving object image of the moving object information of interest is within a predetermined range. That is, when the frame size of the moving body image is small, for example, it is difficult to extract the feature amount used for the person search. Therefore, the detection frame size determination unit 71 uses the feature amount necessary for the person search. Is a frame size within a predetermined range suitable for a person search that can be sufficiently extracted.
- step S24 if the detection frame size determination unit 71 determines that the frame size of the moving body image of the moving body information of interest is within a predetermined range and is not a moving body image inappropriate for human body search, Proceed to step S25. If it is determined in step S24 that the frame size of the moving object image of the moving object information of interest is not within the predetermined determination, the moving object information filtering processing unit 52 discards the moving object information of interest in step S34.
- step S25 the foreground bias determination unit 72 extracts a foreground image from the attention moving body information.
- the foreground bias determination unit 72 calculates the ratio of the foreground region distribution to the entire moving object image in the extracted region that becomes the foreground image, and is higher than a predetermined threshold value. It is determined whether or not there is no bias. That is, for example, in the case of a moving body image as shown by the image P1 in FIG. 6, the back of the person's back and shoulders are shown. In such a case, the foreground image is as shown by an image P2 in FIG. In the case of the image P2, it can be said that the range surrounded by the straight square is an image that is not suitable for person search because the ratio of the foreground region is high.
- the foreground bias determination unit 72 considers that the ratio of the foreground area distribution is higher than a predetermined threshold, the ratio of the foreground area is biased, and is not suitable for person search. In S34, the moving object information of interest is discarded.
- step S26 if it is determined in step S26 that the foreground area is smaller than the predetermined threshold for the moving object image based on the information of the foreground image and the ratio of the foreground area is not biased, the process proceeds to step S27.
- the foreground / background correlation determination unit 73 calculates the correlation between the white area constituting the foreground area and the other area, that is, the background area, in the moving image based on the moving image and the foreground image. .
- the foreground / background correlation determination unit 73 generates, for example, a histogram for each color of the foreground region and the background region, and obtains the reciprocal of the difference between the most frequent colors as the correlation. In this case, the correlation value increases as the correlation increases, and conversely decreases as the correlation decreases.
- step S28 the foreground / background correlation determination unit 73 determines whether or not the obtained correlation is higher than a predetermined threshold, that is, whether or not the foreground image and the background image are not changed and the foreground is regarded as not moving. judge.
- a predetermined threshold that is, whether or not the foreground image and the background image are not changed and the foreground is regarded as not moving. judge.
- the foreground / background correlation determination unit 73 considers that the correlation value is higher than a predetermined threshold value, and the moving object information of interest is discarded in step S34.
- the moving object image P11 and the foreground image P12 as shown in FIG. 7 are, for example, a region where the person is imaged because the moving object that is a person exists at the same position in the imaging range for a long time.
- the person moves out of the imaging range, and the area where the person has existed is detected as a moving object. That is, since a person has existed in the area shown in white in the foreground image P12 for a long time, it was regarded as a background image including the area where the person was present. Thereafter, when the person moves and moves out of the imaging range, the white area of the foreground image where the person has existed is regarded as the moving object existing area.
- the image is a result of detecting a moving object.
- step S28 determines whether the correlation value is lower than the predetermined threshold, the moving object is present in the foreground area, and the other area is the background area. If it is determined in step S28 that the correlation value is lower than the predetermined threshold, the moving object is present in the foreground area, and the other area is the background area, the process proceeds to step S29.
- step S29 the edge length determination unit 74 obtains an edge image including a boundary between the foreground image and the background image based on the foreground image, and further, an edge length that is the length of the edge included in the obtained edge image. Ask for.
- step S30 the edge length determination unit 74 determines whether or not the obtained edge length is shorter than a predetermined threshold value. If the edge length is not shorter than the predetermined threshold value, the process proceeds to step S34, and the moving object information of interest Is destroyed. That is, for example, as shown in the image P21 of FIG. 8, in the case of a back image of a person, the foreground area is correctly extracted because the color of the clothes worn by the person is similar to the color of the background. In some cases, the foreground image may be as shown by the image P22. As shown in the image P22, if the foreground region is extracted as a spot, it is not suitable for person search.
- step S30 determines whether the edge length is shorter than the predetermined length. If it is determined in step S30 that the edge length is shorter than the predetermined length, the process proceeds to step S31.
- step S31 the multiple-person determination unit 75 extracts a foreground image and generates a vector whose element is the sum of pixel values existing in the vertical direction for each coordinate position in the horizontal direction. That is, when the foreground image corresponding to the moving object image as shown by the image P31 in FIG. 9 is the image P32, the horizontal coordinate position of the foreground image as shown at the top of the image P32 is used as the horizontal axis. Thus, a waveform with the vertical axis representing the sum of the vertical pixel values for each horizontal coordinate position is obtained. Therefore, the multiple-person determination unit 75 calculates a vector V corresponding to this waveform.
- V (P (x1, y1) + P (x1, y2) + ... + P (x1, ym), P (x2, y1) + P (x2, y2) + ... + P (x2, ym), P (X3, y1) + P (x3, y2) +... + P (x3, ym) +..., P (xn, y1) + P (xn, y2) + ... + P (xn, ym))
- step S32 the multiple-person determination unit 75 calculates the correlation between the vector V1 and the vector V when a single person is captured in the moving image, and whether the calculated correlation is higher than a predetermined threshold value. Determine whether or not. That is, as shown by an image P31 in FIG. 9, when a single person is captured as a moving object image, the horizontal coordinate position of the foreground image P32 is taken as the horizontal axis, and the vertical coordinate for each horizontal coordinate position is taken.
- the waveform having the sum of pixel values as the vertical axis has an outer shape including one convex waveform including one maximum value.
- the foreground image is the image P42, and therefore the sum of the pixel values in the vertical direction for each coordinate position in the horizontal direction.
- the waveform has an outer shape in which four convex waveforms including four local maximum values corresponding to the number of people are formed.
- the vector V described above corresponds to the outer shape of this waveform
- the vector V1 when a single person is imaged includes one maximum value as shown at the top of the image P32. Since this corresponds to a convex waveform, the correlation is high.
- the outer shape since the outer shape includes a plurality of upwardly convex waveforms including a plurality of maximum values as shown in the upper part of the image P42, The correlation with the vector V1 is low.
- step S32 when the calculated correlation is not higher than the predetermined threshold value, the multiple-person determining unit 75 considers that a plurality of persons are included in the moving image, and the process proceeds to step S34.
- the moving object information is discarded.
- step S32 if the calculated correlation is higher than the predetermined threshold value in step S32, it is considered that only a single person is included in the moving image, and the process proceeds to step S33.
- step S33 the moving body information filtering processing unit 52 regards the target moving body information as moving body information suitable for person search, and causes the moving body information holding unit 53 to hold it.
- step S35 the moving body information filtering processing unit 52 determines whether or not unprocessed moving body information exists in the moving body information acquisition unit 51. If unprocessed moving body information exists, the process returns to step S22. . That is, the processes in steps S22 to S35 are repeated until there is no unprocessed moving body information. If it is determined in step S35 that there is no unprocessed moving body information, the process ends.
- the frame size is within a predetermined range, the foreground image is not biased, and the correlation between the foreground image and the background image is a predetermined threshold value. Only that the edge length in the edge image of the foreground image is shorter than a predetermined length and a single person is considered to be captured is considered to be a moving body image suitable for person search. Only the moving body information including the moving body image is held in the moving body information holding unit 53 as being suitable for the person search by the moving body information filtering process.
- the moving body image included in the moving body information supplied from the monitoring camera 11 is the moving body images P101 to P118 as shown in the state L1 shown in FIG. Will be filtered as follows.
- the information is discarded, and moving body information including the moving body image indicated by the state L3 is left.
- the moving body information including the moving body images P104, P105, P110, and P115 having a high foreground area ratio biased in the foreground image is discarded by the processing in steps S25 and S26.
- the moving body information including the moving body image indicated by L4 is left.
- the moving object image P118 is an image corresponding to the image P11 of FIG. 7, and the foreground image is P12.
- the moving body information including the moving body image P108 in which the foreground area of the foreground image in which the foreground area of the foreground image is not correctly extracted is discarded by the processing in steps S29 and S30, and the moving body including the moving body image illustrated in state L6 is obtained. Information is left.
- the moving object image P108 is an image corresponding to the image P21 in FIG.
- the moving body information including the moving body images P103, P111, and P116 in which a plurality of persons are captured is discarded by the processing in steps S31 and S32, and the moving body image illustrated in the state L7 is included. Moving object information is left.
- the moving body information filtering process can filter only moving body information including moving body images suitable for person search and hold the moving body information in the moving body information holding unit 53, thereby improving the accuracy of person search. Is possible.
- moving object information that is not suitable for person search, it is possible to speed up moving object information matching processing, which will be described later with reference to FIG.
- Moving object information matching processing by the person search and tracking server 12 will be described with reference to the flowchart of FIG.
- step S ⁇ b> 51 the display image generation unit 57 displays a list of moving body images on the moving body information held in the moving body information holding unit 53 on the display unit 58, and selects a moving body image that designates a search target person. Display a prompt display image.
- step S52 the moving object information likelihood calculation processing unit 54 determines whether or not the operation input unit 60 has been operated to select a moving image specifying a person to be searched. If a moving body image that designates a person to be searched is not selected, the process returns to step S51. That is, the processes in steps S51 and S52 are repeated until a moving body image that specifies a person to be searched is specified. In step S52, when a moving body image that specifies a person to be searched is specified, the process proceeds to step S53.
- step S53 the moving body information likelihood calculation processing unit 54 sets the moving body information corresponding to the moving body image specified as the search object among the moving body information held in the moving body information holding unit 53 as the reference moving body information, Other moving body information is set as unprocessed moving body information.
- step S54 the moving body information likelihood calculation processing unit 54 reads any unprocessed moving body information held in the moving body information holding unit 53, and sets it as attention moving body information.
- the moving body information likelihood calculation processing unit 54 calculates a moving body image likelihood that is a likelihood between the moving body image included in the reference moving body information and the moving body image included in the target moving body information, It supplies to the moving body image likelihood threshold value determination part 55 with attention moving body information. That is, the moving body information likelihood calculation processing unit 54, for example, a moving body based on a similarity that becomes higher as the person captured in each moving body image of the reference moving body information and the attention moving body information is similar. Image likelihood is calculated.
- step S56 the moving image likelihood threshold determination unit 55 determines whether or not the supplied moving image likelihood is higher than a predetermined threshold. That is, the moving image likelihood threshold determination unit 55 determines whether the moving image person in the reference moving image information matches the moving object image likelihood based on whether the moving image likelihood is higher than a predetermined threshold. Determine whether or not. If the moving image likelihood is higher than the predetermined threshold in step S56 and the person of the moving image of the reference moving information and the moving object information is considered to match (match), the process proceeds to step S57. move on.
- step S57 the moving object image likelihood threshold determination unit 55 stores the moving object information of interest in the result storage unit 56. At this time, the moving body information likelihood calculation processing unit 54 sets the moving object information held in the moving body information holding unit 53 as the processed moving body information.
- step S56 when the moving image likelihood is not higher than the predetermined threshold in step S56 and the person of the moving image of the reference moving information and the moving object information is considered not to match (does not match), the process is as follows. Proceed to step S59.
- step S59 the moving object image likelihood threshold value determination unit 55 discards the moving object information of interest. Then, the moving object information likelihood calculation processing unit 54 sets the moving object information held in the moving object information holding unit 53 to the processed moving object information.
- step S58 the moving body information likelihood calculation processing unit 54 determines whether or not unprocessed moving body information is held in the moving body information holding unit 53. For example, when unprocessed moving body information is held, The process returns to step S54. That is, the processing of steps S54 to S59 is repeated until there is no unprocessed moving body information in the moving body information holding unit 53, and matching of all moving body information with the reference moving body information in the moving body image is performed. The process is repeated.
- step S58 If it is determined in step S58 that unprocessed moving body information is not held in the moving body information holding unit 53, the process proceeds to step S60.
- step S60 the display image generation unit 57 generates a display image indicating the search tracking result based on the moving object information stored in the result storage unit 56.
- step S61 the display image generation unit 57 displays a display image indicating the search tracking result on the display unit 58.
- the display image indicating the search tracking result is, for example, an image P201 as shown in FIG.
- the moving object of the same person as the person to be searched is located at the position corresponding to the world coordinates in the monitoring area Z.
- Plots A to E indicating the positions where the images are taken are displayed.
- Areas Z1 and Z2 indicate objects in the monitoring area.
- the imaging time when the moving body image included in each of the moving body information is captured is displayed. Yes. In this case, it is shown that the imaging times of the plots A to E are 1:00:00, 1:00:05, 1:00:09, 1:00:18, and 1:00:20. Has been.
- the moving images captured at the respective positions by the lead lines are attached to the plots A to E, respectively, and the moving images PA to PE are attached to the plots A to E, respectively.
- the moving body images PA to PE displayed in the state of being connected by the lead lines from the respective plots A to E in the display image P201 are visually confirmed by the user and confirmed to be a search target person.
- a matching correction process described later is executed with reference to the flowchart of FIG. 13, and the search tracking result can be corrected to obtain a search tracking result with higher accuracy.
- step S62 the matching correction processing unit 59, based on the information of the display image P201 supplied from the display image generation unit 57 and the operation signal supplied from the operation input unit 60, the moving body image PA indicated by the display image P201. Or any one of PEs is selected, confirmation information is input, and it is determined whether matching correction processing is instructed.
- step S62 when any of the moving body images PA to PE indicated by the display image P201 is selected and the confirmation information is input, it is considered that the matching correction process is instructed, and the process proceeds to step S63. .
- step S63 the matching correction processing unit 59 executes the matching correction process and displays the person search tracking result on which the correction process has been performed.
- the matching correction process will be described later in detail with reference to FIG.
- step S64 the matching correction processing unit 59 determines whether or not the end is instructed. If the end is not instructed, the process returns to step S61. That is, the display screen P201 showing the person search tracking result is continuously displayed on the display unit 58. Then, in step S64, when the end is instructed, the process ends.
- the moving body information matching process simply by specifying the moving body image of the person to be searched, the moving body information matching the person to be searched is searched based on the moving body image, and from the moving body information as the search result, It becomes possible to display the search tracking result indicating the movement path of the person to be searched.
- the example of specifying from the moving body image of the moving body information held in the moving body information holding unit 53 when specifying the person to be searched has been described, but other than the image captured by the monitoring camera 11
- the person to be searched may be specified by handling the same image as the moving object image.
- step S81 the matching correction processing unit 59 stores information indicating that the moving body image to which the confirmation information is input by the operation input unit 60 is included in association with the moving body information.
- the moving body information to which the confirmation information is input is determined to be the moving body information appropriately extracted by the moving body image matching process. Therefore, in the subsequent processing, the moving body information is excluded from the search target, and the result storage unit 53 Will continue to be remembered.
- step S ⁇ b> 82 the matching correction processing unit 59 sets all the moving body information for which the confirmation information is not input among the moving body information stored in the result storage unit 56 as unprocessed moving body information.
- step S ⁇ b> 83 the matching correction processing unit 59 sets the moving body information for which the confirmed information has been input most recently among the moving body information stored in the result storage unit 56 as the reference moving body information.
- step S84 the BTF calculating unit 92 determines whether the BTF can be calculated.
- the condition under which the BTF can be calculated is, first, a camera ID that is different from the camera ID included in the reference moving object information, and the moving object information including other moving object images to which the confirmation information is input is There is a time.
- Second there is a moving body image to which two or more pieces of confirmed information are input, each of which is captured by the monitoring camera 11 having a different camera ID. Therefore, the BTF calculation unit 92 determines whether or not the BTF can be calculated based on whether moving object information that satisfies one of the above two conditions exists in the result storage unit 56.
- step S84 if the BTF can be calculated, in step S85, the BTF calculation unit 92 has a different camera ID, but the pixel value between the moving body images included in the moving body image or the reference moving body information to which the definite information is input is input. Using this information, BTF calculation processing is executed to calculate BTF.
- BTF is a function that indicates the mutual correspondence of colors between moving body images including the same subject imaged by the surveillance camera 11 specified by two different camera IDs. Therefore, by calculating the BTF, the color of any moving body image captured by the two monitoring cameras 11 having different camera IDs regardless of the presence or absence of the definite information is converted by using the BTF, so that the color It becomes possible to correct.
- this color correction processing it is possible to correct a color change caused by a difference in imaging environment light in a monitoring area between different monitoring cameras 11 or a solid difference in imaging elements of the imaging unit 31.
- the moving body images captured by the two different monitoring cameras 11 are corrected in color as if they were captured by the same monitoring camera 11, so that the moving body image likelihood between the moving body images can be more accurately determined. It is possible to calculate.
- step S84 for example, when the moving body image to which the confirmation information is input is only captured by one type of surveillance camera 11, it is considered that BTF cannot be calculated, and the process of step S85 is skipped. Is done.
- the spatiotemporal likelihood calculation unit 93 is the moving body information stored in the result storage unit 56, and among the moving body information excluding the reference moving body information, any of the unprocessed moving body information is the attention moving body information. Set to.
- the spatiotemporal likelihood calculation unit 93 calculates a spatiotemporal likelihood based on the information on the world coordinates and the imaging time included in the reference moving body information and the moving object information of interest. More specifically, the spatiotemporal likelihood calculation unit 93 obtains a movement distance from the difference between the world coordinates included in the reference moving body information and the world coordinates included in the target moving body information, for example, and the movement distance is calculated as a human average. The spatiotemporal likelihood of moving object information is calculated based on the ratio of the time between imaging times of moving object information with respect to the average required time required for a specific moving speed.
- the spatiotemporal likelihood threshold determination unit 94 determines whether or not the calculated spatiotemporal likelihood is higher than a predetermined threshold. For example, when the display image showing the person search tracking result is the display image shown in the upper part of FIG. 14, the moving body image PA corresponding to the plot A is selected by the operation input unit 60 as shown by the hand H1. Thus, when the definite information is input, the moving body information corresponding to the plot A becomes the reference moving body information. In the case of the display image of FIG. 14, in the moving body information corresponding to the plots A to E, the imaging times are 1:00:00, 1:00:05, 1:00:10, 1:00:15, And 1:00:05.
- Plots A and B are moving body information corresponding to moving body images PA and PB captured by the same monitoring camera 11 that captures the imaging area CamA indicated by the triangle formed by the same dotted line.
- the plots C and D are moving body information corresponding to the moving body images PC and PD captured by the same monitoring camera 11 that captures the imaging area CamB indicated by the dotted triangle.
- the plot E is moving body information corresponding to the moving body image PE captured by the monitoring camera 11 that captures an imaging area CamC indicated by a triangle formed by a dotted line.
- the moving body information corresponding to the plot B is the attention moving body information
- the distance between the plots AB is the distance AB
- the moving distance with respect to the average required time required by the average moving speed of the human is 5 / (AB / w).
- w is an average moving speed of a human. For example, if the distance AB is 5 m and the average moving speed of a human is 1 m / s, the ratio is 1, and the spatiotemporal likelihood is the highest value.
- the moving body information corresponding to the plots C and D is the attention moving body information
- the distance between the plots AC and AD is the distances AC and AD
- the moving distance is determined by the average moving speed of the human.
- the ratio of the time between imaging times of moving body information to the average required time is 10 / (AC / w) and 15 / (AD / w), respectively.
- the distances AC and AD are 10 m and 15 m, respectively, and the average moving speed of humans is 1 m / s, the ratio is 1 and the space-time likelihood is the highest value. .
- step S88 when the spatiotemporal likelihood is the highest value as described above, the spatiotemporal likelihood threshold determination unit 94 considers that the spatiotemporal likelihood threshold is higher than the predetermined threshold, and the process proceeds to step S89.
- the moving body information corresponding to the plot E is the attention moving body information
- the distance between the plots AE is the distance AE
- the moving body with respect to the average required time required by the average moving speed of the human is calculated.
- the ratio of time between image capturing times of information is 5 / (AE / w). That is, for example, if the distance AE is 25 m and the average human moving speed is 1 m / s, the 25 m is moved in about 5 seconds, so the ratio is 0.2.
- the spatiotemporal likelihood is a low value.
- the spatiotemporal likelihood threshold determination unit 94 considers that the spatiotemporal likelihood threshold is lower than the predetermined threshold, and the process proceeds to step S96.
- step S96 the matching correction processing unit 59 deletes the moving object information from the result storage unit 56. That is, when an image showing a search tracking result as shown by the image P211 in FIG. 14 is displayed, when confirmation information is input to the moving object image PA corresponding to the plot A, based on the spatio-temporal information, As shown in the table at the bottom of the display image in FIG. 14, the moving body information of the plot E where the same person cannot exist is deleted.
- step S89 the BTF image processing unit 95 determines whether or not BTF is obtained. For example, if BTF is not obtained, the process proceeds to step S93.
- step S ⁇ b> 93 the matching correction processing unit 59 sets the moving body information that was the attention moving body information among the moving body information stored in the result storage unit 56 as processed.
- step S94 the display image generation unit 57 updates and generates an image indicating the search tracking result reflecting the update result of the result storage unit 56, and displays it on the display unit 58. That is, for example, as shown in the upper display screen of FIG. 15, the display corresponding to the plot E is erased and displayed. In FIG. 15, in order to indicate that the display is erased, it is indicated by a cross mark, but in reality, the display itself is erased.
- step S95 the spatiotemporal likelihood calculating unit 93 determines whether or not there is unprocessed moving body information among the moving body information stored in the result storage unit 56 and excluding the reference moving body information. If there is unprocessed moving body information, the process returns to step S86. That is, as long as unprocessed moving body information exists, the processes of steps S86 to S96 are repeated. If it is determined in step S95 that there is no unprocessed moving body information, the process proceeds to step S97.
- step S97 the operation input recognizing unit 91 operates the operation input unit 60 to select a moving body image corresponding to any moving body information, and inputs confirmation information, thereby further matching correction processing. Whether or not is instructed is determined.
- step S97 for example, as shown by the hand H2 in FIG. 15, when the operation input unit 60 is operated, the moving body image PC corresponding to the plot C is selected, and the confirmation information is input,
- step S81 further matching correction processing is executed.
- step S84 the moving body image to which the confirmation information is input becomes two moving body images PA and PC, and the camera ID for identifying the monitoring camera 11 that captured each moving body image is different. Is considered to be computable.
- the BTF calculation unit 92 calculates the BTF using the moving body image PC of moving body information corresponding to the plot C and the moving body image PA of moving body information corresponding to the plot A.
- the reference moving body information is the moving body information of the moving body image PC corresponding to the plot C
- the obtained BTF is based on the color of the moving body image PC captured by the monitoring camera 11 that captures the imaging area CamB.
- the color change of the moving body image PA captured by the monitoring camera 11 that captures the imaging area CamA is corrected.
- step S89 when the target moving body information is moving body information corresponding to the moving body image PB imaged in the imaging area CamA, it is considered that the BTF is obtained in step S89, and therefore the process proceeds to step S90. move on.
- step S90 the BTF image processing unit 95 performs color correction by color-converting the moving object image of the moving object information of interest using the calculated BTF. That is, in this case, the BTF image processing unit 95 applies the BTF to the moving body image PB in FIG. 15 so as to correspond to the color of the monitoring camera 11 that has captured the imaging area CamB of the moving body image PC of the reference moving body information. Correct the color.
- the BTF image likelihood calculating unit 96 calculates the BTF image likelihood that is the likelihood of the moving image of the reference moving body information and the moving image of the moving object information of interest that has undergone color conversion by BTF. Calculate the degree.
- the BTF image likelihood is basically the same as the likelihood in the moving object information likelihood calculation processing unit 54.
- step S92 the BTF image likelihood threshold value determination unit 97 performs the moving image of the reference moving object information and the color conversion by the BTF based on whether or not the calculated BTF image likelihood is higher than a predetermined threshold value. It is determined whether or not the moving body information of the attention moving body information matches.
- step S92 when the BTF image likelihood is higher than a predetermined threshold, the process proceeds to step S93. That is, in this case, the moving body information of the moving body information of interest is left in the result storage unit 56.
- step S92 if the BTF image likelihood is lower than the predetermined threshold value in step S92, the process proceeds to step S96. That is, in this case, the moving body information of the moving body information of interest is deleted from the result storage unit 56 by the process of step S96.
- the imaging area A BTF for correcting the color of the image captured by the monitoring camera 11 that captured the imaging area CamA is obtained on the basis of the color of the image captured by the monitoring camera 11 that captured CamB.
- the moving body information corresponding to the plot B including the moving body image having the spatiotemporal likelihood higher than the predetermined threshold and the BTF image likelihood higher than the predetermined threshold is left in the result storage unit 56.
- the moving body information corresponding to the plot D including the moving body image whose spatiotemporal likelihood is higher than the predetermined threshold but whose BTF image likelihood is lower than the predetermined threshold is deleted from the result storage unit 56.
- the reason why the plot A is hatched in the table in the lower part of FIG. 15 is that it is not subject to processing because it is already moving object information for which confirmation information has been input.
- the matching correction process when the user inputs the confirmation information, the matching correction process is repeatedly performed again based on the confirmation information.
- the accuracy can be improved.
- the BTF can be obtained by inputting the definite information with respect to the moving body image captured by the different monitoring cameras 11, the space between the monitoring cameras 11 is further considered in consideration of the spatiotemporal likelihood. Since the matching correction process is performed based on the BTF image likelihood corresponding to the color change at, person search tracking can be performed with higher accuracy.
- the BTF since the BTF only needs to calculate the BTF corresponding to only two surveillance cameras 11 of the moving image of the reference moving object information and the moving image of the moving object information of interest, the processing load for calculating the BTF is reduced. It becomes possible to reduce and improve the processing speed concerning BTF calculation.
- the person to be searched as a suspicious person or suspicious person check the movement history of the suspicious person or suspicious person from the person search tracking results, and check whether there are actually any suspicious points. It becomes possible. More specifically, for example, when a suspicious person outside the company is found in the company, it is possible to check the movement history by setting the suspicious person as a search target person. If you do not have the password, you can check this if you are in an inaccessible place, and you can use it as a so-called security system.
- the person to be searched for to multiple users of the floor in the store, from what the person search tracking results, what kind of travel route the user of each floor is moving in the floor Can be confirmed, and the information can be reflected in the store. More specifically, for example, the user can check the movement history in the customer's floor, and the product layout can be optimally changed based on the movement history. It can be used.
- Second Embodiment> In the above, the example of obtaining the movement history of the person using the moving object information detected by the moving object detection has been described, but by using the person information detected by the person detection instead of the moving object detection, more The movement history may be obtained with high accuracy.
- an example based on person detection will be described as a second embodiment.
- the configuration of the monitoring system shown in FIG. In the following, components having the same functions as those described with reference to FIGS. 1 to 15 are given the same names and the same reference numerals, and the description thereof will be omitted as appropriate. .
- the basic configuration of the monitoring camera 11 of FIG. 16 is the same as that of the monitoring camera 11 of FIG. 2, but instead of the moving object detection unit 32 and the moving object information output unit 37, a person detection unit 121 and a person The difference is that an information output unit 122 is provided.
- the person detection unit 121 extracts a feature amount from each of the images picked up by the image pickup unit 31 and detects a region where the person is picked up based on the extracted feature amount. For example, the detected person is picked up.
- the image information indicating the person's imaging area is extracted as a person image such that the existing area is 1 and the other areas are 0. More specifically, in the case of using HOG (Histograms of Oriented Gradients) as an image feature amount, the person detection unit 121 performs edge extraction processing on a taken image, An edge image for recognizing the silhouette of the subject is extracted. Then, the person detection unit 121 divides the edge-extracted image into sections in the gradient direction for each local region, and takes a histogram as a feature amount.
- the person detection unit 121 determines whether or not the silhouette is a person based on the feature amount extracted in this way, and when it is determined that the person is a person, the area considered to be a person is set to 1. An image in which other areas are set to 0 is generated and detected as a person image.
- the person information output unit 122 captures a captured image, a camera ID of the image capturing unit 31 that captured the image, a world coordinate of the person, and a person image extracted from an image captured by the image capturing unit 31. Person information including these is generated from the information of the imaging time and is output to the person search and tracking server 12 via the network 13.
- the person search tracking server 12 includes a person information acquisition unit 151, a person information filtering processing unit 152, a person information holding unit 153, a person information likelihood calculation processing unit 154, a person information likelihood threshold determination unit 155, a result storage unit 156, and a display.
- the image generation unit 157, the display unit 158, the matching correction processing unit 159, and the operation input unit 160 are configured. Note that the display image generation unit 157, the display unit 158, and the operation input unit 160 have the same configurations as the display image generation unit 57, the display unit 58, and the operation input unit 60, respectively, and thus description thereof is omitted.
- the person information acquisition unit 151 acquires the person information supplied from the monitoring camera 11 via the network 13 and temporarily stores it, and supplies it to the person information filtering processing unit 152.
- the person information filtering processing unit 152 filters the person information supplied from the person information acquisition unit 151 according to a predetermined condition, extracts only person information suitable for searching for a person, and a person information holding unit In addition to being held in 153, unsuitable person information is discarded. More specifically, the person information filtering processing unit 152 performs filtering based on whether or not the person image included in the person information is an image obtained by capturing the whole body of the person, and only the image obtained by capturing the whole body of the person is stored in the person information. In addition to being held in the holding unit 153, unsuitable person information in which the whole body of the person is not imaged is discarded.
- the person information holding unit 153 holds only the person information including the person image in which the whole body of the person is captured, which is suitable for the person search by the person information filtering processing unit 152, and the person information likelihood calculation processing unit as necessary. 154 and the display image generation unit 157.
- the person information likelihood calculation processing unit 154 is a reference person information that is a search target including a person image of the person information designated as a search target among the person images included in the person information held in the person information holding unit 153. With respect to the person images of other person information, the person image likelihood is calculated for each person information and supplied to the person information likelihood threshold determination unit 155.
- the person information likelihood threshold determination unit 155 determines whether the person image likelihood obtained based on the person image calculated by the person information likelihood calculation processing unit 154 is higher than the threshold, and the person image likelihood.
- the person information including the person image having a high is stored in the result storage unit 156. That is, the person information likelihood threshold determination unit 155 performs person image matching processing based on the person image likelihood of other person information with respect to the reference person information to be searched, and a person image having a high person image likelihood. Is extracted as a matching based on a person image. Then, the person information likelihood threshold determination unit 155 stores the extracted person information in the result storage unit 156 as a matching result with the reference person information that is the search target.
- the matching correction processing unit 159 determines that the user is a human image to be searched based on the person image displayed in the search tracking result displayed on the display unit 158, the operation input unit 160 is operated.
- the matching correction processing is executed based on the confirmation information input in the above.
- the matching correction processing unit 159 executes the matching correction processing again when the confirmation information is input again even after the matching correction processing is executed, and repeats the matching correction every time the confirmation information is input. Execute the process.
- the matching correction processing unit 159 includes the operation input recognition unit 181, the other person information holding unit 182, the same person information holding unit 183, the unique feature selection unit 184, the unique feature likelihood calculating unit 185, the unique feature likelihood threshold value.
- a determination unit 186 and a BTF space-time processing unit 187 are provided.
- the operation input recognizing unit 181 recognizes based on the operation signal of the operation input unit 160 that the confirmation information has been input for the selected person information among the search tracking results displayed on the display unit 158.
- the other person information holding unit 182 searches for a person image of the discarded person information, assuming that the person image has not been confirmed and is not a person image designated as a search target by the BTF space-time processing unit 187. It is stored as a person image of another person different from the target person.
- the same person information holding unit 183 includes person information including a person image for which confirmation information has been input and person information including a person image of a person designated as a search target as person information of the same person as the search target person. Hold as there is.
- the unique feature selection unit 184 learns based on the information of the person image registered in the other person information holding unit 182 that is not the person image of the person to be searched, that is, the person image regarded as the person image of the other person, A feature amount of a person image that is not a search target person as a learning result is extracted. Further, the unique feature selection unit 184 learns based on the information of the person image registered in the same person information holding unit 183 and regarded as a person image of the same person as the person to be searched, and the learning result The feature amount of the person image of the search target person is extracted.
- the unique feature selection unit 184 determines, by learning, a feature amount that has a low possibility of being a person image of another person and that is likely to be a person image of the same person, as a unique feature. And supplied to the unique feature likelihood calculation unit 185.
- the unique feature likelihood calculating unit 185 selects a person image of the person information designated as the search target from among the person images included in the person information held in the person information holding unit 153.
- the unique feature likelihood is calculated for each person information for the person image of the other person information with respect to the reference person information that is the search target, and is supplied to the unique feature likelihood threshold determination unit 186.
- the unique feature likelihood threshold determination unit 186 determines whether or not the unique feature likelihood calculated based on the person image calculated by the unique feature likelihood calculation processing unit 185 is higher than the threshold, and the unique feature likelihood.
- the person information including the person image having a high is stored in the result storage unit 156. That is, the unique feature likelihood threshold value determination unit 186 performs person image matching processing based on the unique feature likelihood of other person information with respect to the reference person information to be searched, and a person image having a high unique feature likelihood. Is extracted as a matching based on a person image. Then, the unique feature likelihood threshold determination unit 186 stores the extracted person information in the result storage unit 156 as a matching result with the reference person information to be searched.
- the BTF spatio-temporal processing unit 187 executes a determination process using the spatio-temporal and BTF images in the same manner as the process described in the first embodiment. More specifically, the BTF spatiotemporal processing unit 187 includes a BTF calculating unit 191, a spatiotemporal likelihood calculating unit 192, a spatiotemporal likelihood threshold determining unit 193, a BTF image processing unit 194, a BTF image likelihood calculating unit 195, and A BTF image likelihood threshold determination unit 196 is provided.
- the BTF calculating unit 191, the spatiotemporal likelihood calculating unit 192, the spatiotemporal likelihood threshold determining unit 193, the BTF image processing unit 194, the BTF image likelihood calculating unit 195, and the BTF image likelihood threshold determining unit 196 are respectively , BTF calculation unit 92, spatiotemporal likelihood calculation unit 93, spatiotemporal likelihood threshold determination unit 94, BTF image processing unit 95, BTF image likelihood calculation unit 96, and BTF image likelihood threshold determination unit 97. Therefore, the description thereof will be omitted.
- step S101 the imaging unit 31 of the monitoring camera 11 continuously captures images composed of still images or moving images in the monitoring area that can be monitored from the installed position.
- step S ⁇ b> 102 the person detection unit 121 extracts a feature amount necessary for determining whether or not a person is captured from each image captured by the imaging unit 31, and based on the extracted feature amount.
- An area where a person is imaged is detected.
- the person detection unit 121 extracts, as a person image, image information indicating a person imaging area in which the area where the detected person is imaged is 1 and the other areas are 0.
- the person detection unit 121 performs edge extraction processing on a taken image, An edge image for recognizing the silhouette of the subject is extracted. Then, the person detection unit 121 divides the edge-extracted image into sections in the gradient direction for each local region, and takes a histogram as a feature amount. The person detection unit 121 determines whether or not the silhouette is a person based on the feature amount extracted in this way, and when it is determined that the person is a person, the area considered to be a person is set to 1. A person image in which other areas are set to 0 is generated and output as a detection result.
- HOG Heistograms of Oriented Gradients
- the person detection unit 121 detects the silhouette of the person as shown by the image P101, and is a feature amount composed of a line segment shown by a solid line. Is superimposed on the captured image P111. At this time, a histogram that is divided into sections in the gradient direction for each local region as indicated by a line segment composed of dotted lines in the image P121 is extracted as a feature amount. Then, the person detection unit 121 determines whether or not the person is a person based on the extracted feature amount. If the person detection unit 121 determines that the person is a person, an image including the person region and other regions is used as a person image. Extract. In the case of FIG.
- the image P111 is regarded as having a person because the feature amount indicated by the dotted line matches the feature amount stored in advance.
- the captured images P112 to P114 are also processed in the same manner, and a histogram divided into sections in the gradient direction for each local region including line segments as shown in the images P122 to P124 is extracted as a feature amount.
- the feature amount extraction method for detecting a person may be a method other than HOG.
- Haar Like feature refer to An Extended Set of Haar-like Features for Rapid Object Detection Rainer Lienhart and Jochen Maydt: IEEE ICIP 2002, Vol. 1, pp. 900-903, Sep. 2002.
- a method of using a plurality of feature amounts extracted by HOG refer to Japanese Unexamined Patent Application Publication No. 2009-301104.
- the imaging position coordinate calculation unit 35 calculates the imaging direction, angle, and distance from the position and size of the person in the human image, and further determines the subject from the world coordinates where the person is installed. Calculate the world coordinates of a person. That is, the imaging position coordinate calculation unit 35 calculates, for example, a coordinate position composed of the latitude and longitude of the person on the earth as the world coordinates based on the person image.
- step S104 the imaging time detection unit 36 detects the time information at the timing when the image is captured as the imaging time based on the time information generated by a real time clock (not shown).
- step S105 the person information output unit 122 reads the camera ID from the camera ID storage unit 34, and generates person information by collecting the person image, world coordinates, and imaging time.
- step S106 the person information output unit 122 outputs the generated person information to the person search and tracking server 12 via the network 13 represented by the Internet.
- an image is captured for each monitoring area in each of the monitoring cameras 11, a person in the captured image is detected, and a person image is extracted. Then, along with the person image, person information including the world coordinates of the person in the person image and the information of the imaging time at which the image was captured is generated and supplied to the person search and tracking server 12.
- step S121 the person information acquisition unit 151 acquires and stores the person information sequentially supplied from the monitoring server 11 via the network 13.
- step S122 the person information filtering processing unit 152 sets any unprocessed person information among the person information stored in the person information acquisition unit 151 as the target person information to be processed.
- step S123 the person information filtering processing unit 152 determines whether or not it is a whole body image of a single person from the silhouette of the person image included in the person-of-interest information.
- step S123 when it is determined from the silhouette of the person image included in the person-of-interest information that the image is a whole body image of a single person, the process proceeds to step S124.
- step S124 the person information filtering processing unit 152 regards the target person information as the person information suitable for the person search, and causes the person information holding unit 153 to hold it.
- step S123 when it is determined from the silhouette of the person image included in the person-of-interest information that the image is not a whole body image of a single person, the process proceeds to step S125.
- step S125 the person information filtering processing unit 152 considers that the attention person information is not person information suitable for person search, and discards the person information set in the acquired attention person information.
- step S126 the person information filtering processing unit 152 determines whether or not unprocessed person information exists in the person information acquisition unit 151. If unprocessed person information exists, the process returns to step S122. . That is, the processes in steps S122 to S126 are repeated until there is no unprocessed person information. If it is determined in step S126 that there is no unprocessed person information, the process ends.
- step S151 the display image generation unit 157 displays a list of person images among the person information held in the person information holding unit 153 on the display unit 158, and displays a person image that designates a person to be searched. A display image prompting selection is displayed.
- step S152 the person information likelihood calculation processing unit 154 determines whether or not the operation input unit 160 has been operated to select a person image specifying a person to be searched. If a person image specifying a person to be searched is not selected, the process returns to step S151. That is, the processes of steps S151 and S152 are repeated until a person image that specifies a person to be searched is specified. In step S152, if a person image that specifies a person to be searched is specified, the process proceeds to step S153.
- step S153 the person information likelihood calculation processing unit 154 sets the person information corresponding to the person image designated as the search target among the person information held in the person information holding unit 153 as reference person information, Other person information is set as unprocessed person information.
- the operation input recognizing unit 181 causes the same person information holding unit 183 to hold the person information corresponding to the person image designated as the search target based on the operation signal of the operation input unit 160.
- step S154 the person information likelihood calculation processing unit 154 reads any unprocessed person information held in the person information holding unit 153 and sets it as attention person information.
- the person information likelihood calculation processing unit 154 calculates a person image likelihood that is a likelihood of the person image included in the reference person information and the person image included in the attention person information, It supplies to person information likelihood threshold value determination part 155 with attention person information. That is, the person information likelihood calculation processing unit 154, for example, a person based on a similarity degree that becomes a higher value as the person captured in each person image of the reference person information and the attention person information is similar. Image likelihood is calculated.
- step S156 the person information likelihood threshold determination unit 155 determines whether or not the supplied person image likelihood is higher than a predetermined threshold. In other words, based on whether the person image likelihood is higher than a predetermined threshold, the person information likelihood threshold determination unit 155 determines whether the person images of the reference person information and the attention person information match. Determine whether or not. If it is determined in step S156 that the person image likelihood is higher than the predetermined threshold and the person images of the reference person information and the person of interest information match (match), the process proceeds to step S157. move on.
- step S157 the person information likelihood threshold value determination unit 155 stores the attention person information in the result storage unit 156. At this time, the person information likelihood calculation processing unit 154 sets the attention person information held in the person information holding unit 153 as the processed person information.
- step S156 determines whether the person image likelihood is higher than the predetermined threshold value and the person in the person image of the reference person information and the person of interest information does not match (does not match). If it is determined in step S156 that the person image likelihood is not higher than the predetermined threshold value and the person in the person image of the reference person information and the person of interest information does not match (does not match), the process is as follows. The process proceeds to step S159.
- step S159 the person information likelihood threshold determination unit 155 discards the attention person information. Then, the person information likelihood calculation processing unit 154 sets the attention person information held in the person information holding unit 153 as processed person information.
- step S158 the person information likelihood calculation processing unit 154 determines whether or not unprocessed person information is stored in the person information storage unit 153. For example, when unprocessed person information is stored, The process returns to step S154. In other words, the processes in steps S154 to S159 are repeated until there is no unprocessed person information in the person information holding unit 153, and matching in the person image with the reference person information is performed for all the person information. The process is repeated.
- step S158 If it is determined in step S158 that unprocessed person information is not held in the person information holding unit 153, the process proceeds to step S160.
- step S160 the display image generation unit 157 generates a display image indicating the search tracking result based on the person information stored in the result storage unit 156.
- step S161 the display image generation unit 157 displays a display image indicating the search tracking result on the display unit 158.
- step S ⁇ b> 162 the matching correction processing unit 159 selects one of the human images indicated by the display image based on the display image information supplied from the display image generation unit 157 and the operation signal supplied from the operation input unit 160. Is selected, confirmation information is input, and it is determined whether matching correction processing is instructed.
- step S162 when any one of the person images shown in the display image is selected and the confirmation information is input, it is considered that the matching correction process has been instructed, and the process proceeds to step S163.
- step S163 the matching correction processing unit 159 executes the matching correction process, and displays the person search tracking result on which the correction process has been performed. Details of the matching correction processing will be described later with reference to FIGS.
- step S164 the matching correction processing unit 159 determines whether or not the end is instructed. If the end is not instructed, the process returns to step S161. That is, the display screen showing the person search tracking result is continuously displayed on the display unit 158. In step S164, when an end instruction is given, the process ends.
- the person information matching the person to be searched is searched based on the person image, and the person information as the search result is searched. It becomes possible to display the search tracking result indicating the movement path of the person to be searched.
- a person image of the person information held in the person information holding unit 153 has been described in order to specify a person to be searched.
- the person to be searched may be specified by handling the same image as the person image.
- step S181 the matching correction processing unit 159 stores information indicating that the person image for which the confirmation information is input by the operation input unit 160 is included in association with the person information. Since the person information to which the confirmation information is input is confirmed to be the person information appropriately extracted by the person image matching process, it is excluded from the search target in the subsequent processes, and the result storage unit 153 Will continue to be remembered. At the same time, the operation input recognition unit 183 causes the same person information holding unit 183 to hold the person information to which the confirmation information is input.
- step S182 the matching correction processing unit 159 sets all of the personal information stored in the result storage unit 156, for which the confirmation information is not input, as unprocessed personal information.
- step S183 the matching correction processing unit 159 sets, as reference personal information, the personal information for which the final confirmed information is input among the personal information stored in the result storage unit 156.
- step S184 the BTF calculation unit 191 of the BTF space-time processing unit 187 determines whether or not BTF can be calculated.
- the condition under which the BTF can be calculated is, first, that the camera ID is different from the camera ID included in the reference person information, and that the person information including another person image to which the confirmation information is input is included. There is a time. A second case is when there are person images to which two or more pieces of confirmation information are input, each of which is taken by the monitoring camera 11 having a different camera ID. Therefore, the BTF calculation unit 92 determines whether or not the BTF can be calculated based on whether or not personal information that satisfies one of the above two conditions exists in the result storage unit 156.
- step S185 the BTF calculation unit 191 uses a pixel value between human images included in the human image or reference personal information to which the confirmation information is input although the camera ID is different. Using this information, BTF calculation processing is executed to calculate BTF.
- BTF is a function that indicates the mutual correspondence of colors between human images including the same subject imaged by the monitoring camera 11 specified by two different camera IDs. Therefore, by calculating this BTF, any one of the human images captured by the two monitoring cameras 11 having different camera IDs regardless of the presence or absence of the definite information can be color-converted using the BTF to obtain the color. It becomes possible to correct.
- this color correction processing it is possible to correct a color change caused by a difference in imaging environment light in a monitoring area between different monitoring cameras 11 or a solid difference in imaging elements of the imaging unit 31.
- the color of the person images captured by the two different monitoring cameras 11 is corrected as if they were captured by the same monitoring camera 11, so that the person image likelihood between the person images can be more accurately determined. It is possible to calculate.
- step S184 for example, when the person image to which the confirmation information is input is only captured by one type of monitoring camera 11, it is considered that BTF cannot be calculated, and the process of step S185 is skipped. Is done.
- the spatiotemporal likelihood calculating unit 192 is the person information stored in the result storage unit 156, and among the person information excluding the reference person information, any one of the unprocessed person information is the target person information. Set to.
- the spatiotemporal likelihood calculation unit 192 calculates a spatiotemporal likelihood based on the reference person information and the information on the world coordinates and the imaging time included in the attention person information. More specifically, the spatiotemporal likelihood calculation unit 192 obtains a movement distance from the difference between the world coordinates included in the reference person information and the world coordinates included in the person-of-interest information, for example, and calculates the movement distance as an average of humans. The spatio-temporal likelihood of the person information is calculated based on the ratio of the time between the imaging times of the person information with respect to the average required time required by the typical moving speed.
- step S188 the spatiotemporal likelihood threshold determination unit 94 determines whether or not the calculated spatiotemporal likelihood is higher than a predetermined threshold. For example, when the display image showing the person search tracking result is the display image shown in the upper part of FIG. 24, the operation input unit 160 selects the person image PA corresponding to the plot A as shown by the hand H1. Thus, when the confirmation information is input, the person information corresponding to the plot A becomes the reference person information. In the case of the display image of FIG. 25, in the person information corresponding to the plots A to G, the imaging times are 1:00:00, 1:00:05, 1:00:10, 1:00:15, 1:00:13, 1:00:14, and 1:00:05.
- the plots A and B are person information corresponding to the person images PA and PB captured by the same monitoring camera 11 that captures the imaging area CamA indicated by the triangle formed by the same dotted line.
- the plots C to F are person information corresponding to the person images PC to PF imaged by the same monitoring camera 11 that images the imaging area CamB indicated by a dotted triangle.
- the plot G is person information corresponding to the person image PE imaged by the monitoring camera 11 that images the imaging area CamC indicated by a triangle formed by a dotted line.
- the distance between the plots AB is the distance AB
- the movement distance with respect to the average required time required by the average movement speed of the human is 5 / (AB / w).
- w is an average moving speed of a human. For example, if the distance AB is 5 m and the average moving speed of a human is 1 m / s, the ratio is 1, and the spatiotemporal likelihood is the highest value.
- the movement distance is 10 / (AC / w), 15 / (AD / w), 14 / (AE / w), respectively. , 13 / (AF / w).
- the ratios are all 1, and space-time The likelihood is the highest value.
- step S188 when the spatiotemporal likelihood is the highest value as described above, the spatiotemporal likelihood threshold determination unit 193 considers that the spatiotemporal likelihood threshold is higher than the predetermined threshold, and the process proceeds to step S189.
- the person information corresponding to the plot G is the attention person information
- the person can be compared with the average required time required by the average movement speed of the person.
- the ratio of time between image capturing times of information is 5 / (AE / w). That is, for example, if the distance AG is 25 m and the average human moving speed is 1 m / s, 25 m is moved in about 5 seconds, so the ratio is 0.2.
- the spatiotemporal likelihood is a low value.
- the spatiotemporal likelihood threshold determination unit 193 considers that the spatiotemporal likelihood threshold is lower than the predetermined threshold, and the process proceeds to step S198.
- step S198 the matching correction processing unit 159 deletes the person of interest information from the result storage unit 156 and causes the other person information holding unit 182 to hold it. That is, when the image indicating the search tracking result as shown in the upper part of FIG. 24 is displayed, when the confirmation information is input to the person image PA corresponding to the plot A, the image is displayed based on the spatio-temporal information. As shown in the table below the 24 display images, the person information of the plot G where the same person cannot exist is deleted. By such processing, it becomes possible to eliminate search tracking results that are false detections that occur in the person information matching processing using only human images, and to realize search tracking of a person to be searched with higher accuracy. It becomes possible.
- step S189 the BTF image processing unit 194 determines whether or not BTF is obtained. For example, if BTF is not obtained, the process proceeds to step S197.
- step S197 the matching correction processing unit 159 sets the personal information that has been the attention personal information among the personal information stored in the result storage unit 156 as processed.
- step S199 the display image generation unit 157 updates and generates an image indicating the search tracking result, reflecting the update result of the result storage unit 156, and displays it on the display unit 158. That is, for example, as shown in the upper display screen of FIG. 25, the display corresponding to the plot G is erased and displayed. In FIG. 25, a cross mark is used to indicate that the display is erased, but the display itself is actually erased.
- step S200 the spatiotemporal likelihood calculation unit 192 determines whether or not unprocessed person information exists among the person information stored in the result storage unit 156 and excluding the reference person information. If it is determined that there is unprocessed person information, the process returns to step S186 (FIG. 22). That is, as long as unprocessed person information exists, the processes of steps S186 to S200 are repeated. If it is determined in step S200 that there is no unprocessed person information, the process proceeds to step S201.
- step S201 the operation input recognizing unit 181 operates the operation input unit 160, selects a person image corresponding to any person information, and inputs confirmation information, thereby further matching correction processing. Whether or not is instructed is determined.
- step S201 for example, as shown by the hand H2 in FIG. 25, when the operation input unit 160 is operated, the person image PC corresponding to the plot C is selected, and the confirmation information is input, the process is as follows. Returning to step S181 (FIG. 22), further matching correction processing is executed.
- step S184 since the person images to which the confirmation information is input are two images of the person images PA and PC, and the camera IDs for identifying the monitoring cameras 11 that captured the person images are different, the BTF is different. Is considered to be computable.
- step S185 the BTF calculating unit 191 calculates the BTF using the person image PC of the person information corresponding to the plot C and the person image PA of the person information corresponding to the plot A.
- the reference person information is the person information of the person image PC corresponding to the plot C
- the obtained BTF is based on the color of the person image PC imaged by the monitoring camera 11 that images the imaging area CamB.
- the color change of the person image PA captured by the monitoring camera 11 that captures the imaging area CamA is corrected.
- step S189 when the person-of-interest information is person information corresponding to the person image PB imaged in the imaging area CamA, it is considered that BTF is obtained in step S189. Therefore, the process proceeds to step S190. move on.
- step S190 the BTF image processing unit 194 performs color correction by color-converting the person image of the person-of-interest information using the calculated BTF.
- the BTF image processing unit 194 applies the BTF to the person image PB in FIG. 25 so as to correspond to the color of the monitoring camera 11 that has captured the imaging area CamB of the person image PC of the reference person information. Correct the color.
- step S191 the BTF image likelihood calculation unit 196 calculates the BTF image likelihood that is the likelihood of the person image of the reference person information and the person image of the person-of-interest information that has undergone color conversion by BTF. Calculate the degree.
- the BTF image likelihood is basically the same as the likelihood in the person information likelihood calculation processing unit 154.
- step S192 the BTF image likelihood threshold determination unit 97 performs color conversion using the person image of the reference person information and BTF based on whether or not the calculated BTF image likelihood is higher than a predetermined threshold. It is determined whether or not the person information of the noticed person information matches.
- step S192 when the BTF image likelihood is higher than the predetermined threshold, the process proceeds to step S193.
- step S192 if the BTF image likelihood is lower than the predetermined threshold in step S192, the process proceeds to step S198. That is, in this case, the person information of the person-of-interest information is deleted from the result storage unit 156 and held in the other person information holding unit 182 by the process of step S198.
- the imaging area A BTF for correcting the color of the image captured by the monitoring camera 11 that captured the imaging area CamA is obtained on the basis of the color of the image captured by the monitoring camera 11 that captured CamB.
- the person information corresponding to the plot B including the person image having the spatiotemporal likelihood higher than the predetermined threshold and the BTF image likelihood higher than the predetermined threshold is left in the result storage unit 156.
- the person information corresponding to the plot D including the person image whose spatiotemporal likelihood is higher than the predetermined threshold but the BTF image likelihood is lower than the predetermined threshold is deleted from the result storage unit 156 and the other person information is retained. Held in the portion 182.
- the reason why the plot A is hatched in the table in the lower part of FIG. 25 is that it is not subject to processing because it is personal information for which confirmed information has already been input.
- the unique feature selection unit 184 selects a unique feature by learning based on the person information stored in the other person information storage unit 182 and the person information stored in the same person information storage unit 183. .
- any of the images P211 to P214 in FIG. 26 is characterized by any of the line segments as indicated by images P221 to P224, respectively. Is detected as a person present.
- the unique feature selection unit 184 further analyzes the feature amount information indicated by the line segment by learning, and selects the feature amount that is highly likely to be matched as the same person and is less likely to be matched as another person. The unique feature is selected and the selection result is supplied to the unique feature calculation unit 185.
- the feature amount indicated by the dotted line segment of the images P221 and P222 is the same person. Is specified by learning as a feature quantity that has a high possibility of being matched.
- the feature quantities surrounded by solid line segments in the images P223 and P224 are not the same person. A feature amount that is unlikely to be suitable as another person is identified by learning. Accordingly, in FIG. 26, by such learning, the feature quantity composed of the line segment surrounded by the dashed-dotted ellipse in the image P251 is selected as a useful unique feature that satisfies both conditions.
- the learning is repeated each time new person information is registered in the same person information holding unit 183 and the other person information holding unit 182, so that unique features having higher accuracy are selected. Therefore, each time learning is repeated, the tracking accuracy of the person is improved.
- step S194 the unique feature likelihood calculating unit 185 extracts the feature amount selected as the unique feature from each of the person image of the reference person information and the person image of the target person information.
- step S195 the unique feature likelihood calculating unit 185 calculates the unique feature likelihood using the feature amount extracted as the unique feature from each of the person image of the reference person information and the person image of the target person information.
- the result is supplied to the unique feature likelihood threshold determination unit 186. That is, the unique feature likelihood calculating unit 185 calculates, for example, the mutual similarity based on the unique feature from the person image of the reference person information and the person image of the target person information as the unique feature likelihood.
- step S196 the unique feature likelihood threshold determination unit 186 determines whether or not the calculated unique feature likelihood is higher than a predetermined threshold and they are similar to each other. If it is determined in step S196 that the unique feature likelihood is not higher than the predetermined threshold and is not similar, the process proceeds to step S198.
- the person information of the person-of-interest information is deleted from the result storage unit 156 and held in the other person information holding unit 182 by the process of step S198.
- step S196 determines whether the unique feature likelihood is higher than the predetermined threshold. If it is determined in step S196 that the unique feature likelihood is higher than the predetermined threshold, the process proceeds to step S197.
- the person information of the person-of-interest information is left in the result storage unit 156.
- the imaging area A BTF for correcting the color of the image captured by the monitoring camera 11 that captured the imaging area CamA is obtained on the basis of the color of the image captured by the monitoring camera 11 that captured CamB.
- the person information corresponding to the plots B, D, and E including the person images having the spatiotemporal likelihood higher than the predetermined threshold and the BTF image likelihood higher than the predetermined threshold is left in the result storage unit 156. .
- the person information corresponding to the plot D including the person image whose spatiotemporal likelihood is higher than the predetermined threshold but the BTF image likelihood is lower than the predetermined threshold is deleted from the result storage unit 156 and the other person information is retained. Held in the portion 182. Furthermore, by obtaining the unique feature likelihood using the person information, the plot D having the unique feature likelihood equal to or smaller than the threshold is deleted, and the plots B and F are finally left in the result storage unit 156. .
- the matching correction process when the user inputs the confirmation information, the matching correction process is repeatedly performed again based on the confirmation information.
- the accuracy can be improved.
- the BTF can be obtained by inputting the deterministic information with respect to the person image captured by the different monitoring cameras 11, the space between the monitoring cameras 11 is further considered in consideration of the spatiotemporal likelihood.
- the BTF since the BTF only needs to calculate the BTF corresponding to only two monitoring cameras 11 of the person image of the reference person information and the person image of the attention person information, the processing load for calculating the BTF is increased. It becomes possible to reduce and improve the processing speed concerning BTF calculation.
- the matching correction processing is repeated based on the unique feature likelihood by the unique feature obtained by learning, thereby improving the accuracy of the unique feature, and as a result, with higher accuracy. Person search tracking can be realized.
- the person to be searched as a suspicious person or suspicious person check the movement history of the suspicious person or suspicious person from the person search tracking results, and check whether there are actually any suspicious points. It becomes possible. More specifically, for example, when a suspicious person outside the company is found in the company, it is possible to check the movement history by setting the suspicious person as a search target person. If you do not have the password, you can check this if you are in an inaccessible place, and you can use it as a so-called security system.
- the person to be searched for to multiple users of the floor in the store, from what the person search tracking results, what kind of travel route the user of each floor is moving in the floor Can be confirmed, and the information can be reflected in the store. More specifically, for example, the user can check the movement history in the customer's floor, and the product layout can be optimally changed based on the movement history. It can be used.
- the above-described series of processing can be executed by hardware, but can also be executed by software.
- a program constituting the software may execute various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed from a recording medium in a general-purpose personal computer or the like.
- FIG. 16 shows a configuration example of a general-purpose personal computer.
- This personal computer incorporates a CPU (Central Processing Unit) 1001.
- An input / output interface 1005 is connected to the CPU 1001 via a bus 1004.
- a ROM (Read Only Memory) 1002 and a RAM (Random Access Memory) 1003 are connected to the bus 1004.
- the input / output interface 1005 includes an input unit 1006 including an input device such as a keyboard and a mouse for a user to input an operation command, an output unit 1007 for outputting a processing operation screen and an image of the processing result to a display device, programs, and various types.
- a storage unit 1008 including a hard disk drive for storing data, and a communication unit 1009 configured to perform communication processing via a network represented by the Internet are connected to a LAN (Local Area Network) adapter or the like.
- LAN Local Area Network
- magnetic disks including flexible disks
- optical disks including CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc)), magneto-optical disks (including MD (Mini Disc)), or semiconductors
- a drive 1010 for reading / writing data from / to a removable medium 1011 such as a memory is connected.
- the CPU 1001 is read from a program stored in the ROM 1002 or a removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, installed in the storage unit 1008, and loaded from the storage unit 1008 to the RAM 1003. Various processes are executed according to the program.
- the RAM 1003 also appropriately stores data necessary for the CPU 1001 to execute various processes.
- the CPU 1001 loads the program stored in the storage unit 1008 to the RAM 1003 via the input / output interface 1005 and the bus 1004 and executes the program, for example. Is performed.
- the program executed by the computer (CPU 1001) can be provided by being recorded on the removable medium 1011 as a package medium, for example.
- the program can be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.
- the program can be installed in the storage unit 1008 via the input / output interface 1005 by attaching the removable medium 1011 to the drive 1010. Further, the program can be received by the communication unit 1009 via a wired or wireless transmission medium and installed in the storage unit 1008. In addition, the program can be installed in the ROM 1002 or the storage unit 1008 in advance.
- the program executed by the computer may be a program that is processed in time series in the order described in this specification, or in parallel or at a necessary timing such as when a call is made. It may be a program for processing.
- the system means a set of a plurality of components (devices, modules (parts), etc.), and it does not matter whether all the components are in the same housing. Accordingly, a plurality of devices housed in separate housings and connected via a network and a single device housing a plurality of modules in one housing are all systems. .
- the present technology can take a cloud computing configuration in which one function is shared by a plurality of devices via a network and is jointly processed.
- each step described in the above flowchart can be executed by one device or can be shared by a plurality of devices.
- the plurality of processes included in the one step can be executed by being shared by a plurality of apparatuses in addition to being executed by one apparatus.
- this technique can also take the following structures. (1) Taking an image, detecting a moving object, extracting a moving object image including the detected moving object image, detecting a spatial position coordinate of the moving object based on the moving object image, and detecting the moving object image and the moving object A plurality of imaging units that output moving body information including the spatial position coordinates of the image and the imaging time when the image was captured; A moving object that calculates a moving object image likelihood that is a likelihood of a moving object image included in moving object information other than the search contrast moving object information with respect to a moving object image of search object moving object information that is moving object information including a moving object image of a moving object to be searched An image likelihood calculating unit; For each of the moving image likelihoods calculated by the moving image likelihood calculating unit, it is determined whether or not the moving image likelihood is higher than a predetermined threshold, and the moving object information of the moving image likelihood higher than the predetermined threshold is searched for A moving object image threshold value determining unit that searches as moving object information that is moving object information including a moving object
- An input section Of the moving body information stored as the search result moving body information in the result moving body information storage unit, the fixed moving body information that is the moving body information other than the fixed moving body information that is the moving body information that has been input with the deterministic information, A spatio-temporal likelihood calculation unit that calculates a spatio-temporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time for information; It is determined whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculating unit is lower than a predetermined threshold, and moving object information having a spatiotemporal likelihood lower than the predetermined threshold is obtained as the search result.
- An information processing apparatus comprising: a spatiotemporal likelihood threshold determination unit that is deleted from the moving object information storage unit.
- the operation input unit is determined by the spatiotemporal likelihood threshold determination unit whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold. Thereafter, out of the moving body information stored as the search result moving body information in the result moving body information storage unit, based on the moving body image, the user confirms that the information to be newly determined is the search target moving body information. Accept input, The spatio-temporal likelihood calculation unit is a confirmed unit in which the confirmed information of the moving body information other than the confirmed moving body information is newly input among the moving body information stored as the search result moving body information in the resultant moving body information storage unit.
- a new spatiotemporal likelihood consisting of likelihoods based on the spatial position coordinates and imaging time for moving object information is calculated,
- the spatiotemporal likelihood threshold determination unit determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and is lower than the predetermined threshold.
- the moving body information of space-time likelihood is deleted from the search result moving body information storage unit,
- the operation input unit, the spatiotemporal likelihood calculation unit, and the spatiotemporal likelihood threshold determination unit repeat the same processing every time new information is input by the operation input unit.
- the moving body information further includes an ID for identifying any of the plurality of imaging units that captured the moving body image included therein, Each of the moving objects among the search object moving object information that is moving object information including the moving object image of the moving object to be searched, and the confirmed moving object information that has received the input of the deterministic information for confirming that it is the search object moving object information.
- a BTF calculation unit that calculates a BTF (Brightness Transfer Function) for correcting a color change between the imaging units based on the two moving object images having different IDs for identifying a plurality of imaging units that have captured the images;
- a BTF Bitness Transfer Function
- a BTF processing unit that applies BTF to a moving object image of moving object information,
- a BTF moving image likelihood that calculates a BTF moving image likelihood including a likelihood based on the moving image of moving object information including the moving image subjected to BTF by the BTF processing unit with respect to the moving image of the definite moving information.
- a calculation unit For each of the BTF moving image likelihood calculated by the BTF moving image likelihood calculation unit, further includes a BTF moving image threshold determination unit that determines whether or not it is lower than a predetermined threshold, The spatiotemporal likelihood threshold determination unit determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and lower than the predetermined threshold If not, the BTF moving image threshold determination unit determines whether each of the BTF moving image likelihoods calculated by the BTF moving image likelihood calculation unit is lower than a predetermined threshold, and the BTF moving image image When the likelihood is lower than a predetermined threshold, the moving body information including the moving image whose BTF moving body image likelihood is lower than the predetermined threshold is deleted from the search result moving body information storage unit.
- the moving object image likelihood calculating unit includes a moving object image of moving object information that is moving object information including a moving object image of a moving object to be searched, and a moving object image included in moving object information other than the search contrast moving object information.
- the information processing apparatus according to (1) or (2), wherein a similarity indicating how much each moving object is similar is calculated as the moving image likelihood based on the included moving image.
- the spatio-temporal likelihood calculation unit calculates an average human movement based on a distance between the spatial position coordinates of moving body information other than the fixed moving body information and the fixed moving body information input with the fixed information.
- the information processing apparatus according to any one of (1), (2), and (4), wherein the spatiotemporal likelihood is calculated from a relationship between a required time when moving at a speed and a time between imaging times.
- An image is picked up, a moving object is detected, a moving object image including the detected moving object image is extracted, a spatial position coordinate of the moving object is detected based on the moving object image, and the moving object image and the moving object are detected.
- a moving object that calculates a moving object image likelihood that is a likelihood of a moving object image included in moving object information other than the search contrast moving object information with respect to a moving object image of search object moving object information that is moving object information including a moving object image of a moving object to be searched Image likelihood calculation processing; For each of the moving image likelihoods calculated by the moving image likelihood calculation process, it is determined whether or not the moving image information is higher than a predetermined threshold, and the moving object information having a moving image likelihood higher than the predetermined threshold is searched for.
- a moving object image threshold determination process for searching as moving object information that is moving object information including a moving object image of the same moving object as the moving object of the moving object image of the target moving object information; Search result moving body information storage processing for storing moving body information searched as search result moving body information by the moving body image threshold determination processing; An operation of accepting input of confirmation information for confirming that it is the search target moving body information by the user based on the moving body image among the moving body information stored as the search result moving body information in the result moving body information storing process.
- Input processing Of the moving body information stored as search result moving body information in the result moving body information storage process, the fixed moving body information that is the moving body information other than the fixed moving body information that is the moving body information that has been input with the deterministic information.
- a spatiotemporal likelihood calculation process for calculating a spatiotemporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time for information; It is determined whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation process is lower than a predetermined threshold, and moving object information having a spatiotemporal likelihood lower than the predetermined threshold is obtained as the search result.
- a spatiotemporal likelihood threshold determination process for deleting from the moving object information stored by the moving object information storage process. (7) An image is picked up, a moving object is detected, a moving object image including the detected moving object image is extracted, a spatial position coordinate of the moving object is detected based on the moving object image, and the moving object image and the moving object are detected.
- the user inputs confirmation information for confirming that the moving body information is the search target moving body information.
- Taking an image, detecting a person, extracting a person image composed of the detected person image, detecting a spatial position coordinate of the person based on the person image, and detecting the person image and the person A plurality of imaging units that output human information including the spatial position coordinates of the image and the imaging time when the image was captured;
- the person who calculates the person image likelihood that is the likelihood of the person image included in the person information other than the search reference person information with respect to the person image of the search target person information that is the person information including the person image of the person to be searched An image likelihood calculating unit; For each of the person image likelihoods calculated by the person image likelihood calculating unit, it is determined whether or not the person image likelihood is higher than a predetermined threshold, and the person information having a person image likelihood higher than the predetermined threshold is searched for A person image threshold value determination unit for searching as search result person information that is person information including a person image of the same person as the person image of the person information of the target person information;
- a search result person information storage unit for storing person
- An input section Of the person information stored as the search result person information in the result person information storage unit, the confirmed person to which the confirmed information is input of the person information other than the confirmed person information that is the person information to which the confirmed information is input
- a spatio-temporal likelihood calculation unit that calculates a spatio-temporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time for information; It is determined whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and the person information having a spatiotemporal likelihood lower than the predetermined threshold is obtained as the search result.
- An information processing apparatus comprising: a spatiotemporal likelihood threshold determination unit that is deleted from the person information storage unit.
- the operation input unit is determined by the spatiotemporal likelihood threshold determination unit whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold. After that, out of the person information stored as the search result person information in the result person information storage unit, based on the person image, a confirmation information for newly confirming that the person information to be searched is newly determined by the user Accept input, The spatio-temporal likelihood calculation unit is a confirmed information in which the confirmed information is newly inputted of person information other than the confirmed person information among the person information stored as the search result person information in the result person information storage unit.
- a new spatio-temporal likelihood composed of likelihoods based on the spatial position coordinates and the imaging time for the person information determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and is lower than the predetermined threshold.
- the person information of space-time likelihood is deleted from the search result person information storage unit,
- the operation input unit, the spatiotemporal likelihood calculation unit, and the spatiotemporal likelihood threshold determination unit repeat the same processing every time new confirmation information is input by the operation input unit.
- Information processing device is
- the person information further includes an ID for identifying any of the plurality of imaging units that captured the included person image,
- Each of the search target person information which is personal information including the person image of the person to be searched, and the confirmed person information for which input of the confirmation information for confirming that the search target person information is accepted
- a BTF calculating unit that calculates a BTF (Brightness Transfer Function) for correcting a color change between the imaging units based on the two human images having different IDs for identifying a plurality of imaging units that have captured images;
- the person image captured by the imaging unit of the ID for which the BTF is required A BTF processing unit that applies BTF to the person image of the included person information;
- BTF person image likelihood for calculating BTF person image likelihood composed of likelihoods based on the person image of person information including the person image subjected to BTF by the BTF processing unit with respect to the person image of the confirmed person information
- a calculation unit for calculating BTF person image likelihood composed of likelihoods based on the person image of person information including the person
- (9) apparatus (11) Search target person information that is person information including a person image of the person to be searched, and confirmed person information for which input of confirmation information for confirming that the search target person information is accepted, The same person information holding unit that holds the same person information as the person to be searched; Each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, or about each of the BTF human image likelihood calculated by the BTF human image likelihood calculation unit A person information including a person image lower than the predetermined threshold; a person information holding unit for holding person information, which is person information of another person from the search target person; A unique feature for searching for the search target person is selected by learning based on the person image in the person information held in the same person information holding unit and the person image in the person information held in the other person information holding unit.
- a unique feature search unit to A unique feature likelihood calculating unit that calculates a unique feature likelihood that is a likelihood based on the unique feature of a person image included in person information other than the search reference person information with respect to a person image of the search target person information; , It is determined whether or not each of the unique feature likelihoods calculated by the unique feature likelihood calculating unit is lower than a predetermined threshold value, and the personal information having a lower characteristic feature likelihood than the predetermined threshold value is obtained as the search result.
- a unique feature likelihood threshold determination unit to be deleted from the person information storage unit determines whether each of the spatiotemporal likelihood newly calculated by the spatiotemporal likelihood calculation unit is lower than a predetermined threshold, and lower than the predetermined threshold If not, the BTF person image threshold determination unit determines whether each of the BTF person image likelihoods calculated by the BTF person image likelihood calculation unit is lower than a predetermined threshold, and the BTF person image When the likelihood is not lower than a predetermined threshold, the unique feature likelihood threshold determination unit determines whether or not each of the unique feature likelihoods calculated by the unique feature likelihood calculation unit is lower than a predetermined threshold.
- the information processing apparatus wherein determination is performed and person information having a unique feature likelihood lower than the predetermined threshold is deleted from the search result person information storage unit.
- the unique feature likelihood threshold determination unit deletes the personal information having a unique feature likelihood lower than the predetermined threshold from the search result personal information storage unit, the other person information holding unit (11)
- the information processing apparatus according to (11), wherein person information having a unique feature likelihood lower than a threshold value is held as other person information that is person information of another person.
- the unique feature retrieval unit performs the same by performing learning based on the person image in the person information held in the same person information holding unit and the person image in the person information held in the other person information holding unit.
- the feature amount is such that the likelihood of the person image in the person information held by the person information holding unit and the person image of the person to be searched is high, and is held by the other person information holding unit.
- the information processing apparatus according to (8), wherein a feature quantity that reduces a likelihood of a person image in person information and a person image of the search target person is selected as a unique feature.
- the unique feature likelihood calculating unit includes a person image of search target person information that is person information including a person image of a person to be searched, and a person image included in person information other than the search reference person information.
- the information processing apparatus wherein similarity indicating how similar each person is based on the unique feature of the person image included in each of the images is calculated as the unique feature likelihood.
- the person image likelihood calculating unit includes a person image of search target person information that is person information including a person image of a person to be searched, and a person image included in person information other than the search reference person information.
- a similarity indicating how similar each person is based on the person images included in each is calculated as the person image likelihood.
- the spatio-temporal likelihood calculating unit calculates an average human movement based on a distance between the spatial position coordinates between the person information other than the confirmed person information and the confirmed person information to which the confirmed information is input.
- a person image threshold value determination process for searching as search result person information that is person information including a person image of the same person as the person image of the person information of the target person information;
- a search result person information storage process for storing the person information searched as the search result person information by the person image threshold determination process;
- Input processing Of the person information stored as the search result person information by the result person information storage process, the confirmed person to which the confirmed information is input, of the person information other than the confirmed person information that is the person information to which the confirmed information is input
- a spatiotemporal likelihood calculation process for calculating a spatiotemporal likelihood composed of likelihoods based on the spatial position coordinates and imaging time for information; It is determined whether or not each of the spatiotemporal likelihoods calculated by the spatiotemporal likelihood calculation process is lower than a predetermined threshold, and the person information having a spatiotemporal likelihood lower than the predetermined threshold is obtained as the search result.
- An information processing method comprising: a spatiotemporal likelihood threshold determination process for deleting the person information stored by the person information storage process.
Abstract
Description
1. 第1の実施の形態(動体検出を用いた一例)
2. 第2の実施の形態(人物検出を用いた一例)
[監視システムの構成例]
図1は、本技術を適用した監視システムの一実施の形態の構成例を示している。図1の監視システム1は、人物の検索や追尾が必要となる監視エリア内における複数のエリアを撮像し、撮像された画像に基づいて、監視エリア内の人物の検索と、その人物の移動経路を追尾するものである。監視システム1は、監視カメラ11-1乃至11-n、人物検索追尾サーバ12、およびネットワーク13より構成されている。
次に、図2のブロック図を参照して、監視カメラ11の第1の構成例について説明する。
次に、図3のブロック図を参照して、人物検索追尾サーバ12の第1の構成例について説明する。
次に、図4のフローチャートを参照して、図2の監視カメラ11による撮像処理について説明する。
次に、図5のフローチャートを参照して、人物検索追尾サーバ12による動体情報フィルタリング処理について説明する。
次に、図11のフローチャートを参照して、人物検索追尾サーバ12による動体情報マッチング処理について説明する。
次に、図13のフローチャートを参照して、人物検索追尾サーバ12による図11のマッチング修正処理につい説明する。
以上においては、動体検出により検出された動体情報を利用して人物の移動履歴を求める例について説明してきたが、動体検出に代えて、人物検出により検出された人物情報を利用することにより、より高い精度で移動履歴を求めるようにしてもよい。以下、第2の実施の形態として、人物検出による例について説明するが、図1で示される監視システムの構成については、同一であるので、その説明は省略するものとする。また、以降において、図1乃至図15を参照して説明した構成と同一の機能を備えた構成については、同一の名称、および同一の符号を付すものとし、その説明は適宜省略するものとする。
次に、図16のブロック図を参照して、監視カメラ11の第2の構成例について説明する。
次に、図17のブロック図を参照して、人物検索追尾サーバ12の第2の構成例について説明する。
次に、図4のフローチャートを参照して、図16の監視カメラ11による撮像処理について説明する。
次に、図20のフローチャートを参照して、人物検索追尾サーバ12による人物情報フィルタリング処理について説明する。
次に、図21のフローチャートを参照して、人物検索追尾サーバ12による人物情報マッチング処理について説明する。尚、人物情報マッチング処理の流れは、図11のフローチャートを参照して説明した動体情報マッチング処理と類似した処理となる。
次に、図22,図23のフローチャートを参照して、人物検索追尾サーバ12による図21のマッチング修正処理につい説明する。
(1) 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部と、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出部と、
前記動体画像尤度算出部により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定部と、
前記動体画像閾値判定部により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出部と、
前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶部より削除する時空間尤度閾値判定部と
を含む情報処理装置。
(2) 前記操作入力部は、前記時空間尤度閾値判定部により、前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定された後、前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを新たに確定する確定情報の入力を受け付け、
前記時空間尤度算出部は、前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定動体情報以外の動体情報の、前記確定情報が新たに入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を新たに算出し、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶部より削除し、
前記操作入力部、前記時空間尤度算出部、および前記時空間尤度閾値判定部は、前記操作入力部により新たに確定情報が入力される度に、同様の処理を繰り返す
(1)に記載の情報処理装置。
(3) 前記動体情報には、含まれている動体画像を撮像した前記複数の撮像部のいずれかを識別するIDをさらに含み、
前記検索対象となる動体の動体画像を含む動体情報である検索対象動体情報、および、前記検索対象動体情報であることを確定する確定情報の入力が受け付けられた確定動体情報のうち、それぞれの動体画像を撮像した複数の撮像部を識別するIDが異なる2の前記動体画像に基づいて、前記撮像部間の色変化を補正するBTF(Brightness Transfer Function)を計算するBTF計算部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定動体情報以外の動体情報のうち、前記BTFが求められているIDの撮像部により撮像された動体画像を含む動体情報の動体画像にBTFを施すBTF処理部と、
前記確定動体情報の動体画像に対する、前記BTF処理部によりBTFが施された動体画像を含む動体情報の、前記動体画像に基づいた尤度からなるBTF動体画像尤度を算出するBTF動体画像尤度算出部と、
前記BTF動体画像尤度算出部により算出されたBTF動体画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定するBTF動体画像閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF動体画像閾値判定部は、前記BTF動体画像尤度算出部により算出されたBTF動体画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF動体画像尤度が所定の閾値よりも低いとき、前記BTF動体画像尤度が所定の閾値よりも低い動体画像を含む動体情報を、前記検索結果動体情報記憶部より削除する
(2)に記載の情報処理装置。
(4)
前記動体画像尤度算出部は、検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像と、前記検索対照動体情報以外の動体情報に含まれる動体画像とのそれぞれに含まれる動体画像に基づいて、それぞれの動体がどの程度類似しているのかを示す類似度を、前記動体画像尤度として算出する
(1)または(2)に記載の情報処理装置。
(5) 前記時空間尤度算出部は、前記確定動体情報以外の動体情報と、前記確定情報が入力された確定動体情報との、前記空間位置座標間の距離を、平均的な人間の移動速度で移動したときの所要時間と、撮像時刻間の時間との関係から前記時空間尤度を算出する
(1),(2),(4)のいずれかに記載の情報処理装置。
(6) 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部を含む情報処理装置の情報処理方法において、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出処理と、
前記動体画像尤度算出処理により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定処理と、
前記動体画像閾値判定処理により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶処理と、
前記結果動体情報記憶処理で検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力処理と、
前記結果動体情報記憶処理で検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出処理と、
前記時空間尤度算出処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶処理により記憶された動体情報より削除する時空間尤度閾値判定処理と
を含む情報処理方法。
(7) 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部を含む情報処理装置を制御するコンピュータに実行させるプログラムであって、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出ステップと、
前記動体画像尤度算出ステップの処理により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定ステップと、
前記動体画像閾値判定ステップの処理により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶ステップと、
前記結果動体情報記憶ステップの処理で検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力ステップと、
前記結果動体情報記憶ステップの処理で検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出ステップと、
前記時空間尤度算出ステップの処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶ステップの処理により記憶された動体情報より削除する時空間尤度閾値判定ステップと
をコンピュータに実行させるためのプログラム。
(8) 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部と、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出部と、
前記人物画像尤度算出部により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定部と、
前記人物画像閾値判定部により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出部と、
前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除する時空間尤度閾値判定部と
を含む情報処理装置。
(9) 前記操作入力部は、前記時空間尤度閾値判定部により、前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定された後、前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを新たに確定する確定情報の入力を受け付け、
前記時空間尤度算出部は、前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定人物情報以外の人物情報の、前記確定情報が新たに入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を新たに算出し、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除し、
前記操作入力部、前記時空間尤度算出部、および前記時空間尤度閾値判定部は、前記操作入力部により新たに確定情報が入力される度に、同様の処理を繰り返す
(8)に記載の情報処理装置。
(10) 前記人物情報には、含まれている人物画像を撮像した前記複数の撮像部のいずれかを識別するIDをさらに含み、
前記検索対象となる人物の人物画像を含む人物情報である検索対象人物情報、および、前記検索対象人物情報であることを確定する確定情報の入力が受け付けられた確定人物情報のうち、それぞれの人物画像を撮像した複数の撮像部を識別するIDが異なる2の前記人物画像に基づいて、前記撮像部間の色変化を補正するBTF(Brightness Transfer Function)を計算するBTF計算部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定人物情報以外の人物情報のうち、前記BTFが求められているIDの撮像部により撮像された人物画像を含む人物情報の人物画像にBTFを施すBTF処理部と、
前記確定人物情報の人物画像に対する、前記BTF処理部によりBTFが施された人物画像を含む人物情報の、前記人物画像に基づいた尤度からなるBTF人物画像尤度を算出するBTF人物画像尤度算出部と、
前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定するBTF人物画像閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF人物画像閾値判定部は、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF人物画像尤度が所定の閾値よりも低いとき、前記BTF人物画像尤度が所定の閾値よりも低い人物画像を含む人物情報を、前記検索結果人物情報記憶部より削除する
(9)に記載の情報処理装置。
(11) 前記検索対象となる人物の人物画像を含む人物情報である検索対象人物情報、および、前記検索対象人物情報であることを確定する確定情報の入力が受け付けられた確定人物情報を、前記検索対象人物と同一人物の人物情報として保持する同一人物情報保持部と、
前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか、または、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、前記所定の閾値よりも低い人物画像を含む人物情報を、前記検索対象人物とは他人の人物情報である他人情報を保持する他人情報保持部と、
前記同一人物情報保持部に保持された人物情報における人物画像と、前記他人情報保持部に保持された人物情報における人物画像とに基づいた学習により前記検索対象人物を検索するための固有特徴を選択する固有特徴検索部と、
前記検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の、前記固有特徴に基づいた尤度である固有特徴尤度を算出する固有特徴尤度算出部と、
前記固有特徴尤度算出部により算出された固有特徴尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い固有特徴尤度の人物情報を、前記検索結果人物情報記憶部より削除する固有特徴尤度閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF人物画像閾値判定部は、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF人物画像尤度が所定の閾値よりも低くないとき、前記固有特徴尤度閾値判定部は、前記固有特徴尤度算出部により算出された固有特徴尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い固有特徴尤度の人物情報を、前記検索結果人物情報記憶部より削除する
(10)に記載の情報処理装置。
(12) 前記固有特徴尤度閾値判定部により前記所定の閾値よりも低い固有特徴尤度の人物情報が、前記検索結果人物情報記憶部より削除されるとき、前記他人情報保持部は、前記所定の閾値よりも低い固有特徴尤度の人物情報を他人の人物情報である他人情報として保持する
(11)に記載の情報処理装置。
(13) 前記固有特徴検索部は、前記同一人物情報保持部に保持された人物情報における人物画像と、前記他人情報保持部に保持された人物情報における人物画像とに基づいた学習により、前記同一人物情報保持部により保持されている人物情報における人物画像と、前記検索対象人物の人物画像との尤度が高くなるような特徴量であって、かつ、前記他人情報保持部により保持されている人物情報における人物画像と、前記検索対象人物の人物画像との尤度が低くなるような特徴量を、固有特徴として選択する
(8)に記載の情報処理装置。
(14) 前記固有特徴尤度算出部は、検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像と、前記検索対照人物情報以外の人物情報に含まれる人物画像とのそれぞれに含まれる人物画像の固有特徴に基づいて、それぞれの人物がどの程度類似しているのかを示す類似度を、前記固有特徴尤度として算出する
(8)に記載の情報処理装置。
(15) 前記人物画像尤度算出部は、検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像と、前記検索対照人物情報以外の人物情報に含まれる人物画像とのそれぞれに含まれる人物画像に基づいて、それぞれの人物がどの程度類似しているのかを示す類似度を、前記人物画像尤度として算出する
(8)に記載の情報処理装置。
(16) 前記時空間尤度算出部は、前記確定人物情報以外の人物情報と、前記確定情報が入力された確定人物情報との、前記空間位置座標間の距離を、平均的な人間の移動速度で移動したときの所要時間と、撮像時刻間の時間との関係から前記時空間尤度を算出する
(8)に記載の情報処理装置。
(17) 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部を含む情報処理装置の情報処理方法において、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出処理と、
前記人物画像尤度算出処理により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定処理と、
前記人物画像閾値判定処理により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶処理と、
前記結果人物情報記憶処理により検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力処理と、
前記結果人物情報記憶処理により検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出処理と、
前記時空間尤度算出処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶処理により記憶された人物情報を削除する時空間尤度閾値判定処理と
を含む情報処理方法。
(18) 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部を含む情報処理装置を制御するコンピュータに実行させるプログラムであって、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出ステップと、
前記人物画像尤度算出ステップの処理により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定ステップと、
前記人物画像閾値判定ステップの処理により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶ステップと、
前記結果人物情報記憶ステップの処理により検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力ステップと、
前記結果人物情報記憶ステップの処理により検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出ステップと、
前記時空間尤度算出ステップの処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除する時空間尤度閾値判定ステップと
を含む処理をコンピュータに実行させるプログラム。
Claims (18)
- 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部と、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出部と、
前記動体画像尤度算出部により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定部と、
前記動体画像閾値判定部により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出部と、
前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶部より削除する時空間尤度閾値判定部と
を含む情報処理装置。 - 前記操作入力部は、前記時空間尤度閾値判定部により、前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定された後、前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを新たに確定する確定情報の入力を受け付け、
前記時空間尤度算出部は、前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定動体情報以外の動体情報の、前記確定情報が新たに入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を新たに算出し、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶部より削除し、
前記操作入力部、前記時空間尤度算出部、および前記時空間尤度閾値判定部は、前記操作入力部により新たに確定情報が入力される度に、同様の処理を繰り返す
請求項1に記載の情報処理装置。 - 前記動体情報には、含まれている動体画像を撮像した前記複数の撮像部のいずれかを識別するIDをさらに含み、
前記検索対象となる動体の動体画像を含む動体情報である検索対象動体情報、および、前記検索対象動体情報であることを確定する確定情報の入力が受け付けられた確定動体情報のうち、それぞれの動体画像を撮像した複数の撮像部を識別するIDが異なる2の前記動体画像に基づいて、前記撮像部間の色変化を補正するBTF(Brightness Transfer Function)を計算するBTF計算部と、
前記結果動体情報記憶部に検索結果動体情報として記憶されている動体情報のうち、前記確定動体情報以外の動体情報のうち、前記BTFが求められているIDの撮像部により撮像された動体画像を含む動体情報の動体画像にBTFを施すBTF処理部と、
前記確定動体情報の動体画像に対する、前記BTF処理部によりBTFが施された動体画像を含む動体情報の、前記動体画像に基づいた尤度からなるBTF動体画像尤度を算出するBTF動体画像尤度算出部と、
前記BTF動体画像尤度算出部により算出されたBTF動体画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定するBTF動体画像閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF動体画像閾値判定部は、前記BTF動体画像尤度算出部により算出されたBTF動体画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF動体画像尤度が所定の閾値よりも低いとき、前記BTF動体画像尤度が所定の閾値よりも低い動体画像を含む動体情報を、前記検索結果動体情報記憶部より削除する
請求項2に記載の情報処理装置。 - 前記動体画像尤度算出部は、検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像と、前記検索対照動体情報以外の動体情報に含まれる動体画像とのそれぞれに含まれる動体画像に基づいて、それぞれの動体がどの程度類似しているのかを示す類似度を、前記動体画像尤度として算出する
請求項1に記載の情報処理装置。 - 前記時空間尤度算出部は、前記確定動体情報以外の動体情報と、前記確定情報が入力された確定動体情報との、前記空間位置座標間の距離を、平均的な人間の移動速度で移動したときの所要時間と、撮像時刻間の時間との関係から前記時空間尤度を算出する
請求項1に記載の情報処理装置。 - 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部を含む情報処理装置の情報処理方法において、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出処理と、
前記動体画像尤度算出処理により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定処理と、
前記動体画像閾値判定処理により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶処理と、
前記結果動体情報記憶処理で検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力処理と、
前記結果動体情報記憶処理で検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出処理と、
前記時空間尤度算出処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶処理により記憶された動体情報より削除する時空間尤度閾値判定処理と
を含む情報処理方法。 - 画像を撮像し、動体を検出し、検出した前記動体の画像からなる動体画像を抽出し、前記動体画像に基づいて、前記動体の空間位置座標を検出し、前記動体画像および前記動体の空間位置座標、および前記画像を撮像した撮像時刻とを含む動体情報を出力する複数の撮像部を含む情報処理装置を制御するコンピュータに実行させるプログラムであって、
検索対象となる動体の動体画像を含む動体情報である検索対象動体情報の動体画像に対する、前記検索対照動体情報以外の動体情報に含まれる動体画像の尤度である動体画像尤度を算出する動体画像尤度算出ステップと、
前記動体画像尤度算出ステップの処理により算出された動体画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い動体画像尤度の動体情報を、前記検索対象動体情報の動体画像の動体と同一の動体の動体画像を含む動体情報である検索結果動体情報として検索する動体画像閾値判定ステップと、
前記動体画像閾値判定ステップの処理により検索結果動体情報として検索された動体情報を記憶する検索結果動体情報記憶ステップと、
前記結果動体情報記憶ステップの処理で検索結果動体情報として記憶されている動体情報のうち、前記動体画像に基づいて、使用者により、前記検索対象動体情報であることを確定する確定情報の入力を受け付ける操作入力ステップと、
前記結果動体情報記憶ステップの処理で検索結果動体情報として記憶されている動体情報のうち、前記確定情報が入力された動体情報である確定動体情報以外の動体情報の、前記確定情報が入力された確定動体情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出ステップと、
前記時空間尤度算出ステップの処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の動体情報を、前記検索結果動体情報記憶ステップの処理により記憶された動体情報より削除する時空間尤度閾値判定ステップと
をコンピュータに実行させるためのプログラム。 - 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部と、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出部と、
前記人物画像尤度算出部により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定部と、
前記人物画像閾値判定部により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出部と、
前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除する時空間尤度閾値判定部と
を含む情報処理装置。 - 前記操作入力部は、前記時空間尤度閾値判定部により、前記時空間尤度算出部により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定された後、前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを新たに確定する確定情報の入力を受け付け、
前記時空間尤度算出部は、前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定人物情報以外の人物情報の、前記確定情報が新たに入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を新たに算出し、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除し、
前記操作入力部、前記時空間尤度算出部、および前記時空間尤度閾値判定部は、前記操作入力部により新たに確定情報が入力される度に、同様の処理を繰り返す
請求項8に記載の情報処理装置。 - 前記人物情報には、含まれている人物画像を撮像した前記複数の撮像部のいずれかを識別するIDをさらに含み、
前記検索対象となる人物の人物画像を含む人物情報である検索対象人物情報、および、前記検索対象人物情報であることを確定する確定情報の入力が受け付けられた確定人物情報のうち、それぞれの人物画像を撮像した複数の撮像部を識別するIDが異なる2の前記人物画像に基づいて、前記撮像部間の色変化を補正するBTF(Brightness Transfer Function)を計算するBTF計算部と、
前記結果人物情報記憶部に検索結果人物情報として記憶されている人物情報のうち、前記確定人物情報以外の人物情報のうち、前記BTFが求められているIDの撮像部により撮像された人物画像を含む人物情報の人物画像にBTFを施すBTF処理部と、
前記確定人物情報の人物画像に対する、前記BTF処理部によりBTFが施された人物画像を含む人物情報の、前記人物画像に基づいた尤度からなるBTF人物画像尤度を算出するBTF人物画像尤度算出部と、
前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定するBTF人物画像閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF人物画像閾値判定部は、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF人物画像尤度が所定の閾値よりも低いとき、前記BTF人物画像尤度が所定の閾値よりも低い人物画像を含む人物情報を、前記検索結果人物情報記憶部より削除する
請求項9に記載の情報処理装置。 - 前記検索対象となる人物の人物画像を含む人物情報である検索対象人物情報、および、前記検索対象人物情報であることを確定する確定情報の入力が受け付けられた確定人物情報を、前記検索対象人物と同一人物の人物情報として保持する同一人物情報保持部と、
前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか、または、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、前記所定の閾値よりも低い人物画像を含む人物情報を、前記検索対象人物とは他人の人物情報である他人情報を保持する他人情報保持部と、
前記同一人物情報保持部に保持された人物情報における人物画像と、前記他人情報保持部に保持された人物情報における人物画像とに基づいた学習により前記検索対象人物を検索するための固有特徴を選択する固有特徴検索部と、
前記検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の、前記固有特徴に基づいた尤度である固有特徴尤度を算出する固有特徴尤度算出部と、
前記固有特徴尤度算出部により算出された固有特徴尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い固有特徴尤度の人物情報を、前記検索結果人物情報記憶部より削除する固有特徴尤度閾値判定部とをさらに含み、
前記時空間尤度閾値判定部は、前記時空間尤度算出部により新たに算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低くない場合、前記BTF人物画像閾値判定部は、前記BTF人物画像尤度算出部により算出されたBTF人物画像尤度のそれぞれについて、所定の閾値よりも低いか否かを判定し、前記BTF人物画像尤度が所定の閾値よりも低くないとき、前記固有特徴尤度閾値判定部は、前記固有特徴尤度算出部により算出された固有特徴尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い固有特徴尤度の人物情報を、前記検索結果人物情報記憶部より削除する
請求項10に記載の情報処理装置。 - 前記固有特徴尤度閾値判定部により前記所定の閾値よりも低い固有特徴尤度の人物情報が、前記検索結果人物情報記憶部より削除されるとき、前記他人情報保持部は、前記所定の閾値よりも低い固有特徴尤度の人物情報を他人の人物情報である他人情報として保持する
請求項11に記載の情報処理装置。 - 前記固有特徴検索部は、前記同一人物情報保持部に保持された人物情報における人物画像と、前記他人情報保持部に保持された人物情報における人物画像とに基づいた学習により、前記同一人物情報保持部により保持されている人物情報における人物画像と、前記検索対象人物の人物画像との尤度が高くなるような特徴量であって、かつ、前記他人情報保持部により保持されている人物情報における人物画像と、前記検索対象人物の人物画像との尤度が低くなるような特徴量を、固有特徴として選択する
請求項8に記載の情報処理装置。 - 前記固有特徴尤度算出部は、検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像と、前記検索対照人物情報以外の人物情報に含まれる人物画像とのそれぞれに含まれる人物画像の固有特徴に基づいて、それぞれの人物がどの程度類似しているのかを示す類似度を、前記固有特徴尤度として算出する
請求項8に記載の情報処理装置。 - 前記人物画像尤度算出部は、検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像と、前記検索対照人物情報以外の人物情報に含まれる人物画像とのそれぞれに含まれる人物画像に基づいて、それぞれの人物がどの程度類似しているのかを示す類似度を、前記人物画像尤度として算出する
請求項8に記載の情報処理装置。 - 前記時空間尤度算出部は、前記確定人物情報以外の人物情報と、前記確定情報が入力された確定人物情報との、前記空間位置座標間の距離を、平均的な人間の移動速度で移動したときの所要時間と、撮像時刻間の時間との関係から前記時空間尤度を算出する
請求項8に記載の情報処理装置。 - 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部を含む情報処理装置の情報処理方法において、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出処理と、
前記人物画像尤度算出処理により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定処理と、
前記人物画像閾値判定処理により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶処理と、
前記結果人物情報記憶処理により検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力処理と、
前記結果人物情報記憶処理により検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出処理と、
前記時空間尤度算出処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶処理により記憶された人物情報を削除する時空間尤度閾値判定処理と
を含む情報処理装置。 - 画像を撮像し、人物を検出し、検出した前記人物の画像からなる人物画像を抽出し、前記人物画像に基づいて、前記人物の空間位置座標を検出し、前記人物画像および前記人物の空間位置座標、および前記画像を撮像した撮像時刻とを含む人物情報を出力する複数の撮像部を含む情報処理装置を制御するコンピュータに実行させるプログラムであって、
検索対象となる人物の人物画像を含む人物情報である検索対象人物情報の人物画像に対する、前記検索対照人物情報以外の人物情報に含まれる人物画像の尤度である人物画像尤度を算出する人物画像尤度算出ステップと、
前記人物画像尤度算出ステップの処理により算出された人物画像尤度のそれぞれについて、所定の閾値よりも高いか否かを判定し、前記所定の閾値よりも高い人物画像尤度の人物情報を、前記検索対象人物情報の人物画像の人物と同一の人物の人物画像を含む人物情報である検索結果人物情報として検索する人物画像閾値判定ステップと、
前記人物画像閾値判定ステップの処理により検索結果人物情報として検索された人物情報を記憶する検索結果人物情報記憶ステップと、
前記結果人物情報記憶ステップの処理により検索結果人物情報として記憶されている人物情報のうち、前記人物画像に基づいて、使用者により、前記検索対象人物情報であることを確定する確定情報の入力を受け付ける操作入力ステップと、
前記結果人物情報記憶ステップの処理により検索結果人物情報として記憶されている人物情報のうち、前記確定情報が入力された人物情報である確定人物情報以外の人物情報の、前記確定情報が入力された確定人物情報に対する、前記空間位置座標および撮像時刻に基づいた尤度からなる時空間尤度を算出する時空間尤度算出ステップと、
前記時空間尤度算出ステップの処理により算出された時空間尤度のそれぞれについて所定の閾値よりも低いか否かを判定し、前記所定の閾値よりも低い時空間尤度の人物情報を、前記検索結果人物情報記憶部より削除する時空間尤度閾値判定ステップと
を含む処理をコンピュータに実行させるプログラム。
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201380005113.4A CN104041017B (zh) | 2012-01-17 | 2013-01-09 | 信息处理设备和方法 |
US14/370,812 US9412180B2 (en) | 2012-01-17 | 2013-01-09 | Information processing apparatus, information processing method, and program |
EP13738625.6A EP2806634B1 (en) | 2012-01-17 | 2013-01-09 | Information processing device and method, and program |
JP2013554271A JP5999105B2 (ja) | 2012-01-17 | 2013-01-09 | 情報処理装置および方法、並びにプログラム |
RU2014128431/08A RU2600540C2 (ru) | 2012-01-17 | 2013-01-09 | Устройство и способ обработки информации, и программа |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2012007538 | 2012-01-17 | ||
JP2012-007538 | 2012-01-17 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2013108686A1 true WO2013108686A1 (ja) | 2013-07-25 |
Family
ID=48799112
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2013/050209 WO2013108686A1 (ja) | 2012-01-17 | 2013-01-09 | 情報処理装置および方法、並びにプログラム |
Country Status (6)
Country | Link |
---|---|
US (1) | US9412180B2 (ja) |
EP (1) | EP2806634B1 (ja) |
JP (1) | JP5999105B2 (ja) |
CN (1) | CN104041017B (ja) |
RU (1) | RU2600540C2 (ja) |
WO (1) | WO2013108686A1 (ja) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20150050172A (ko) * | 2013-10-31 | 2015-05-08 | 한국전자통신연구원 | 관심 객체 추적을 위한 다중 카메라 동적 선택 장치 및 방법 |
JP2015184810A (ja) * | 2014-03-20 | 2015-10-22 | 株式会社東芝 | 画像処理装置、画像処理方法および画像処理プログラム |
JP2016072964A (ja) * | 2014-09-30 | 2016-05-09 | キヤノン株式会社 | 被写体再識別のためのシステム及び方法 |
JP2018081515A (ja) * | 2016-11-17 | 2018-05-24 | 日本電信電話株式会社 | リソース検索装置およびリソース検索方法 |
JP2018129736A (ja) * | 2017-02-10 | 2018-08-16 | キヤノン株式会社 | 情報処理装置、情報処理方法、およびプログラム |
JP2018190132A (ja) * | 2017-05-01 | 2018-11-29 | 富士通株式会社 | 画像認識用コンピュータプログラム、画像認識装置及び画像認識方法 |
JP2019083579A (ja) * | 2019-02-14 | 2019-05-30 | キヤノン株式会社 | 情報処理装置、撮像装置、情報処理方法、およびプログラム |
JP2020014210A (ja) * | 2013-04-16 | 2020-01-23 | 日本電気株式会社 | 情報処理システム、情報処理方法及びプログラム |
US10636153B2 (en) | 2016-11-22 | 2020-04-28 | Fujitsu Limited | Image processing system, image processing apparatus, and image processing method for object tracking |
US10863142B2 (en) | 2016-03-25 | 2020-12-08 | Panasonic Intellectual Property Management Co., Ltd. | Monitoring device and monitoring system |
US10961825B2 (en) * | 2014-11-14 | 2021-03-30 | National Oilwell Vargo Norway As | Drilling rig |
JPWO2021245747A1 (ja) * | 2020-06-01 | 2021-12-09 | ||
US11429985B2 (en) | 2017-03-21 | 2022-08-30 | Kabushiki Kaisha Toshiba | Information processing device calculating statistical information |
Families Citing this family (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111723668A (zh) * | 2013-09-06 | 2020-09-29 | 日本电气株式会社 | 安全系统、安全方法和非暂时性计算机可读介质 |
US11615460B1 (en) | 2013-11-26 | 2023-03-28 | Amazon Technologies, Inc. | User path development |
US20150215583A1 (en) * | 2013-12-04 | 2015-07-30 | Rasilient Systems, Inc. | Cloud Video Surveillance |
WO2015145917A1 (ja) * | 2014-03-28 | 2015-10-01 | 日本電気株式会社 | 画像補正装置、画像補正方法およびプログラム記録媒体 |
WO2016093553A1 (ko) * | 2014-12-12 | 2016-06-16 | 서울대학교 산학협력단 | 이벤트 데이터를 수집하는 시스템, 이벤트 데이터를 수집하는 방법, 이벤트 데이터를 수집하는 서비스 서버 및 카메라 |
US10134004B1 (en) * | 2014-12-23 | 2018-11-20 | Amazon Technologies, Inc. | Processing image data from a camera cluster |
JP6589321B2 (ja) * | 2015-03-24 | 2019-10-16 | 富士通株式会社 | システム、検索方法およびプログラム |
US10679177B1 (en) * | 2015-03-25 | 2020-06-09 | Amazon Technologies, Inc. | Using depth sensing cameras positioned overhead to detect and track a movement of a user within a materials handling facility |
US10586203B1 (en) | 2015-03-25 | 2020-03-10 | Amazon Technologies, Inc. | Segmenting a user pattern into descriptor regions for tracking and re-establishing tracking of a user within a materials handling facility |
US10810539B1 (en) | 2015-03-25 | 2020-10-20 | Amazon Technologies, Inc. | Re-establishing tracking of a user within a materials handling facility |
US11205270B1 (en) * | 2015-03-25 | 2021-12-21 | Amazon Technologies, Inc. | Collecting user pattern descriptors for use in tracking a movement of a user within a materials handling facility |
RU2602386C1 (ru) * | 2015-05-26 | 2016-11-20 | Общество с ограниченной ответственностью "Лаборатория 24" | Способ визуализации объекта |
JP6389803B2 (ja) * | 2015-05-27 | 2018-09-12 | 富士フイルム株式会社 | 画像処理装置、画像処理方法、プログラムおよび記録媒体 |
WO2017087551A1 (en) * | 2015-11-16 | 2017-05-26 | Cuica Llc | Inventory management and monitoring |
EP3391339A2 (en) * | 2015-12-18 | 2018-10-24 | Iris Automation, Inc. | Real-time visual situational awareness system |
JP6700791B2 (ja) * | 2016-01-05 | 2020-05-27 | キヤノン株式会社 | 情報処理装置、情報処理方法及びプログラム |
JP6284086B2 (ja) * | 2016-02-05 | 2018-02-28 | パナソニックIpマネジメント株式会社 | 追跡支援装置、追跡支援システムおよび追跡支援方法 |
DE102016114168A1 (de) * | 2016-08-01 | 2018-02-01 | Connaught Electronics Ltd. | Verfahren zum Erfassen eines Objekts in einem Umgebungsbereich eines Kraftfahrzeugs mit Vorhersage der Bewegung des Objekts, Kamerasystem sowie Kraftfahrzeug |
EP3533229A4 (en) * | 2016-10-28 | 2020-06-24 | Axon Enterprise, Inc. | SYSTEMS AND METHODS FOR ENRICHING CAPTURED DATA |
JP6659524B2 (ja) * | 2016-11-18 | 2020-03-04 | 株式会社東芝 | 移動体追跡装置、表示装置および移動体追跡方法 |
JP6933015B2 (ja) * | 2017-06-20 | 2021-09-08 | 富士通株式会社 | 情報処理方法、情報処理プログラム及び情報処理装置 |
JP7105246B2 (ja) * | 2017-10-23 | 2022-07-22 | パナソニック インテレクチュアル プロパティ コーポレーション オブ アメリカ | 再構成方法および再構成装置 |
US11328513B1 (en) | 2017-11-07 | 2022-05-10 | Amazon Technologies, Inc. | Agent re-verification and resolution using imaging |
CN110099237B (zh) * | 2018-01-31 | 2021-08-17 | 腾讯科技(深圳)有限公司 | 图像处理方法、电子装置及计算机可读存储介质 |
JP7009252B2 (ja) * | 2018-02-20 | 2022-01-25 | キヤノン株式会社 | 画像処理装置、画像処理方法およびプログラム |
US11010597B1 (en) | 2018-05-10 | 2021-05-18 | Ism Connect, Llc | Entry prevention of persons of interest from venues and events using facial recognition |
US11544965B1 (en) | 2018-05-10 | 2023-01-03 | Wicket, Llc | System and method for access control using a plurality of images |
JP7216487B2 (ja) * | 2018-06-21 | 2023-02-01 | キヤノン株式会社 | 画像処理装置およびその制御方法 |
EP3866770A1 (en) * | 2018-10-18 | 2021-08-25 | Johnson & Johnson Consumer Inc. | Novel dosage form |
CN111291585B (zh) * | 2018-12-06 | 2023-12-08 | 杭州海康威视数字技术股份有限公司 | 一种基于gps的目标跟踪系统、方法、装置及球机 |
US11386306B1 (en) | 2018-12-13 | 2022-07-12 | Amazon Technologies, Inc. | Re-identification of agents using image analysis and machine learning |
JP2021145164A (ja) * | 2020-03-10 | 2021-09-24 | 株式会社日立製作所 | 映像解析システム、及び、映像解析方法 |
US11151390B1 (en) * | 2020-05-21 | 2021-10-19 | Ism Connect, Llc | Self-correcting face detection pipeline-based method and apparatus for censusing a crowd |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006245795A (ja) | 2005-03-01 | 2006-09-14 | Toa Corp | 動体追跡装置及び動体追跡表示装置 |
JP2009301104A (ja) | 2008-06-10 | 2009-12-24 | Chube Univ | 物体検出装置 |
JP2010049296A (ja) * | 2008-08-19 | 2010-03-04 | Secom Co Ltd | 移動物体追跡装置 |
JP2010257451A (ja) * | 2009-03-31 | 2010-11-11 | Sogo Keibi Hosho Co Ltd | 人物検索装置、人物検索方法、及び人物検索プログラム |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
RU2193825C2 (ru) * | 2000-08-10 | 2002-11-27 | Открытое акционерное общество "Научно-конструкторское бюро вычислительных систем" | Способ обработки сигналов для определения координат объектов, наблюдаемых в последовательности телевизионных изображений, и устройство для его осуществления (варианты) |
US7999849B2 (en) * | 2006-05-17 | 2011-08-16 | The Boeing Company | Moving object detection |
JP5253102B2 (ja) * | 2008-11-13 | 2013-07-31 | 将文 萩原 | 物体判別方法および物体判別装置 |
JP5483899B2 (ja) * | 2009-02-19 | 2014-05-07 | 株式会社ソニー・コンピュータエンタテインメント | 情報処理装置および情報処理方法 |
GB201113633D0 (en) * | 2011-08-08 | 2011-09-21 | Vision Semantics Ltd | User-primed iterative discovery of object associations in distributed multi-source data |
US9111147B2 (en) * | 2011-11-14 | 2015-08-18 | Massachusetts Institute Of Technology | Assisted video surveillance of persons-of-interest |
JP6364743B2 (ja) * | 2013-10-31 | 2018-08-01 | 株式会社Jvcケンウッド | 情報処理装置、制御方法、プログラム、及び情報システム |
-
2013
- 2013-01-09 JP JP2013554271A patent/JP5999105B2/ja active Active
- 2013-01-09 WO PCT/JP2013/050209 patent/WO2013108686A1/ja active Application Filing
- 2013-01-09 CN CN201380005113.4A patent/CN104041017B/zh active Active
- 2013-01-09 EP EP13738625.6A patent/EP2806634B1/en active Active
- 2013-01-09 RU RU2014128431/08A patent/RU2600540C2/ru active
- 2013-01-09 US US14/370,812 patent/US9412180B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006245795A (ja) | 2005-03-01 | 2006-09-14 | Toa Corp | 動体追跡装置及び動体追跡表示装置 |
JP2009301104A (ja) | 2008-06-10 | 2009-12-24 | Chube Univ | 物体検出装置 |
JP2010049296A (ja) * | 2008-08-19 | 2010-03-04 | Secom Co Ltd | 移動物体追跡装置 |
JP2010257451A (ja) * | 2009-03-31 | 2010-11-11 | Sogo Keibi Hosho Co Ltd | 人物検索装置、人物検索方法、及び人物検索プログラム |
Non-Patent Citations (6)
Title |
---|
"Human Tracking Using Similar Image Search in Sparsely Distributed Camera Environment", TECHNICAL REPORT, vol. 110, no. 330, December 2010 (2010-12-01) |
0. JAVED; K. SHAFIQUE; M. SHAH: "Appearance modeling for tracking in multiple non-overlapping cameras", IEEE INT. CONF. ON COMPUTER VISION AND PATTERN RECOGNITION, 2005, pages 26 - 33, XP010817496, DOI: doi:10.1109/CVPR.2005.71 |
J. FRIEDMAN; T. HASTIE; R. TIBSHIRAN: "Annals of statistics", vol. 38, 2000, article "A decision-theoretic generalization of on-line learning and an application to boosting, Unpublished manuscript available electronically by Yoav Freund and Robert E. Schapir'' and ''Additive logistic r", pages: 337 - 374 |
N. DALAL; B. TRIGG: "Histograms of Oriented Gradients for Human Detection", COMPUTER VISION AND PATTERN RECOGNITION, SAN DIEGO, CA, 20 June 2005 (2005-06-20) |
RAINER LIENHART; JOCHEN MAYDT: "An Extended Set of Haar-like Features for Rapid Object Detection", IEEE ICIP 2002, vol. 1, September 2002 (2002-09-01), pages 900 - 903 |
See also references of EP2806634A4 |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11483521B2 (en) | 2013-04-16 | 2022-10-25 | Nec Corporation | Information processing system, information processing method, and program |
JP2020014210A (ja) * | 2013-04-16 | 2020-01-23 | 日本電気株式会社 | 情報処理システム、情報処理方法及びプログラム |
KR20150050172A (ko) * | 2013-10-31 | 2015-05-08 | 한국전자통신연구원 | 관심 객체 추적을 위한 다중 카메라 동적 선택 장치 및 방법 |
KR102105189B1 (ko) * | 2013-10-31 | 2020-05-29 | 한국전자통신연구원 | 관심 객체 추적을 위한 다중 카메라 동적 선택 장치 및 방법 |
JP2015184810A (ja) * | 2014-03-20 | 2015-10-22 | 株式会社東芝 | 画像処理装置、画像処理方法および画像処理プログラム |
JP2016072964A (ja) * | 2014-09-30 | 2016-05-09 | キヤノン株式会社 | 被写体再識別のためのシステム及び方法 |
US10961825B2 (en) * | 2014-11-14 | 2021-03-30 | National Oilwell Vargo Norway As | Drilling rig |
US11885204B2 (en) | 2014-11-14 | 2024-01-30 | National Oilwell Varco Norway As | Drilling rig |
US10863142B2 (en) | 2016-03-25 | 2020-12-08 | Panasonic Intellectual Property Management Co., Ltd. | Monitoring device and monitoring system |
JP2018081515A (ja) * | 2016-11-17 | 2018-05-24 | 日本電信電話株式会社 | リソース検索装置およびリソース検索方法 |
US10636153B2 (en) | 2016-11-22 | 2020-04-28 | Fujitsu Limited | Image processing system, image processing apparatus, and image processing method for object tracking |
JP2018129736A (ja) * | 2017-02-10 | 2018-08-16 | キヤノン株式会社 | 情報処理装置、情報処理方法、およびプログラム |
US11429985B2 (en) | 2017-03-21 | 2022-08-30 | Kabushiki Kaisha Toshiba | Information processing device calculating statistical information |
JP2018190132A (ja) * | 2017-05-01 | 2018-11-29 | 富士通株式会社 | 画像認識用コンピュータプログラム、画像認識装置及び画像認識方法 |
JP2019083579A (ja) * | 2019-02-14 | 2019-05-30 | キヤノン株式会社 | 情報処理装置、撮像装置、情報処理方法、およびプログラム |
JPWO2021245747A1 (ja) * | 2020-06-01 | 2021-12-09 | ||
WO2021245747A1 (ja) * | 2020-06-01 | 2021-12-09 | 日本電気株式会社 | 追跡装置、追跡方法、および記録媒体 |
JP7332047B2 (ja) | 2020-06-01 | 2023-08-23 | 日本電気株式会社 | 追跡装置、追跡システム、追跡方法、およびプログラム |
Also Published As
Publication number | Publication date |
---|---|
US9412180B2 (en) | 2016-08-09 |
EP2806634A1 (en) | 2014-11-26 |
CN104041017B (zh) | 2017-08-11 |
JPWO2013108686A1 (ja) | 2015-05-11 |
RU2600540C2 (ru) | 2016-10-20 |
RU2014128431A (ru) | 2016-02-10 |
EP2806634A4 (en) | 2015-08-19 |
CN104041017A (zh) | 2014-09-10 |
US20150146921A1 (en) | 2015-05-28 |
EP2806634B1 (en) | 2020-08-12 |
JP5999105B2 (ja) | 2016-09-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5999105B2 (ja) | 情報処理装置および方法、並びにプログラム | |
US11594031B2 (en) | Automatic extraction of secondary video streams | |
KR101173802B1 (ko) | 대상물 추적 장치, 대상물 추적 방법, 및 제어 프로그램이 기록된 기록 매체 | |
JP6428266B2 (ja) | 色補正装置、色補正方法および色補正用プログラム | |
JP6800628B2 (ja) | 追跡装置、追跡方法、及びプログラム | |
US20100165112A1 (en) | Automatic extraction of secondary video streams | |
CN110008795B (zh) | 图像目标追踪方法及其系统与计算机可读取记录介质 | |
JP2011146796A5 (ja) | ||
JP6551226B2 (ja) | 情報処理システム、情報処理方法及びプログラム | |
US20120106784A1 (en) | Apparatus and method for tracking object in image processing system | |
JP6924064B2 (ja) | 画像処理装置およびその制御方法、ならびに撮像装置 | |
WO2013054130A1 (en) | Aerial survey video processing | |
JP6521626B2 (ja) | 被写体追跡装置、方法およびプログラム | |
JP5539565B2 (ja) | 撮像装置及び被写体追跡方法 | |
JP2016081095A (ja) | 被写体追跡装置、その制御方法、撮像装置、表示装置及びプログラム | |
JP6899986B1 (ja) | オブジェクト領域特定装置、オブジェクト領域特定方法、及びオブジェクト領域特定プログラム | |
Majumdar et al. | Registration and mosaicing for images obtained from UAV | |
JP2014053859A (ja) | 移動体観測装置 | |
JP5247419B2 (ja) | 撮像装置および被写体追跡方法 | |
Kim et al. | Fast ROI Detection for Speed up in a CNN based Object Detection | |
Korovin et al. | Point Features Detector of Brightness Anomalies in Monochrome Images of a Real-time Video Sequence | |
JP2018019156A (ja) | チャンネル識別システム、装置、方法及びプログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 13738625 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2013554271 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2013738625 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14370812 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2014128431 Country of ref document: RU Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |