JP2018106236A - Video analysis apparatus, video analysis method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique capable of more accurately identifying an object from a video.SOLUTION: Extraction means extracts a first feature quantity related to a specified object from a video imaged by an image means. Determination means compares a second feature quantity related an object detected from a video different from the video that the first feature quantity is extracted with the first feature quantity and determines that the object detected from the different video is the specified object. Acquisition means acquires a first status information indicating a status related to the video that the first feature quantity is extracted and a second status information indicating a status related to the video that the second feature quantity is extracted. Update means determines whether the first feature quantity is updated based on the second feature quantity by comparing the first status information and the second status information acquired by the acquisition means and update them when the determination means determines that the object detected from the different video is the specified object.SELECTED DRAWING: Figure 5

Description

  The present invention relates to a video analysis technique for detecting a specific object from a video.

  In recent years, video surveillance systems such as network cameras are required to monitor a wide range of shopping streets, shopping malls, parks and the like as well as conventional stores such as convenience stores and rental video stores.

  In addition, a technique for detecting a person or the like from an image of a network camera and tracking the detected person or the like is known. A technique for identifying whether a detected person is a specific person using a face authentication technique is also known.

  In particular, in a wide-area monitoring system that targets a wide area where many people pass, there is a demand for video analysis technology for tracking and identifying a person based on characteristics related to the person obtained from the video.

  Patent Document 1 describes a video analysis technique for tracking and identifying a person. In Patent Document 1, a person is detected from an image, a face feature amount and an appearance feature amount are calculated from the detected person, and these are compared with the feature amount of a monitoring target person. Then, based on the comparison result, it is determined whether or not the detected person is a monitoring target person. In addition, when it is determined that the person is a monitoring target by comparing the facial feature amounts, the appearance feature amount is updated to follow the change in the appearance feature amount due to a change in clothes or the like.

JP2013-196423

  However, in the method of Patent Document 1 described above, the situation when each feature amount is acquired is not considered at the time of comparison and update of the feature amount.

  Accordingly, an object of the present invention is to provide a technique capable of identifying an object from a video with higher accuracy.

  In order to solve the above problems, the video analysis apparatus of the present invention has the following configuration. That is, an extraction unit that extracts a first feature amount related to a designated object that is a designated object from a video imaged by the imaging unit, and a video that is different from the video from which the first feature amount is extracted A determination unit that compares the first feature amount with the second feature amount related to the selected object and determines whether the object detected from the different video is the designated object; and the first feature Acquisition means for acquiring first situation information indicating the situation relating to the video from which the amount has been extracted and second situation information indicating the situation relating to the video from which the second feature quantity has been extracted, and detection from the different video When the determination unit determines that the determined object is the designated object, the first situation information and the second situation information acquired by the acquisition unit To determine whether to update the first feature value based on the second feature value, and when it is determined to update, the first feature value is determined based on the second feature value. And updating means for updating.

  According to the present invention, it is possible to provide a technique capable of identifying an object from a video with higher accuracy.

1 is a block diagram illustrating an overall configuration of a video monitoring system in Embodiment 1. FIG. It is a figure which shows an example of the video analysis process which tracks and identifies a person. It is a figure which shows an example of the update method of the person feature-value of the video analysis process in Embodiment 1. FIG. It is a block diagram which shows a hardware configuration. 3 is a functional block diagram showing a configuration of Embodiment 1. FIG. It is a figure which shows an example of the acquisition method of status information. It is a figure which shows an example of each data table. 4 is a flowchart illustrating details of video analysis processing in the first embodiment. FIG. 10 is a diagram illustrating an example of a feature amount adding method according to the second embodiment. 10 is a flowchart illustrating details of video analysis processing in the second embodiment.

  Embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, the structure shown in the following embodiment is an example, and this invention is not limited to the structure demonstrated in the following embodiment. In each embodiment, a person is described as an example of an object to be identified, but the present invention can also be applied to other objects such as a car.

<Embodiment 1>
The first embodiment will be described below. First, the configuration of the video monitoring system 101 in this embodiment will be described with reference to FIG.

  The video monitoring system 101 captures images of, for example, supermarkets, convenience stores, rental video stores, shopping streets, shopping malls, and parks.

  The video monitoring system 101 includes a client PC (personal computer) 102, a video analysis server 103, a recording server 104, and a network camera (imaging device) 105.

  The client PC 102 receives an operation input from the user. Then, an instruction (command) corresponding to the operation input is output to each of the video analysis server 103, the recording server 104, and the network camera 105, and an operation corresponding to the instruction is executed. In addition, information and images acquired from each device are displayed on the display.

  The video analysis server 103 receives an instruction from the client PC 102 and performs video analysis processing based on the instruction. The video analysis process here is an analysis process for the video acquired from the recording server 104 or the network camera 105 to determine a target such as a person shown in the video, or to track the movement of the target. , Or to analyze the behavior of the target. Video analysis processing for tracking and identifying a person will be described later. Here, the identification is to determine whether a person detected in a certain video and a person detected from another video different from the video are the same person. The other video referred to here may be a video captured by the same network camera 105 or a video captured by another network camera 105. When the image is taken by another network camera 105, the situation of the subject to be imaged and the imaging situation such as brightness are different. Even when images are captured by the same network camera 105, the imaging situation changes if the imaging direction, ambient brightness, or the like changes. When the imaging state changes, the feature amount extracted from the captured image also changes.

  The recording server 104 receives an instruction from the client PC 102 or the video analysis server 103 and performs video recording processing based on the instruction. Here, the video recording process refers to a video output instruction to a plurality of network cameras 105 connected to the same network as the recording server 104, and acquires and acquires a video output according to the instruction. Refers to the process of saving video. The video recording process also includes a process of storing meta information obtained from the network camera 105. The meta information includes, for example, time information, position information, camera setting information, sensor information corresponding to the network camera 105, and the like.

  The network camera 105 is provided, for example, on the ceiling, wall, or utility pole of a building. Each network camera 105 captures an image and acquires surrounding video. The network camera 105 receives an instruction from the client PC 102, the video analysis server 103, or the recording server 104, and performs surrounding imaging, camera rotation control (pan, tilt), audio recording, and the like. Various sensors such as a human sensor, a temperature sensor, a voice sensor, and an infrared sensor may be connected to the network camera 105 in order to acquire the above-described meta information. Each of the devices 102 to 105 is connected to a network and can communicate with each other. Alternatively, the functions may be integrated into a single PC or server, and the functions of a plurality of devices may be realized by a single device.

  An example of video analysis processing for tracking and identifying a person will be described with reference to FIGS. FIG. 5 is a block diagram including functional blocks of the video analysis server 103 and functional blocks of the recording server 104. The network camera 105a performs imaging at “position A”. The network camera 105b performs imaging at “position B” different from “position A”.

  The video acquisition unit 502 of the recording server 104 transmits an instruction (command) to the network camera 105 to transmit the video. Then, as a response to the instruction, an image captured by each network camera 105 is acquired. The acquired video may be stored in the recording server 104 or may be output to a functional block (person detection unit 503 or the like) that performs subsequent processing without being stored.

  Next, the person detection unit 503 of the video analysis server 103 analyzes the acquired video and detects a person.

  Here, in FIG. 2, a person 201 and a person 202 represent persons at each position. The person detection unit 503 detects the person 201 from the video at the position A captured by the network camera 105a. In addition, the person detection unit 503 detects the person 202 from the video at the position B captured by the network camera 105b. As a method of detecting a person, various known methods such as a method using pattern matching can be used.

  The feature extraction unit 504 of the video analysis server 103 extracts the feature amount of the person from the image of the person area detected by the person detection unit 503. Here, the feature amount refers to extracting a prominent feature from the image. The feature amount of the person may be a feature amount of the face or the whole body. Further, the feature amount of a person may be information such as a stride, a gait, a height, a system, a stride, a clothes, and the like detected by analyzing an image. The feature amount may be information indicating a color histogram of an image of a person, an edge position, or the like. Anything can be used as long as it can be used for the person identification process.

  When extracting a facial feature, the feature extraction unit 504 extracts the position, size, shape, and the like of each part (eyes, nose, etc.) of the face from the image as a feature amount, and converts it into data. Similarity can be calculated by matching data features, and the same face can be identified using the similarity. As described above, various known techniques can be used for the feature quantity extraction method.

  In FIG. 2, the feature extraction unit 504 extracts person feature amounts 205 and 206 from the images 203 and 204 in the region corresponding to the detected person.

  The feature matching unit 505 determines, for example, whether the person 201 and the person 202 are the same person by collating the feature quantity 205 obtained at the position A and the feature quantity 206 obtained at the position B. Specifically, the feature quantities are compared with each other, and when they are equal to or higher than a predetermined similarity, it is determined that they are the same person. It is possible to identify a person detected in the video surveillance system 101 by extracting feature amounts from time to time for all videos acquired from each network camera 105 and collating the extracted feature amounts. Furthermore, it is possible to estimate the movement path of the person using the identified result.

  Here, FIG. 3 is a diagram illustrating an example of the update of the feature amount of the person executed by the feature update unit 507 according to the first embodiment.

  The situation information 207 is information indicating a situation when the image 203 from which the feature amount 205 is extracted is captured. The status information 208 is information indicating a status when the image 204 from which the feature amount 206 is extracted is captured. The situation information is acquired by the situation information acquisition unit 506 of the video analysis server 103.

  The situation information is information regarding the video acquisition status such as the distance between the camera and the subject, brightness, and the degree of congestion, and may be acquired by analyzing the video or may be acquired from a sensor or the like.

  Here, the identification target information 303 includes feature amount information indicating the feature amount of the identification target, and situation information when a video having acquired the feature amount is captured.

  The feature matching unit 505 compares (matches) the information indicating the feature quantity included in the identification target information 303 with the feature quantity 205 acquired at the position A, so that the person detected at the position A can be identified. Determine if you are a person.

  When the feature matching unit 505 determines that the person is the same person, the feature update unit 507 compares the situation information included in the identification target information 303 with the situation information 205 acquired at the position A, and according to the comparison result. The feature amount and status information of the identification target information are updated.

  When the feature update unit 507 determines that the feature amount 205 acquired at the position A is acquired under a condition with higher reliability than the feature amount of the identification target information 303, the feature update unit 507 sets the identification target information 303 as the feature amount 205. Update with status information 207. In addition, the reliability said here is the reliability showing whether it can identify with a sufficient precision.

  That is, as the identification target information used for the person identification process at the next position B, the identification target information 209 obtained by updating the identification target information 303 with the feature amount 205 and the situation information 207 is used.

  Here, the situation information and its reliability will be described in more detail with reference to FIG. FIG. 6 shows the status information to be acquired and the acquisition method. FIG. 6 also shows an example of the feature quantity that is affected when the situation information fluctuates, and the conditions under which the reliability of the feature quantity increases.

  First, the situation information regarding brightness will be described. When the brightness information is acquired by analyzing the video, the situation information acquisition unit 506 acquires the change in the illumination light by obtaining the background difference of the image. Moreover, when acquiring brightness information from a sensor, the status information acquisition unit 506 can acquire information indicating a change in illuminance value from the illuminance sensor. The situation information regarding brightness may be a luminance value. For example, the average value of the luminance of the entire screen may be used.

  Further, when the situation information 207 of the video from which the feature amount 205 is acquired indicates that the darkness is equal to or less than the situation information of the current identification target information 303, the feature update unit 507 determines that the feature amount is low because the reliability is low. Do not update. On the other hand, when the status information 207 indicates that the status information of the current identification target information 303 is brighter, the feature update unit 507 updates the feature amount. For example, since the shadow is not clear in a dark situation, the feature amount acquired in that situation is not updated. On the other hand, when the feature amount is acquired in a situation where the brightness is bright and the difference in brightness within the screen is small, the feature amount is updated.

  Next, status information regarding the distance between the network camera 105 and a person will be described. When the distance information between the network camera 105 and the person is acquired by analyzing the video, the situation information acquisition unit 506 can estimate the distance from the size of the subject and the change in the size. When obtaining the distance from the sensor, the situation information acquisition unit 506 can acquire the distance from the target using the distance measuring sensor.

  When the situation information 207 of the video from which the feature amount 205 is acquired indicates that the distance is greater than the distance indicated by the current identification target information 303, the feature update unit 507 does not update the feature amount. On the other hand, when the status information 207 of the video from which the feature amount 205 is acquired indicates that the distance is indicated by the status information of the current identification target information 303, the feature update unit 507 updates the feature amount. This is because, for example, when the distance between the network camera 105 and the subject is long, information indicating the details of the facial organs may not be included in the face feature amount, and reliability is reduced. Therefore, the distance between the network camera 105 and the subject is as close as possible, and the feature quantity acquired at a distance where the facial organ is clearly visible has higher reliability.

  Next, situation information regarding the degree of congestion will be described. When analyzing the video and acquiring the congestion level, the situation information acquisition unit 506 can detect the person included in the image and acquire the congestion level from the area and the number of people corresponding to the imaging range. The status information regarding the degree of congestion is, for example, the number of people per predetermined area. When obtaining the degree of congestion from the sensor, the situation information acquisition unit 506 can measure the degree of congestion by looking at the amount of change in the infrared sensor.

  When the situation information 207 of the video from which the feature amount 205 has been acquired indicates that the degree of congestion is greater than the degree of congestion indicated by the situation information of the current identification target information 303, the feature update unit 507 does not update the feature amount. On the other hand, when the status information 207 of the video from which the feature amount 205 has been acquired indicates that the congestion level indicated by the status information of the current identification target information 303 is less crowded, the feature update unit 507 updates the feature amount. To do.

  This is because, for example, when the degree of congestion is high, it is difficult to walk with a natural body, and thus the reliability of feature amounts such as stride and gait decreases. Therefore, it can be said that the degree of congestion that allows walking smoothly without being influenced by the surroundings is higher reliability.

  What is necessary is just to determine whether a feature-value is updated using at least one of the above situation information. However, if a plurality of situation information is used, it can be determined whether the feature amount is updated more accurately. For example, the update may be performed when it is determined to update based on two or more pieces of situation information among the three pieces of situation information. Moreover, you may make it update, when it determines with all three status information being updated.

  Also, the situation information acquisition method and the conditions for increasing the reliability have been described, but this may differ depending on the situation where each network camera 105 is installed. For example, if all the locations where the network camera 105 is installed are in a dim situation, the higher the degree of brightness, the higher the reliability. Moreover, although the example in which the information indicating the condition is stored in advance has been described, the condition may be changed by receiving an input specifying the condition. The above is the explanation of the situation information and its reliability.

  Next, a data table 701 indicating video analysis results held by the video monitoring system 101 will be described with reference to FIG. The data table 701 includes a target ID, an image, a feature amount, a situation information ID, and a position as information.

  “Target ID” is an ID for uniquely identifying a person detected from the acquired image. “Image” is information indicating an image when a person is detected. This may be a still image or a moving image. The “feature amount” is a feature amount obtained by performing feature extraction processing on the detected person.

  The “situation information ID” is an ID for uniquely identifying the situation information when the target image is acquired. “Position” is information indicating the position where the target is detected.

  Next, the data table 702 will be described. The data table 702 is a data table for holding status information. The data table 702 includes status information ID, brightness, distance from the camera, and congestion level as information. “Brightness” is status information indicating the brightness when an image is acquired. The situation information indicating the brightness may be held as an illuminance value, or may be expressed in large, medium, or small. “Distance to camera” indicates status information indicating the distance between the camera and the target (person). “Congestion degree” is status information indicating the degree of congestion when an image is acquired. The data table 701 and the data table 702 may be held by any of the devices 102 to 104. This completes the description of each table.

  Next, the flow of processing in the video monitoring system 101 will be described using the flowchart of FIG. Here, a process will be described in which a person found at position A is designated as an identification target, collated with a person found at another position B, and the feature amount is updated. It is identified whether or not the designated object designated as the identification target is included in each video. This will be described with reference to FIGS. 3 and 7 as necessary. Note that an example in which the identification processing is performed with the feature amount 205 and the situation information 207 acquired at the position A in and after S802 will be described.

  In step S <b> 801, the identification target designating unit 509 receives a user input and sets an identification target. Details of S801 will be described with reference to FIG. FIG. 8B illustrates a flow when the network camera 501a at the position A detects the person 201 and the detected person is set as an identification target by the user's input.

  In step S811, the video acquisition unit 502 acquires the video captured at the position A.

  In step S812, the situation information acquisition unit 506 acquires the situation information 207 corresponding to the video imaged at the position A. Whether the status information 207 is status information at the time of extracting the feature amount 205 can be determined to be updated more accurately.

  In step S813, the person detection unit 503 detects a person from the video acquired in step S811.

  In step S814, the feature extraction unit 504 performs feature extraction processing on the person 201 detected in step S813, and extracts a feature amount 205. The feature amount to be extracted may be any as long as it can identify a person, but in this example, description will be made assuming that a facial feature amount is used. The processing of S811 to S814 is repeatedly performed as necessary.

  In step S815, the identification target designating unit 509 accepts user input and sets an identification target. The user's input reception is performed, for example, by displaying a list of images of the persons detected in S813 and receiving a designation from the user. The above is the details of S801.

  Returning to FIG. 8A, in S802, the video acquisition unit 502 acquires the video of the position B.

  In step S <b> 803, the situation information acquisition unit 506 acquires the situation information 208 corresponding to the video imaged at the position B.

  In step S804, the person detection unit 503 detects the person 202 from the video acquired in step S802.

  In step S <b> 805, the feature extraction unit 504 performs feature extraction processing on the person 202 detected in step S <b> 804 and extracts the feature amount 206 of the person 202. Although the camera at the position B has been described here, the processing of S802 to S805 is repeatedly performed for each camera at each position.

  In step S806, the feature matching unit 505 performs matching (comparison) between the feature amount of the person identification target information 209 specified in step S801 and the feature amount 208 obtained in step S805.

  In step S807, the feature matching unit 505 determines the matching result in step S806 and determines whether the person 201 to be identified and the person 202 detected at the position B are the same. Whether or not they are the same is determined based on the similarity between the feature amounts. For example, if the similarity between the feature amounts is higher than a threshold, it is determined that they are the same. If it is determined that they are the same, the process proceeds to S808, and if it is determined that they are not the same, the process proceeds to S810.

  In step S808, the feature matching unit 505 compares the situation information of the identification target information 209 with the situation information 208 when the feature amount related to the person 202 determined to be the same in step S807 is acquired. In this example, the situation information 207 (001ENV in FIG. 7) acquired at the position A is compared with the situation information 208 (002ENV in FIG. 7) acquired at the position B. As a result of the comparison, if it is determined that the condition information 207 acquired at the position A is a better condition, the process proceeds to S810; otherwise, the process proceeds to S809. The comparison of the situation information is performed in accordance with the conditions for increasing the reliability of the feature amount shown in the table of FIG. According to the table of the data table 702, the status information 207 (001ENV) is “low” brightness, “near” the distance from the camera, and “low” congestion. The status information 208 (002ENV) includes brightness “large”, distance to the camera “near”, and congestion degree “small”. As described above, in the case of the facial feature amount, the reliability is higher when it is acquired in a brighter situation, and thus it can be said that the situation information 208 (002ENV) is a better situation than the situation information 207 (001ENV). . Therefore, in this example, the process proceeds to S809.

  As another example, when the brightness is “small” at the position A, the distance is “near” with the camera, the brightness is “large” at the position B, and the distance is “far” with the camera, The imaging situation may be weighted for comparison. If the brightness is a more important condition than the distance from the camera, it can be said that the imaging situation at position B is better in this example.

  In step S809, the feature update unit 507 receives the results of the processing in steps S807 and S808, and updates the feature quantity to be identified. The feature amount update processing is performed by replacing the feature amount 205 acquired at the position A set as the identification target with the feature amount 206 acquired at the position B.

  In step S810, the display unit 508 causes the display to display the result of the feature amount collation and update (S806 to S809). In this example, as a result of collation, it is determined that the person is the same person, and since it is determined that the situation acquired at the position A is better than the position B, the fact that the feature amount has been updated is output as a result. . As described above, the feature quantity can be updated only in a better imaging situation while taking into account the difference in imaging situation.

  In the present embodiment, only the facial feature amount has been described. Also good.

  In S809, the feature amount update replaces the feature amount. However, an average of the feature amounts of the position A and the position B may be calculated and replaced with the average feature amount.

  As described above, according to the video monitoring system 101 of the first embodiment, the feature amount can be updated in consideration of the imaging state when the feature amount is acquired. Thereby, since it can prevent updating to the feature-value with low reliability, the precision of person identification and tracking can be improved.

<Embodiment 2>
Next, Embodiment 2 will be described. In the first embodiment, it has been described that the situation information at the time of obtaining the feature quantity is obtained and the feature quantity is updated in consideration of the situation information. In the second embodiment, not only the case of updating the feature quantity to be identified, but also an example of additionally storing the feature quantity will be described. Note that description of portions similar to those of the first embodiment is omitted as appropriate.

  Hereinafter, Embodiment 2 will be described with reference to the flowcharts of FIGS. 9 and 10. In the following description, a case will be described in which the identification target information 210 and the identification target information (identification target information 209 in FIG. 9) of the target ID “001” in FIG. 7 are held as identification target information. A case will be described in which matching is newly performed with the feature amount (feature amount 206 in FIG. 9) of the target ID “002”.

  Since S1001 to S1005 are the same as S801 to S805, description thereof will be omitted.

  In step S <b> 1006, the feature matching unit 505 matches the feature quantities 209 and 210 of the plurality of identification target information held with the feature quantity 206.

  In step S1007, the feature matching unit 505 determines whether the person set as the identification target and the person 202 detected at the position B are the same based on the plurality of matching results in step S1006. Whether or not they are identical is determined by comparing the similarity between the feature quantity of the identification target information 209 and the feature quantity 206 and the similarity between the feature quantity of the identification target information 210 and the feature quantity 206 with a threshold value. If the degree of similarity is higher than the threshold, it is determined that they are the same. If it is determined that they are the same, the process proceeds to S1008, and if it is determined that they are not the same, the process proceeds to S1010.

  In step S <b> 1008, the feature matching unit 505 compares the situation information 208 when the feature amount 206 is acquired with the situation information of the identification target information currently held for the person determined to be the same in step S <b> 1007.

  The comparison of the situation information is performed according to the condition that the reliability of the feature amount shown in the table of FIG. 6 is high. A comparison between the status information of the identification target information 209 and the status information 208 and a comparison between the status information of the identification target information 210 and the status information 208 are executed.

  In step S1009, the feature update unit 507 performs processing for adding or updating feature amounts in response to the processing results in steps S1007 and S1008.

  First, the feature update unit 507 confirms the number of identification target information (retention number) currently held as a comparison target, determines whether the retention number is less than a threshold value, It is added as identification target information for the next verification. The holding number may be set in advance or may be specified by the user.

  Note that only feature quantities acquired under conditions better than the identification target information 209 and 210 that are already held may be added as identification target information. Further, in the case of situation information different from the identification target information 209 and 210 that are already held, they may be additionally held.

  On the other hand, if the holding number is equal to or greater than the threshold value, it is replaced (updated) with the worst condition (situation information) among the identification targets held. There are various methods not limited to which identification target information is replaced when the threshold value is held above the threshold value. For example, there is a method of replacing the oldest acquired time or the closest status information, and any method can be used as long as it improves the collation accuracy. The replacement method based on time is different depending on the characteristics of each feature amount, such as when the face feature amount is better in the imaging situation even if the time is older, and the whole body feature amount is better even if the imaging situation is somewhat worse .

  In step S <b> 1010, the display unit 508 causes the display to display the result of feature amount collation and feature amount update or addition (S <b> 1006 to S <b> 1009). In this example, as a result of collation, it is determined that the person is the same person, and that the feature amount is additionally held is output as a result. As described above, it is possible to add or update the feature quantity to be identified while taking into account the difference in the imaging situation.

  As described above, according to the video monitoring system of the second embodiment, it is possible to add and hold a feature quantity considering the situation information by obtaining and holding the situation information when the feature quantity is obtained. become. Accordingly, by holding a plurality of feature amounts for person identification and tracking, it is possible to increase the probability of more accurate identification. In addition, since not all feature quantities are added, the number of feature quantities held as collation targets does not increase excessively, and the calculation amount of feature quantity collation can be suppressed.

(Other examples)
Next, a hardware configuration for realizing each function of each embodiment will be described with reference to FIG. Each of the devices 102 to 104 can also be realized by the hardware configuration shown in FIG. 5, but may be realized by an embedded system, a tablet terminal, or the like.

  In FIG. 4, a CPU (Central Processing Unit) 401 is a central processing unit and controls the operation of the entire information processing apparatus in cooperation with other components based on a computer program. A ROM (Read only memory) 402 is a read-only memory, and stores a computer program, data used for processing defined by the program, and the like. A RAM (Random Access Memory) 403 is a writable memory and functions as a work area of the CPU 401.

  The external storage device 404 can access a recording medium and load a computer program and data stored in a medium (recording medium) such as a USB memory into the video monitoring system 101. The storage 405 is a device that functions as a large-capacity memory such as an HDD (Hard Disk Drive) or an SSD (Solid state Drive). The storage 405 stores various computer programs and data.

  The operation unit 406 is a device that receives an instruction and a command input from a user, and can be realized by a keyboard, a pointing device, a touch panel, or the like.

  The display 407 is a display device that displays a command input from the operation unit 406 and a response output of the information processing apparatus with respect to the command. An interface (I / F) 408 is a device that relays data exchange with an external device. A system bus 409 is a data bus for exchanging data in the information processing apparatus.

  The present invention can also be realized by processing in which one or more processors read and execute a program that realizes one or more functions of the above-described embodiment. The program may be supplied to a system or apparatus having a processor via a network or a storage medium. The present invention can also be realized by a circuit (for example, ASIC) that realizes one or more functions of the above-described embodiments.

  Each functional block shown in FIG. 5 may be realized by the hardware shown in FIG. 4 or may be realized by software.

The present invention is not limited to the embodiments described above, and various modifications can be made without departing from the scope of the present invention. For example, what combined each embodiment is also contained in the content of an indication of this specification.

502 video acquisition unit 503 person detection unit 504 feature extraction unit 505 feature collation unit 506 status information acquisition unit 507 feature update unit

Claims (10)

Extraction means for extracting a first feature amount relating to a designated object, which is a designated object, from a video imaged by the imaging means;
The second feature amount relating to the object detected from the video different from the video from which the first feature amount is extracted is compared with the first feature amount, and the object detected from the different video is Determining means for determining whether the object is the designated object;
Acquisition means for acquiring first situation information indicating a situation relating to the video from which the first feature amount has been extracted and second situation information indicating a situation relating to the video from which the second feature quantity has been extracted;
When the determination unit determines that the object detected from the different video is the specified object, the first situation information acquired by the acquisition unit is compared with the second situation information. Updating means for determining whether to update the first feature quantity based on the second feature quantity and updating the first feature quantity based on the second feature quantity when it is determined to update A video analysis apparatus characterized by comprising: The situation information includes the situation related to the distance between the imaged object and the imaging means that imaged the object, the situation related to the degree of congestion around the imaged object, and the brightness of the video obtained by imaging the object. The video analysis apparatus according to claim 1, wherein the video analysis apparatus is at least one of such situations. The video analysis apparatus according to claim 1, wherein the different video is a video captured by an imaging unit different from the imaging unit. The updating unit determines whether the second situation information is more suitable for processing in the determination unit than the first situation information. When the updating unit determines that the second situation information is more suitable, The video analysis device according to claim 1, wherein the video analysis device is updated based on the second status information. The updating means determines that the brightness indicated by the second situation information is more suitable for the processing in the determination means when the brightness indicated by the first situation information is brighter. The video analysis apparatus according to claim 4. The updating means is configured such that the distance between the imaged object indicated by the second situation information and the image pickup means obtained by imaging the object indicates the imaged object and the object indicated by the first situation information. The video analysis apparatus according to claim 4 or 5, wherein the video analysis apparatus determines that the determination unit is suitable for processing when the distance to the image pickup unit is shorter than the image pickup unit. The update unit is more suitable for the process in the determination unit when the congestion level indicated by the second situation information indicates that the congestion level is less than the congestion level indicated by the first status information. The video analysis device according to claim 4, wherein the video analysis device is determined as follows. The determination unit compares a plurality of comparison target feature amounts including the first feature amount with the second feature amount. The update unit has a predetermined number of the comparison target feature amounts. The video analysis device according to any one of claims 1 to 7, wherein when the number is smaller than the number, the second feature amount is added to the plurality of comparison target feature amounts used by the determination unit. . An extraction step of extracting a first feature amount relating to a designated object, which is a designated object, from a video imaged by the imaging means;
Comparing the first feature quantity with a second feature quantity associated with an object detected from a video different from the video from which the first feature quantity has been extracted,
An acquisition step of acquiring first situation information indicating a situation relating to the video from which the first feature amount has been extracted and second situation information indicating a situation relating to the video from which the second feature amount has been extracted;
When the determination step determines that the object detected from the different video is the designated object, the first situation information acquired by the acquisition step is compared with the second situation information. The updating step of determining whether to update the first feature quantity based on the second feature quantity and updating the first feature quantity based on the second feature quantity when it is determined to update. And a video analysis method characterized by comprising:   The program for functioning a computer as each means of the video-analysis apparatus of any one of Claims 1-8.

Images