WO2015133159A1 - 画像処理装置、画像処理方法、および、画像処理プログラム - Google Patents
画像処理装置、画像処理方法、および、画像処理プログラム Download PDFInfo
- Publication number
- WO2015133159A1 WO2015133159A1 PCT/JP2015/050222 JP2015050222W WO2015133159A1 WO 2015133159 A1 WO2015133159 A1 WO 2015133159A1 JP 2015050222 W JP2015050222 W JP 2015050222W WO 2015133159 A1 WO2015133159 A1 WO 2015133159A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- head
- image
- person
- input image
- area
- Prior art date
Links
Images
Classifications
-
- 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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- 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
- 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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Definitions
- the present disclosure relates to an image processing apparatus, an image processing method, and an image processing program capable of detecting a person from an input image.
- the technology for detecting a person from an image includes, for example, a monitoring camera for detecting an abnormality such as an intrusion of a person, a monitoring camera for detecting a fall of a person in a nursing facility, a sports video form analysis device, It is applied to various things such as image processing devices.
- image processing apparatuses are made on the assumption that an image from a camera installed on a ceiling or a pillar is analyzed. That is, these image processing apparatuses detect a person so that the processing time does not increase, assuming that the person appears in a predetermined direction.
- Non-Patent Document 1 discloses a technique for detecting a person from an input image obtained by photographing a standing person from a predetermined direction.
- Non-Patent Document 2 discloses a technique for detecting a person from an input image obtained by photographing a sleeping person from a predetermined direction.
- Kiyoshi Hashimoto Tomoyuki Kagaya, Hiroo Kataoka, Yuji Sato, Masaki Tanabe, Kyoko Oshima, Mitsuko Fujita, Yoshimitsu Aoki, “Multi-person tracking with a backbone link model considering posture changes”, Information Processing Society of Japan Research Report, 2011-CVIM-177 (19) Ching-WeiWang, Andrew Hunter, “Robust Pose Recognition of the Obscured Human Body”, International Journal of Computer Vision December 2010, Volume 90, Issue 3, p 313-330, graduate Institute of Technological Science of Science
- Non-Patent Document 1 can detect only a person captured in a predetermined direction, the camera installation position with respect to the position of the person is limited. The same can be said for the technique disclosed in Non-Patent Document 2.
- the conventional person detection image processing technique can detect only a person appearing in a certain direction in the input image.
- This disclosure has been made to solve the above-described problems, and an object in one aspect is to detect a person regardless of the positional relationship between the camera and the person while suppressing an increase in processing time. It is to provide an image processing apparatus capable of performing the above.
- an image processing apparatus for detecting a human region included in an input image from the input image.
- the image processing apparatus includes a head detection unit for detecting the position of a person's head in an input image, and a feature amount extracted from a peripheral image region of the position of the head detected by the head detection unit.
- the specifying unit for specifying the relative positional relationship between the head and other parts in the input image, the direction for searching for the person region according to the relative relationship specified by the specifying unit, and the person
- a person detection unit is provided for setting at least one of areas to be searched for and detecting a person area from an input image.
- the specifying unit searches for a feature value indicating a human torso from the feature values extracted from the peripheral image area of the head, and specifies the direction of the torso relative to the head as a relative positional relationship.
- the input image includes a distance image including a distance to each point on the surface of the subject.
- the feature amount of the peripheral image area of the head extracted by the specifying unit includes a distribution of normal vectors with respect to the surface of the subject extracted from the distance image.
- the specifying unit specifies a relative positional relationship from the direction in which the distribution of the normal vector to the surface of the human torso exists in the distribution of the normal vector in the image region around the head.
- the input image includes a distance image including a distance to each point on the surface of the subject.
- the feature amount of the peripheral image area of the head extracted by the specifying unit includes the relative distance of the subject from the position of the head extracted from the distance image.
- the specifying unit specifies a relative positional relationship from a direction in which an area having a certain number of pixels having a relative distance smaller than a predetermined value exists in the peripheral image area of the head.
- the input image includes a distance image including a distance to each point on the surface of the subject.
- the feature amount of the peripheral image region of the head extracted by the specifying unit includes a distance difference from the background obtained by subtracting the background distance image obtained by photographing the background when no person is included from the distance image.
- the specifying unit specifies a relative positional relationship in a direction in which an area having a certain number of pixels having a distance difference larger than a predetermined value exists in the peripheral image area of the head.
- the feature amount of the peripheral image region of the head extracted by the specific unit includes edge information extracted from the input image.
- the specifying unit specifies a relative positional relationship from the direction in which two parallel edges exist in the edge information in the peripheral image region of the head.
- the head detection unit further detects the size of the person's head in the input image.
- the specifying unit determines the size of the peripheral image region of the head from which the feature amount is extracted according to the size of the head.
- the relative positional relationship includes a human body direction indicating a direction of another part with respect to the head in the input image.
- the human detection unit detects a human region in the input image by rotating either a template used for searching the human region or an input image according to the human body direction and performing matching processing.
- the human detection unit searches for a person region according to at least one of a position of a torso in an input image obtained by searching, a position of a head in the input image, and a size of the head.
- the person area in the input image is detected by limiting at least one of the direction to search and the area to search for the person area.
- an image processing method to be executed by a computer for detecting a human region included in an input image from the input image.
- An image processing method detects a position of a person's head in an input image and uses a feature amount extracted from a peripheral image region of the head position to Specifying the relative positional relationship between the images, and setting at least one of the direction for searching for the person area and the area for searching for the person area according to the relative position relationship, Detecting a region.
- an image processing program for detecting a human region included in an input image from the input image.
- the image processing program detects the position of the person's head in the input image and uses the feature amount extracted from the peripheral image area of the head position to detect the head and other parts in the input image. And specifying at least one of a direction for searching for a person area and an area for searching for a person area according to the relative position relation, and specifying an input image And detecting a person area in the image.
- the present invention it is possible to detect a person regardless of the positional relationship between the camera and the person while suppressing an increase in processing time.
- FIG. 1 is a diagram illustrating a state in which a camera 110 connected to the image processing apparatus 100 is photographing a subject.
- FIG. 2 is a diagram showing an input image obtained by the camera 110 photographing a person from various directions.
- FIG. 3 is a conceptual diagram showing an outline of person detection processing according to the related art.
- FIG. 4 is a conceptual diagram showing an outline of person detection processing of the image processing apparatus 100.
- Image processing apparatus 100 can detect a person regardless of the direction of the human body in the input image.
- FIG. 1 shows a state in which the camera 110 connected to the image processing apparatus 100 is photographing a person 50 existing in a certain space.
- the input image 60 is obtained by the camera 110 photographing the person 50.
- the input image 60 obtained from the camera 110 may be a normal two-dimensional image (hereinafter, also referred to as “luminance image”) composed of luminance values, and each distance from the camera 110 to each point of the subject. 3D information (hereinafter also referred to as “distance image”). Further, the input image 60 obtained from the camera 110 may be a still image or a moving image.
- the person shown in the input image appears in various directions.
- the posture (A) of FIG. 2 when a person sleeping on the bed is targeted, the person 50 may move in the direction of the input image 60 depending on the positional relationship between the bed and the camera 110. It looks sideways.
- the camera 110 is installed on the foot side of the person 50 in order to obtain an image in which the person 50 appears in a certain direction (for example, the vertical direction with respect to the input image 60) in the input image 60.
- the state in which the person 50 is sleeping has a higher degree of freedom of posture than the state in which the person 50 is standing, the person 50 is shown in the posture (B) and posture (C) of FIG. Does not necessarily appear in a certain direction (vertical direction with respect to the input image 60) in the input image 60.
- Image processing apparatus 100 can detect a person region even if the direction of a person is in various directions in the image.
- FIG. 3 is a conceptual diagram showing an outline of person detection processing according to the related art.
- the image processing technology according to the related art searches for a person region while scanning the person search window 80 in the input image 60 and rotating the person search window 80 in small increments.
- the image processing technology according to the related technology specifies the position of the search window as a person area when, for example, a feature amount extracted from the search window matches a predetermined template (feature amount).
- feature amount a predetermined template
- the image processing technique according to the related technique can detect the person region regardless of the human body direction in the image.
- the image processing technique needs to perform image processing while sequentially rotating the search window, which may take a lot of calculation time.
- Image processing apparatus 100 can detect a human region without taking a calculation time even if the human body direction is an unknown image. More specifically, as shown in FIG. 4, the image processing apparatus 100 first detects a head position 72 from the input image 60. Next, the image processing apparatus 100 uses the feature amount extracted from the body search window 70 that is the peripheral region of the detected head position 72 to determine the relative relationship between the head and other parts in the input image 60. Specify the positional relationship.
- the relative positional relationship between the head and the other part includes, for example, the human body direction indicating the direction of the other part with respect to the head in the input image 60.
- the human body direction includes, for example, the body direction relative to the head and the foot direction relative to the head in the input image 60.
- the image processing apparatus 100 sets at least one of a direction for searching for the person area and an area for searching for the person area in accordance with the relative positional relationship between the head and other parts.
- a person area is searched from the image 60.
- the image processing apparatus 100 sets the person search window 80 according to the position of the head of the person 50 or according to the specified human body direction.
- the image processing apparatus 100 searches for a person area from the set person search window 80 and detects a person area from the input image 60.
- the image processing apparatus 100 identifies the human body direction from the feature amount of the peripheral region of the head before detecting the human region.
- the image processing apparatus 100 does not need to search for a person area while rotating the person search window in small increments, and compared with the person detection process according to the related technology, The calculation time can be greatly reduced.
- FIG. 5 is a block diagram illustrating an example of a hardware configuration of the image processing apparatus 100. The hardware configuration of the image processing apparatus 100 will be described with reference to FIG.
- the image processing apparatus 100 is mainly mounted on a computer having a general-purpose architecture.
- the image processing apparatus 100 includes, as main components, a ROM (Read Only Memory) 1, a CPU (Central Processing Unit) 2, a RAM (Random Access Memory) 3, a camera interface (I / F) 4, and a memory card interface.
- (I / F) 5 network interface (I / F) 6, and storage device 20 are included.
- the ROM 1 stores an initial program (boot program) that is executed when the image processing apparatus 100 is started up.
- the CPU 2 controls the entire image processing apparatus 100 by executing various programs such as an operating system (OS) and an image processing program 24 stored in the ROM 1 and the storage device 20.
- the RAM 3 functions as a working memory for executing a program by the CPU 2 and temporarily stores various data necessary for executing the program.
- the camera I / F 4 mediates data communication between the CPU 2 and the camera 110.
- the camera I / F 4 includes an image buffer and temporarily stores a distance image transmitted from the camera 110.
- the camera I / F 4 transfers the accumulated data to the storage device 20 or the ROM 1.
- the camera I / F 4 gives an imaging command to the camera 110 in accordance with an internal command generated by the CPU 2.
- the camera 110 includes, for example, a stereo camera, a distance image sensor that detects position information of a subject in three dimensions, and a camera that can acquire the positions of other subjects in three dimensions.
- the camera 110 may be incorporated in the image processing apparatus 100. In this case, the camera I / F 4 is not an essential configuration.
- the memory card I / F 5 reads / writes data from / to various memory cards (nonvolatile storage media) 5A such as an SD (Secure Digital) card and a CF (Compact Flash (registered trademark)) card.
- memory cards nonvolatile storage media
- SD Secure Digital
- CF Compact Flash
- the memory card I / F 5 is mounted with a memory card 5A storing a distance image acquired by another device, and the distance image read from the memory card 5A is stored in the storage device 20. .
- the network I / F 6 exchanges data with other devices (such as a server device) via various communication media such as the antenna 6A. More specifically, the network I / F 6 is connected via a wired line (LAN (Local Area Network), WAN (Wide Area Network), etc.) such as Ethernet (registered trademark) and / or a wireless line such as a wireless LAN. Data communication.
- LAN Local Area Network
- WAN Wide Area Network
- Ethernet registered trademark
- wireless line such as a wireless LAN.
- the storage device 20 typically includes a large-capacity magnetic storage medium such as a hard disk.
- the storage device 20 stores an image processing program 24 for realizing various types according to the present embodiment. Further, the storage device 20 may store a program such as an operating system.
- the camera 110 may be incorporated in the image processing apparatus 100 instead of being externally attached, and the image processing apparatus 100 main body may have a function of capturing an image of a subject.
- the image processing apparatus 100 may be configured to acquire a distance image using a mechanism similar to a camera and input the acquired distance image to the image processing apparatus 100 by an arbitrary method. In this case, the distance image is input to the image processing apparatus 100 via the memory card I / F 5 or the network I / F 6.
- the image processing program 24 stored in the storage device 20 is stored in a storage medium such as a CD-ROM (Compact Disk-Read Only Memory) and distributed, or distributed from a server device or the like via a network.
- the image processing program 24 may call a required module among program modules provided as a part of the operating system executed by the image processing apparatus 100 at a predetermined timing and order to realize the processing.
- the image processing program 24 itself does not include modules provided by the operating system, and image processing is realized in cooperation with the operating system.
- the image processing program 24 may be provided by being incorporated in a part of an arbitrary program instead of a single program. Even in such a case, the image processing program 24 itself does not include a module that is commonly used in an arbitrary program, and image processing is realized in cooperation with the arbitrary program. Even such an image processing program 24 that does not include some modules does not depart from the spirit of the image processing apparatus 100 according to the present embodiment. Furthermore, some or all of the functions provided by the image processing program 24 may be realized by dedicated hardware.
- the image processing apparatus 100 may not necessarily perform processing in real time.
- the image processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server apparatus realizes processing according to the present embodiment.
- the distance image is transmitted to the server device (cloud side), and the server device performs image processing according to the present embodiment on the received distance image.
- the server device side it is not necessary for the server device side to perform all functions (processing), and the user side terminal and the server device may cooperate to realize the image processing according to the present embodiment.
- FIG. 6 is a block diagram illustrating an example of a functional configuration of the image processing apparatus 100.
- the image processing apparatus 100 includes a head detection unit 210, a specifying unit 220, and a human detection unit 230.
- the image processing apparatus 100 can use the position of the head as a starting point when detecting the person region.
- the head detection unit 210 detects the position of the person's head in the input image 60.
- the position of the head is detected as coordinate information in the input image 60, for example.
- the head detecting unit 210 outputs the detected head position to the specifying unit 220.
- Specific methods of head detection include, for example, methods that focus on facial parts such as eyes and nose, such as template matching using a face database and Haar-Like features.
- the head detection unit 210 can automatically specify the direction of the human body because the head direction can be specified from the position of the facial parts.
- the image processing apparatus 100 needs to photograph the face almost from the front.
- characteristic face parts such as eyes and nose are not captured in the input image 60, and thus the head detection unit 210 may not be able to identify the face orientation.
- the head detection unit 210 needs to search for the head while rotating the face detection template in small increments in order to match the positional relationship between the facial parts. There is such a possibility.
- the head detection unit 210 uses the distance image (three-dimensional information) obtained by photographing the subject as the input image 60 and performs head detection from the distance distribution state of the distance image. That is, the head detection unit 210 detects a spherical shape indicating the outer shape of the face in the distance image. Since the head resembles a spherical shape, the outer shape does not change greatly even if it is rotated. For this reason, the head detection unit 210 can detect the head position without rotating the head detection template by detecting the spherical shape. Therefore, the head detection unit 210 can realize a fast and stable head detection process.
- the process executed by the head detection unit 210 is not limited to the above, and any process that can detect the position of the head from the input image 60 may be used.
- the head detection unit 210 may perform head detection using a model having a head shape instead of a spherical shape.
- the head detection unit 210 may use distance information from the camera to the head.
- the head detection unit 210 can accurately detect the size of the head on the input image 60 by detecting the head using a head shape model or distance information to the head.
- the head detection unit 210 may detect the position of the head from the input image 60 by detecting the outer edge of the face shown in the luminance image. In this case, the head detection unit 210 detects a circular shape indicating the outer shape of the face in the input image 60.
- the specifying unit 220 uses the feature amount extracted from the peripheral image region of the position of the head detected by the head detecting unit 210, and the relative positional relationship between the head and other parts in the input image 60. Is identified. Typically, the specifying unit 220 specifies the human body direction indicating the direction of the trunk relative to the head as the positional relationship. The identification unit 220 outputs the identified human body direction to the human detection unit 230. Details of the method of specifying the human body direction of the specifying unit 220 will be described later.
- the human detection unit 230 detects a human region from the input image 60 using the head detection result by the head detection unit 210 and the human body direction specified by the specification unit 220. For example, in order to shorten the processing time, the person detection unit 230 detects a person region by limiting at least one of the direction to be searched and the region according to the human body direction specified by the specification unit 220.
- the detected person area is shown as coordinate information in the input image 60, for example.
- the human detection unit 230 performs input by rotating and performing matching processing according to the human body direction in which one of the template used for searching the human region and the input image 60 is specified. A person region in the image 60 is detected.
- the template includes edge information, HOG (Histogram of Oriented Gradient), or other feature amount extracted from an image showing a person's standing posture.
- the person detection unit 230 performs the person detection process by rotating the template in the direction along the specified human body direction with reference to the head position.
- the human detection unit 230 can specify the scanning direction in the input image 60 from the human body direction, the existing human detection processing can be applied almost as it is. That is, even if the existing human detection process is applied, the human detection unit 230 does not need to rotate the template or the input image 60 each time, and can prevent an increase in calculation amount. Further, the person detection unit 230 may detect the person region by tracking the convex shape in the distance image and detecting the person.
- the identifying unit 220 determines the human body direction from the input image 60 using the detection result of the head detecting unit 210.
- the human head and torso are connected by a neck, and the positional relationship between the head and the torso changes slightly by tilting the neck, but the positional relationship that the torso exists around the head is unchanged. .
- the specifying unit 220 specifies the human body direction using the continuity between the human body parts of the head and the torso. That is, the specifying unit 220 searches for a feature value indicating the human torso from the feature values extracted in the peripheral image area of the head, and specifies the direction of the torso relative to the head as the human body direction.
- Various methods are conceivable for the specifying process of the human body direction by the specifying unit 220. Below, the specific example of the specific process of the human body direction by the head detection part 210 is demonstrated in order.
- the method for specifying the human body direction when a person is in a sleeping position will be described.
- specification part 220 can pinpoint the direction of a human body.
- FIG. 7 is a conceptual diagram showing an outline of processing for specifying a human body direction from an input image obtained by photographing a person.
- FIG. 8 is a diagram visually showing a search area around the head.
- FIG. 9 is a conceptual diagram showing an outline of processing for specifying a human body direction using distance information.
- the position of the subject is shown in two dimensions, that is, the position (x axis) and the distance (z axis) for the sake of simplicity.
- the position of the subject can be obtained as three-dimensional information indicated by the x-axis, the z-axis, and the y-axis (not shown) orthogonal to the x-axis and the z-axis.
- the specifying unit 220 can detect the torso position or direction, the specifying unit 220 can specify the human body direction in the input image 60 from the relative positional relationship between the head position and the torso position.
- the field of view of the camera includes only half of the human torso surface, for example, as shown in FIG. 7, the distance measurement result of the sleeping person is a half-cylinder shape or along the human body surface. Distributed in a convex shape. For this reason, the specifying unit 220 regards the human torso as a semi-cylindrical shape or a convex shape.
- the specifying unit 220 can specify the human body direction by searching for a semi-cylindrical shape or a convex shape in a region adjacent to the head.
- the term “half-cylinder shape” when used, the concept of “convex shape” may be included. Further, when the term “convex shape” is used, the concept of “half-cylinder shape” may be included.
- the specifying unit 220 determines a region (that is, a torso search window 70) for searching for a torso based on the position of the head in the input image 60 detected by the head detecting unit 210. For example, as illustrated in FIG. 8, the specifying unit 220 searches for the half-cylinder shape of the person in the peripheral region 74 of the head of the person 50. In the example shown in FIG. 8, the body is detected on the lower side of the paper surface of the head. The specifying unit 220 specifies the direction of the trunk relative to the head as the human body direction. Or the specific
- the identifying unit 220 uses distance information for the bed plane.
- the human body is considered to have a semi-cylindrical convex distance distribution on the bed surface. For this reason, the specifying unit 220 can specify the direction of the trunk by searching for an area having a convex distance distribution around the head.
- FIG. 10 is a diagram showing an outline of processing for specifying a human body direction using normal line information with respect to a subject surface calculated from a distance image.
- the position of the subject is shown in two dimensions, that is, the position (x-axis) and the distance (z-axis) in order to simplify the description.
- the position of the subject can be obtained as three-dimensional information of the position (x axis, y axis) and distance (z axis).
- the specifying unit 220 can specify the direction of the trunk by searching for a fan-shaped normal distribution from the image area around the head.
- the specifying unit 220 extracts a normal vector distribution with respect to the surface of the subject from the peripheral image region of the head.
- the specifying unit 220 specifies the human body direction from the direction in which the normal vector distribution with respect to the surface of the human torso exists in the extracted normal vector distribution.
- “distribution of normal vectors with respect to the surface of the human torso” means, for example, the distribution of normal vectors with respect to the semi-cylindrical surface, the distribution of normal vectors with respect to the shape of the human torso obtained by learning in advance, and Includes normal vector distributions showing other torso.
- the identifying unit 220 detects an area where a normal vector distribution with respect to the surface of the human torso exists as an image of the human torso from an image area around the head, and the direction of the torso area with respect to the head position. Is specified as the human body direction.
- the specifying unit 220 may specify the human body direction or the body direction from the distribution of normal vectors in the detected body region without using the head position.
- the specifying unit 220 can reduce the influence of the distance change caused by the futon or the like by using the normal line information calculated from the distance image, as compared with the case of using the distance information as it is. That is, the specifying unit 220 can reduce the influence of noise and can realize stable detection. In addition, since the specifying unit 220 uses the normal line information, a highly accurate determination using a human three-dimensional shape is possible.
- Still another example of the processing of the specifying unit 220 includes a method using distance difference information. For example, paying attention to the fact that the distance between the head and the torso from the bed plane is the same, the specifying unit 220 calculates the distance from the bed plane around the head, and the distance is greater than a predetermined value. The direction including many small areas is determined as the trunk direction.
- the predetermined value may be determined from the thickness of the human body and the futon, for example. Alternatively, as the predetermined value, the values of the thickness of the human body and the futon may be used as they are.
- the specifying unit 220 calculates a relative distance with respect to the bed plane using a distance image including distances to points on the surface of the subject. That is, the specifying unit 220 calculates the distance from each point of the subject around the head to the bed plane. Thereby, the specifying unit 220 extracts a relative distance from each point of the subject around the head to the bed plane as a feature amount. The specifying unit 220 uses the extracted feature amount to specify the human body direction from the direction of a region having a certain number or more of pixels having a relative distance smaller than a predetermined value.
- the specifying unit 220 detects a region having a certain number or more of pixels having a value smaller than a predetermined value as a human torso region, and specifies the direction of the torso region with respect to the head position as the human body direction.
- the specifying unit 220 may specify the human body direction or the torso direction from the detected shape of the torso region without using the head position.
- the specifying unit 220 can efficiently exclude areas that are not related to the human area such as the background area by using the relative distance from the bed plane, and can specify the human body direction with high accuracy. become.
- the specifying unit 220 specifies the human body direction using the relative distance from the bed plane. However, even if the human body direction is specified using the relative distance from the head in the peripheral image region of the head. Good. More specifically, the specifying unit 220 calculates a relative distance with respect to the head using a distance image including distances to points on the surface of the subject. That is, the specifying unit 220 subtracts the distance at the head position from the distance at each point of the subject around the head. Thereby, the specifying unit 220 extracts the relative distance of the subject around the head with respect to the head as a feature amount.
- the specifying unit 220 specifies the human body direction from the direction of an area having a certain number or more of pixels having a relative distance smaller than a predetermined value, using the extracted feature amount. Typically, the specifying unit 220 detects a region having a certain number or more of pixels having a value smaller than a predetermined value as a human torso region, and specifies the direction of the torso region with respect to the head position as the human body direction. The specifying unit 220 may specify the human body direction or the torso direction from the detected shape of the torso region without using the head position.
- Still another example of the processing of the specifying unit 220 includes a method using background difference information obtained by subtracting a background distance image from the input image 60 (distance image). Focusing on the fact that the person is discontinuous with the background, the specifying unit 220 determines the direction in which the distance difference between the image area around the head and the background distance image includes many areas that are larger than a predetermined value. It is determined that the body direction.
- the specifying unit 220 obtains a distance from the background obtained by subtracting a background distance image obtained by photographing a background when no person is included from a distance image acquired at a certain timing. The difference is extracted as a feature amount. Then, the specifying unit 220 specifies the direction of the region having a distance difference larger than a predetermined value as the human body direction in the peripheral image region of the head. Typically, the identifying unit 220 detects a region having a certain number or more of pixels having a distance difference larger than a predetermined value as a human torso region, and determines the direction of the torso region with respect to the head position in the input image 60. Specify as direction. The specifying unit 220 may specify the human body direction or the torso direction from the detected shape of the torso region without using the head position.
- the background distance image for example, an image obtained in advance when a person is not included in the field of view of the camera is used.
- an image obtained by photographing at a timing when the head detecting unit 210 does not detect the head in the input image 60 may be used.
- the specifying unit 220 can efficiently exclude regions that are not related to the human region such as the background region by using the distance image obtained by performing the background subtraction, and specifies the human body direction with high accuracy. It becomes possible.
- FIG. 11 is a conceptual diagram showing an outline of processing for specifying the human body direction using edge information acquired from the input image 60.
- a method using edge information acquired from the input image 60 can be considered.
- the human torso is considered to be cylindrical, the distance from the camera 110 to the subject changes rapidly on both sides of the human torso. That is, it is considered that two parallel long distance edges occur on both sides of the fuselage.
- the specifying unit 220 can obtain the direction of the trunk by searching for two parallel long distance edges in the image region adjacent to the head. That is, the specifying unit 220 specifies the human body direction from the direction in which two parallel edges exist in the edge information extracted in the peripheral image region of the head.
- the specifying unit 220 performs edge extraction from the input image 60 (distance image). Edge extraction is realized by, for example, differentiating a distance image to obtain a gradient of pixel values. Typically, the identifying unit 220 performs edge extraction by convolving a differential filter, a Prewitt filter, a Sobel filter, a Laplacian filter, a LOG filter, and other filters that can extract edges into a distance image. .
- the specifying unit 220 extracts edge information of an image area around the head as a feature amount, and uses the extracted edge information to extract an image area around the head. An existing straight line is detected. If the identifying unit 220 detects two straight lines that are parallel, or are facing in a similar direction, the identifying unit 220 identifies a region between the two straight lines as a human torso region.
- the specifying unit 220 can efficiently exclude a region unrelated to a human region having no distance change or luminance change by using the edge information, and can specify the human body direction with high accuracy. Become. Also, since the edge information can be obtained from a low-resolution image, the specifying unit 220 can specify the human body direction even if the image has a low resolution.
- Still another example of the processing of the specifying unit 220 includes a method using an image composed of luminance values (that is, a luminance image). Unlike the case where the distance image is used, the specifying unit 220 cannot obtain the three-dimensional shape when the luminance image is used as the input image 60, and thus it is difficult to search for the convex shape. However, the specifying unit 220 can specify the direction of the human body by searching for a characteristic characteristic of the body from the luminance image.
- the specifying unit 220 may use a method using a luminance edge extracted from a luminance image. If the body is considered to be cylindrical, the distance from the camera changes abruptly on both sides of the body. If the distance changes on both sides of the fuselage, there is a high possibility that the brightness will change on both sides. That is, even in the case of a luminance image, two parallel edges can be observed on both sides of the body.
- the specifying unit 220 specifies the human body direction from the direction in which two parallel edges exist in the edge information extracted in the peripheral image region of the head.
- Still another example of the processing of the specifying unit 220 includes a method using luminance difference information.
- the specifying unit 220 can specify a person region by detecting a portion different from the background. That is, the specifying unit 220 determines a direction including many regions around the head where the luminance difference from the background is larger than a predetermined value as the trunk direction.
- the specifying unit 220 uses, as a feature amount, a luminance difference from a background obtained by taking a luminance image obtained by photographing a background when no person is included from a luminance image acquired at a certain timing. Extract. Then, the specifying unit 220 specifies the human body direction from the direction of the region having a luminance difference larger than a predetermined value in the peripheral image region of the head. Typically, the specifying unit 220 detects a region having a certain number or more of pixels having a luminance difference larger than a predetermined value as a human torso region, and specifies the direction of the torso region with respect to the head position as the human body direction. The specifying unit 220 may specify the human body direction or the body direction from the detected shape of the body region.
- the identification unit 220 may utilize prior knowledge such as a human body shape model and bed arrangement when searching for a peripheral region of the head.
- the head is usually present on the side of the pillow, and it seems that the body is located somewhere in the long side direction, so that the search range can be limited accordingly.
- the specifying unit 220 combines the plurality of processes described in the specific examples 1 to 8 of the above processes, integrates the plurality of processing results, and specifies the human body direction in the input image 60. Also good. For example, the specifying unit 220 searches for an image region around the head and specifies the human body direction by searching for two parallel edges. Thus, the specifying unit 220 can specify the human body direction more accurately by combining a plurality of processes.
- FIG. 12 is a flowchart showing a part of processing executed by the image processing apparatus 100.
- the processing in FIG. 12 is realized by the CPU 2 executing a program. In other aspects, some or all of the processing may be performed by circuit elements or other hardware.
- step S510 the CPU 2 acquires an input image. Typically, the CPU 2 acquires a distance image or a luminance image as an input image.
- step S512 the CPU 2 detects the position of the head of the person in the acquired input image as the head detection unit 210.
- step S520 the CPU 2 determines, as the head detection unit 210, whether or not a human head has been detected in the input image. When CPU 2 determines that the head has been detected (YES in step S520), control is switched to step S522. If not (NO in step S520), CPU 2 returns control to step S510.
- step S522 the CPU 2 specifies the human body direction in the input image using the feature amount extracted from the peripheral image region of the detected head position as the specifying unit 220.
- step S524 the CPU 2 searches the input image as the human detection unit 230 by limiting at least one of the search direction and the area according to the specified human body direction, and detects the person area.
- step S530 CPU 2 determines whether or not to end the image processing according to the present embodiment. For example, when receiving a user operation to end the image processing, the CPU 2 ends the image processing. When CPU 2 determines to end the image processing according to the present embodiment (YES in step S530), CPU 2 ends the image processing. If not (NO in step S530), CPU 2 sequentially executes the processes from step S510 to step S530 again.
- the image processing apparatus 100 can identify the human body direction in the input image and detect the person area according to the human body direction. A person can be detected without depending on it. Further, the image processing apparatus 100 can specify the direction of the person in the input image by a process with a relatively small calculation amount. As a result, the image processing apparatus 100 does not need to search for a person area while sequentially rotating the person search window, so that the calculation time can be greatly reduced.
- Image processing apparatus 100A according to the present embodiment is different from image processing apparatus 100 according to the first embodiment in that the search range of the body area and the person area in the input image is appropriately changed. Since the hardware configuration is the same as that of image processing apparatus 100 according to the first embodiment, description thereof will not be repeated.
- each part of the human body correlates to some extent with the size of the head. For this reason, the size or ratio of the human body part (particularly the trunk) relative to the head can be determined to some extent from the size of the head. Focusing on this point, the image processing apparatus 100A uses a body search window 70 (see FIG. 4) for searching for a human body direction according to at least one of the identified head size and head position. Change the size accordingly. Thereby, the image processing apparatus 100A can specify the human body direction by excluding the area where no person is shown. That is, the image processing apparatus 100A can greatly reduce the processing time required for specifying the human body direction, and can further reduce false detection of the human body direction.
- the image processing apparatus 100A uses not only the human body direction but also at least one information among the head size, the head position, and the torso position information to detect a person area 80 ( The size is changed as appropriate. Thereby, the image processing apparatus 100A can search for a person area by excluding an area where no person is shown. That is, the image processing apparatus 100A can greatly reduce the processing time required for detecting a person area, and can further reduce false detection of a person area.
- FIG. 13 is a block diagram illustrating an example of a functional configuration of the image processing apparatus 100A.
- the image processing apparatus 100A includes a head detection unit 210, a specification unit 220, and a human detection unit 230.
- the specifying unit 220 includes an area setting unit 222 for setting the body search window 70 in the input image 60.
- the person detection unit 230 includes an area setting unit 232 for setting the person search window 80 in the input image 60.
- the head detection unit 210 detects the head position in the input image 60 by the method described above, and detects the size of the head in the input image 60.
- the head detection unit 210 can obtain the size of the head accompanying the detection of the head position.
- the head detection unit 210 detects the size of the head from a template enlargement rate or a template size in a head detection process such as a template matching process.
- the size of the head in the input image 60 is indicated by, for example, the number of pixels included in the head region, the head diameter, and the like.
- the region setting unit 222 determines the size of the peripheral image region of the head from which the feature amount is extracted (that is, the body search window 70) according to the size of the head obtained from the head detection unit 210. Details of the processing of the area setting unit 222 will be described later.
- the specifying unit 220 searches the body search window 70 set in the input image 60 by the region setting unit 222 and specifies the human body direction.
- the area setting unit 232 detects the person area by limiting the area to search for the person area according to at least one of the position of the trunk in the input image 60 and the position of the head in the input image 60.
- the person search window 80 is set in the input image 60 so as to include the head position obtained from the head detection unit 210 and the torso position. Details of the processing of the area setting unit 232 will be described later.
- the human detection unit 230 detects the human body region by searching the person search window 80 set in the input image 60 by the region setting unit 232.
- FIG. 14 is a conceptual diagram showing an outline of processing for determining the shape of the body search window 70 according to the position of the head.
- FIG. 15 is a conceptual diagram showing an outline of processing for determining the size of the body search window 70 in accordance with the size of the head.
- the image processing apparatus 100A can estimate the size or ratio of the human body part (for example, the torso) with respect to the head to some extent from the head size. Focusing on this, the region setting unit 222 sets the body search window 70 having a shape suitable for the region connected to the head. More specifically, the region setting unit 222 determines at least one of the position and size of the body search window 70 according to the head size. For example, as shown in the posture (B) of FIG. 14, the region setting unit 222 sets the torso search windows 70A to 70H so as to include the head position at one end around the head.
- the region setting unit 222 may limit the search range and window size by using a human body shape model in the search around the head. Furthermore, the region setting unit 222 may limit the size of the convex shape and the radius of the convex shape that are determined as the body portion according to the size of the head. By using prior knowledge such as a model, the search range, body search window size, and the conditions of the convex shape to be recognized are limited, so that the setting of the window according to the size of the human torso and the influence of the non-human convex shape Can be eliminated.
- the area setting unit 222 may be configured to change the size of the body search window 70 in accordance with the size of the head. As described above, the size and ratio of other parts of the person relative to the size of the head can be estimated in advance from the head size. For this reason, the region setting unit 222 can perform a search by changing parameters such as a body search range and a body part search window size from the size of the head. Since the image processing apparatus 100A can optimize the parameters in accordance with the size of the head, for example, it is possible to cope with a change in the size of the body search window 70 due to a change in magnification or a difference in physique. .
- the size of the torso search window 70 is set by the region setting unit 222 so as to increase as the head size in the input image 60 increases.
- the size of the region of the torso search window 70 is set by the region setting unit 222 so as to decrease as the head size in the input image 60 decreases.
- the region setting unit 222 enlarges or reduces the size of the input image 60 itself while keeping the size of the body search window 70 constant. It may be configured.
- FIG. 16 is a concept showing an outline of processing for setting the person search window 80 not only according to the human body direction but also according to at least one of the size of the head in the input image, the position of the head, and the position of the torso.
- FIG. 16 is a concept showing an outline of processing for setting the person search window 80 not only according to the human body direction but also according to at least one of the size of the head in the input image, the position of the head, and the position of the torso.
- the area setting unit 232 sets a person area according to the human body direction specified by the specifying unit 220. Typically, the region setting unit 232 determines the direction of the person search window 80 along the human body direction.
- the region setting unit 232 determines the position of the person search window 80 according to the head position and the torso position in the input image 60. Typically, the region setting unit 232 sets the person search window 80 so that the position of the person search window 80 includes the head position and the torso position in the input image 60.
- the region setting unit 232 may determine the size of the person search window 80 according to the head size in the input image 60.
- the size of the person search window 80 is set by the area setting unit 232 such that the size of the head in the input image 60 increases as the head size increases.
- the size of the area of the person search window 80 is set by the area setting unit 232 such that the smaller the head size in the input image 60 is, the smaller the head size is. .
- the area setting unit 232 sets the person search window 80 not only according to the human body direction but also according to the position of the head, the position of the torso, and the size of the head. It can be excluded from the search target area. As a result, the processing time can be greatly shortened, and further, erroneous detection of the person area can be reduced.
- the area setting unit 232 enlarges or reduces the size of the input image 60 itself while keeping the size of the person search window 80 constant. It may be configured.
- FIG. 17 is a flowchart showing a part of the process executed by image processing apparatus 100A.
- the processing in FIG. 17 is realized by the CPU 2 executing a program. In other aspects, some or all of the processing may be performed by circuit elements or other hardware.
- step S510 the CPU 2 acquires an input image. Typically, the CPU 2 acquires a distance image or a luminance image as an input image.
- step S512 the CPU 2 detects the position of the head of the person in the acquired input image as the head detection unit 210.
- step S520 the CPU 2 determines, as the head detection unit 210, whether or not a human head has been detected in the input image. If CPU 2 determines that the head has been detected (YES in step S520), it switches control to step S610. If not (NO in step S520), CPU 2 returns control to step S510.
- step S610 the CPU 2 uses the region setting unit 222 as the region setting unit 222 to extract a feature amount used for detecting a human torso in accordance with the obtained head size, and an image region around the head. That is, the size of the body search window 70 and the position of the body search window 70 are determined.
- step S ⁇ b> 522 the CPU 2 scans the set body search window 70 as the specifying unit 220 and specifies the human body direction using the feature amount extracted from the body search window 70.
- step S612 the CPU 2 determines the direction of the person search window 80 according to the specified human body direction as the area setting unit 232, and also determines the size of the head, the position of the head, and the body part in the input image 60.
- the position of the person search window 80 is determined according to at least one of the positions.
- step S612 the CPU 2 detects the person area as the area setting unit 232 according to at least one of the specified human body direction, the head size, the head position, and the torso position. The size and position of the image area (that is, the person search window 80) used for the determination are determined.
- step S ⁇ b> 524 the CPU 2 scans the set person search window 80 as the person detection unit 230, and detects a person area using the feature amount extracted from the person search window 80.
- step S530 CPU 2 determines whether or not to end the image processing according to the present embodiment. For example, when receiving a user operation to end the image processing, the CPU 2 ends the image processing. When CPU 2 determines to end the image processing according to the present embodiment (YES in step S530), CPU 2 ends the image processing. If not (NO in step S530), CPU 2 sequentially executes the processes from step S510 to step S530 again.
- image processing apparatus 100A changes the size and position of the body search window according to the position and size of the head in the input image. As a result, it is possible to exclude an area unrelated to the person in the input image, and to greatly reduce the calculation time for detecting the trunk. At the same time, it is possible to reduce the detection error of the body by excluding the inner area related to the person in the input image.
- the image processing apparatus 100A determines the direction of the person search window according to the specified human body direction, and changes the size and position of the person search window according to the position and size of the head in the input image. .
Abstract
Description
[概要]
図1~図4を参照して、第1の実施の形態に従う画像処理装置100の概要について説明する。図1は、画像処理装置100に接続されたカメラ110が被写体を撮影している様子を表している図である。図2は、カメラ110が様々な方向から人物を撮影して得られた入力画像を示した図である。図3は、関連技術に従う人物検出処理の概略を示した概念図である。図4は、画像処理装置100の人物検出処理の概略を示した概念図である。
図3を参照して、本実施の形態に従う画像処理装置100の理解を深めるために、関連技術に従う人体検出処理について説明する。図3は、関連技術に従う人物検出処理の概略を示した概念図である。
本実施の形態に従う画像処理装置100は、人体方向が未知の画像であっても計算時間をかけずに人物領域を検出することができる。より具体的には、画像処理装置100は、図4に示されるように、画像処理装置100は、まず、入力画像60から頭部の位置72を検出する。次に、画像処理装置100は、検出した頭部の位置72の周辺領域である胴体探索ウィンドウ70から抽出した特徴量を用いて、入力画像60内における頭部とその他の部位との相対的な位置関係を特定する。頭部とその他の部位との相対的な位置関係とは、たとえば、入力画像60内における、頭部に対するその他部位の方向を示す人体方向を含む。人体方向は、たとえば、入力画像60内における、頭部に対する胴体の方向、頭部に対する足の方向を含む。
図5は、画像処理装置100のハードウェア構成の一例を示すブロック図である。図5を参照して、画像処理装置100のハードウェア構成について説明する。
図6を参照して、第1の実施の形態に従う画像処理装置100の機能構成について説明する。図6は、画像処理装置100の機能構成の一例を示すブロック図である。画像処理装置100は、頭部検出部210と、特定部220と、人検出部230とを備える。
(特定部220の概要)
以下、図7~図11を参照して、特定部220の具体的な処理について説明する。特定部220は、頭部検出部210の検出結果を用いて入力画像60から人体方向を判定する。人の頭部と胴体とは首で繋がっており、頭部と胴体との位置関係は、首を傾けることで多少変化するものの、頭部の周囲に胴体が存在するという位置関係は不変である。
図7~図9を参照して、特定部220の処理の一例について説明する。図7は、人物を撮影して得られた入力画像から人体方向を特定する処理についての概略を示した概念図である。図8は、頭部周辺において探索する領域を視覚的に示した図である。図9は、距離情報を用いて人体方向を特定する処理についての概略を示した概念図である。なお、図9においては、説明を簡単にするために、被写体の位置が、位置(x軸)および距離(z軸)の二次元で示されているが、実際には、特定部220は、x軸と、z軸と、x軸およびz軸に直交するy軸(図示しない)とで示される三次元の情報として被写体の位置を得ることができる。
図10を参照して、特定部220の処理の具体例について説明する。図10は、距離画像から算出した被写体表面に対する法線情報を用いて人体方向を特定する処理についての概略を示した図である。なお、図10においては、説明を簡単にするために、被写体の位置が、位置(x軸)および距離(z軸)の二次元で示されているが、実際には、特定部220は、位置(x軸、y軸)および距離(z軸)の三次元の情報として被写体の位置を得ることができる。
特定部220の処理のさらに他の例としては、距離差情報を用いる方法が挙げられる。たとえば、頭部と胴体とのベッド平面からの距離が同じである点に着目して、特定部220は、頭部周辺においてベッド平面からの距離を算出し、当該距離が、所定の値よりも小さい領域を多く含む方向を胴体方向と判定する。当該所定の値は、たとえば、人体および布団の厚みから決定されてもよい。あるいは、当該所定の値は、人体および布団の厚みの値をそのまま用いてもよい。
特定部220の処理のさらに他の例として、入力画像60(距離画像)から背景距離画像を差分した背景差分情報を用いる方法が挙げられる。人物は、背景とは非連続であることに着目して、特定部220は、頭部周辺の画像領域と、背景距離画像との距離差が、所定の値よりも大きい領域を多く含む方向を胴体方向と判定する。
図11を参照して、特定部220の処理のさらに他の具体例について説明する。図11は、入力画像60から取得したエッジ情報を用いて人体方向を特定する処理についての概略を示した概念図である。
特定部220の処理のさらに他の例として、輝度値からなる画像(すなわち、輝度画像)を用いた方法が挙げられる。特定部220は、距離画像を用いた場合とは異なり、輝度画像を入力画像60として用いた場合には、立体形状を得ることができないために凸形状を探索することは難しい。しかしながら、特定部220は、身体らしい特徴を輝度画像から探索することで、人体方向を特定することができる。
特定部220の処理のさらに他の例として、輝度差分情報を用いた方法が挙げられる。特定部220は、ベッドなどを写した背景画像が予め得られる場合には、背景とは異なる部分を検出することで人物領域を特定することができる。すなわち、特定部220は、頭部周辺において背景との輝度差が所定の値よりも大きい領域を多く含む方向を胴体方向と判定する。
他にも、特定部220は、頭部の周辺領域を探索する際に、人体の形状モデルやベッドの配置などの事前知識を活用してもよい。ベッドの配置が既知の場合、頭は枕の側に通常存在し、長辺方向の何処かに胴体があると思われるため、探索範囲をその分限定することができる。
他にも、特定部220は、上記の処理の具体例1~8に記載の複数の処理を組み合わせて、これらの複数の処理結果を統合して、入力画像60内において人体方向を特定してもよい。たとえば、特定部220は、頭部周辺の画像領域に、半筒形状を探索するとともに、2本の平行なエッジを探索することで人体方向を特定する。このように、特定部220は、複数の処理を組み合わせることで、より正確に人体方向を特定することが可能になる。
図12を参照して、画像処理装置100の制御構造について説明する。図12は、画像処理装置100が実行する処理の一部を表わすフローチャートである。図12の処理は、CPU2がプログラムを実行することにより実現される。他の局面において、処理の一部又は全部が、回路素子その他のハードウェアによって実行されてもよい。
以上のようにして、本実施の形態に従う画像処理装置100は、入力画像内における人体方向を特定し、人体方向に応じて人物領域を検出することができるので、カメラと人物との位置関係に依らずに人物を検出することができるようになる。また、画像処理装置100は、比較的計算量が少ない処理で入力画像内における人物の方向を特定できる。これにより、画像処理装置100は、人物探索ウィンドウを逐次回転させながら人物領域を探索する必要がなくなるため、計算時間を大幅に短縮することが可能になる。
[概要]
以下、第2の実施の形態に従う画像処理装置100Aの概要について説明する。本実施の形態に従う画像処理装置100Aは、入力画像内における胴体領域および人物領域の探索範囲を適宜変更する点で第1の実施の形態に従う画像処理装置100と異なる。なお、ハードウェア構成については第1の実施の形態に従う画像処理装置100と同じであるので説明を繰り返さない。
図13を参照して、第2の実施の形態に従う画像処理装置100Aの機能構成について説明する。図13は、画像処理装置100Aの機能構成の一例を示すブロック図である。画像処理装置100Aは、頭部検出部210と、特定部220と、人検出部230とを備える。特定部220は、胴体探索ウィンドウ70を入力画像60内に設定するための領域設定部222を含む。人検出部230は、人物探索ウィンドウ80を入力画像60内に設定するための領域設定部232を含む。
図14および図15を参照して、領域設定部222による胴体領域の探索範囲の限定方法の詳細について説明する。図14は、頭部の位置に応じて胴体探索ウィンドウ70の形状を決定する処理の概略を示した概念図である。図15は、頭部のサイズに応じて胴体探索ウィンドウ70のサイズを決定する処理の概略を示した概念図である。
図16を参照して、領域設定部232による人物領域の探索範囲の限定方法について説明する。図16は、人体方向だけでなく、入力画像内の頭部のサイズ、頭部の位置、胴体部の位置のうち少なくとも1つに応じて人物探索ウィンドウ80を設定する処理の概略を示した概念図である。
図17を参照して、画像処理装置100Aの制御構造について説明する。図17は、画像処理装置100Aが実行する処理の一部を表わすフローチャートである。図17の処理は、CPU2がプログラムを実行することにより実現される。他の局面において、処理の一部又は全部が、回路素子その他のハードウェアによって実行されてもよい。
以上のようにして、本実施の形態に従う画像処理装置100Aは、胴体探索ウィンドウのサイズおよび位置を、入力画像内の頭部の位置およびサイズに適応して変更する。これにより、入力画像内における人物とは関係の無い領域を除外することができ、胴体を検出するための計算時間を大幅に短縮することが可能になる。同時に、入力画像内における人物とは関係の内領域を除外することで胴体の誤検出も減らすことができる。
Claims (11)
- 入力画像内において、人物の頭部の位置を検出するための頭部検出部と、
前記頭部検出部により検出された頭部の位置の周辺画像領域から抽出した特徴量を用いて、前記入力画像内における頭部とその他の部位との相対的な位置関係を特定するための特定部と、
前記特定部により特定された相対関係に応じて、人物領域を探索する方向と、人物領域を探索する領域とのうちの少なくとも一方を設定し、前記入力画像から人物領域を検出するための人検出部とを備える、画像処理装置。 - 前記特定部は、前記頭部の周辺画像領域で抽出した特徴量から人物の胴体を示す特徴量を探索し、前記頭部に対する胴体の方向を前記相対的な位置関係として特定する、請求項1に記載の画像処理装置。
- 前記入力画像は、被写体の表面上の各点までの距離からなる距離画像を含み、
前記特定部が抽出する前記頭部の周辺画像領域の前記特徴量は、前記距離画像から抽出した、前記被写体の表面に対する法線ベクトルの分布を含み、
前記特定部は、前記頭部周辺の画像領域の前記法線ベクトルの分布において、人の胴体の表面に対する法線ベクトルの分布が存在する方向から前記相対的な位置関係を特定する、請求項1または2に記載の画像処理装置。 - 前記入力画像は、被写体の表面上の各点までの距離からなる距離画像を含み、
前記特定部が抽出する前記頭部の周辺画像領域の前記特徴量は、前記距離画像から抽出した、前記頭部の位置からの前記被写体の相対距離を含み、
前記特定部は、前記頭部の周辺画像領域において、前記相対距離が所定値よりも小さい画素を一定数以上有する領域が存在する方向から前記相対的な位置関係を特定する、請求項1~3のいずれか1項に記載の画像処理装置。 - 前記入力画像は、被写体の表面上の各点までの距離からなる距離画像を含み、
前記特定部が抽出する前記頭部の周辺画像領域の前記特徴量は、人物が含まれないときの背景を撮影して得られた背景距離画像を前記距離画像から差分した、背景からの距離差を含み、
前記特定部は、前記頭部の周辺画像領域において、前記距離差が所定値よりも大きい画素を一定数以上有する領域が存在する方向から前記相対的な位置関係を特定する、請求項1~4のいずれか1項に記載の画像処理装置。 - 前記特定部が抽出する前記頭部の周辺画像領域の前記特徴量は、前記入力画像から抽出した、エッジ情報を含み、
前記特定部は、前記頭部の周辺画像領域における前記エッジ情報において、2本の平行するエッジが存在する方向から前記相対的な位置関係を特定する、請求項1~5のいずれか1項に記載の画像処理装置。 - 前記頭部検出部は、前記入力画像における人物の頭部のサイズをさらに検出し、
前記特定部は、前記頭部のサイズに応じて、前記特徴量を抽出する前記頭部の周辺画像領域のサイズを決定する、請求項1~6のいずれか1項に記載の画像処理装置。 - 前記相対的な位置関係は、入力画像内における頭部に対する、その他の部位の方向を示す人体方向を含み、
前記人検出部は、人物領域の探索に用いるテンプレート、および、前記入力画像のいずれか一方を前記人体方向に応じて回転するとともにマッチング処理を行い、前記入力画像内における人物領域を検出する、請求項1~7のいずれか1項に記載の画像処理装置。 - 前記人検出部は、探索して得た前記入力画像内の胴体の位置、前記入力画像内の頭部の位置、および、前記頭部のサイズのうちの少なくとも1つに応じて、人物領域を探索する方向と、人物領域を探索する領域とのうちの少なくとも一方を限定して、前記入力画像における人物領域を検出する、請求項7または8に記載の画像処理装置。
- 入力画像内において、人物の頭部の位置を検出することと、
前記頭部の位置の周辺画像領域から抽出した特徴量を用いて、前記入力画像内における頭部とその他の部位との相対的な位置関係を特定することと、
前記相対的な位置関係に応じて、人物領域を探索する方向と、人物領域を探索する領域とのうちの少なくとも一方を設定し、前記入力画像内で人物領域を検出することとを備える、画像処理方法。 - 画像処理プログラムであって、
前記画像処理プログラムは、コンピュータに、
入力画像内において、人物の頭部の位置を検出することと、
前記頭部の位置の周辺画像領域から抽出した特徴量を用いて、前記入力画像内における頭部とその他の部位との相対的な位置関係を特定することと、
前記相対的な位置関係に応じて、人物領域を探索する方向と、人物領域を探索する領域とのうちの少なくとも一方を設定し、前記入力画像内で人物領域を検出することとを実行させる、画像処理プログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US15/123,235 US10445566B2 (en) | 2014-03-05 | 2015-01-07 | Image processing apparatus, image processing method, and non-transitory recording medium storing computer readable program |
JP2016506153A JP6406345B2 (ja) | 2014-03-05 | 2015-01-07 | 画像処理装置、画像処理方法、および、画像処理プログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2014042929 | 2014-03-05 | ||
JP2014-042929 | 2014-03-05 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015133159A1 true WO2015133159A1 (ja) | 2015-09-11 |
Family
ID=54054971
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/050222 WO2015133159A1 (ja) | 2014-03-05 | 2015-01-07 | 画像処理装置、画像処理方法、および、画像処理プログラム |
Country Status (3)
Country | Link |
---|---|
US (1) | US10445566B2 (ja) |
JP (1) | JP6406345B2 (ja) |
WO (1) | WO2015133159A1 (ja) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2019144997A (ja) * | 2018-02-23 | 2019-08-29 | エイアイビューライフ株式会社 | 情報処理装置 |
CN111666786A (zh) * | 2019-03-06 | 2020-09-15 | 杭州海康威视数字技术股份有限公司 | 图像处理方法、装置、电子设备及存储介质 |
WO2022123919A1 (ja) * | 2020-12-11 | 2022-06-16 | ソニーグループ株式会社 | 情報処理装置、情報処理方法、情報処理プログラム |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6627680B2 (ja) * | 2016-07-27 | 2020-01-08 | 株式会社Jvcケンウッド | 人物検出装置、人物検出システム、人物検出方法及び人物検出プログラム |
JP6717235B2 (ja) * | 2017-03-02 | 2020-07-01 | オムロン株式会社 | 見守り支援システム及びその制御方法 |
JP6922410B2 (ja) * | 2017-05-19 | 2021-08-18 | 富士通株式会社 | 姿勢判定プログラム、姿勢判定装置及び姿勢判定方法 |
CN109614848B (zh) * | 2018-10-24 | 2021-07-20 | 百度在线网络技术(北京)有限公司 | 人体识别方法、装置、设备及计算机可读存储介质 |
JP7005546B2 (ja) * | 2019-03-15 | 2022-01-21 | 株式会社東芝 | 画像認識装置、画像認識方法および画像認識プログラム |
US11381730B2 (en) * | 2020-06-25 | 2022-07-05 | Qualcomm Incorporated | Feature-based image autofocus |
CN113642520B (zh) * | 2021-08-31 | 2023-05-30 | 中国人民解放军国防科技大学 | 一种带有头部信息的双任务行人检测方法 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH09139936A (ja) * | 1995-11-16 | 1997-05-27 | Nissan Motor Co Ltd | 車両用カメラ |
JP2006254321A (ja) * | 2005-03-14 | 2006-09-21 | Matsushita Electric Ind Co Ltd | 人物追尾装置及び人物追尾プログラム |
JP2009026146A (ja) * | 2007-07-20 | 2009-02-05 | Canon Inc | 画像処理装置及び画像処理方法 |
JP2013054739A (ja) * | 2011-08-31 | 2013-03-21 | Nielsen Co (Us) Llc | 画像中の人をカウントするための方法及び装置 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7386150B2 (en) * | 2004-11-12 | 2008-06-10 | Safeview, Inc. | Active subject imaging with body identification |
US7613324B2 (en) * | 2005-06-24 | 2009-11-03 | ObjectVideo, Inc | Detection of change in posture in video |
JP4516516B2 (ja) * | 2005-12-07 | 2010-08-04 | 本田技研工業株式会社 | 人物検出装置、人物検出方法及び人物検出プログラム |
US8781162B2 (en) * | 2011-01-05 | 2014-07-15 | Ailive Inc. | Method and system for head tracking and pose estimation |
US9538158B1 (en) * | 2012-10-16 | 2017-01-03 | Ocuvera LLC | Medical environment monitoring system |
KR101800617B1 (ko) * | 2013-01-02 | 2017-12-20 | 삼성전자주식회사 | 디스플레이 장치 및 이의 화상 통화 방법 |
-
2015
- 2015-01-07 JP JP2016506153A patent/JP6406345B2/ja active Active
- 2015-01-07 US US15/123,235 patent/US10445566B2/en active Active
- 2015-01-07 WO PCT/JP2015/050222 patent/WO2015133159A1/ja active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH09139936A (ja) * | 1995-11-16 | 1997-05-27 | Nissan Motor Co Ltd | 車両用カメラ |
JP2006254321A (ja) * | 2005-03-14 | 2006-09-21 | Matsushita Electric Ind Co Ltd | 人物追尾装置及び人物追尾プログラム |
JP2009026146A (ja) * | 2007-07-20 | 2009-02-05 | Canon Inc | 画像処理装置及び画像処理方法 |
JP2013054739A (ja) * | 2011-08-31 | 2013-03-21 | Nielsen Co (Us) Llc | 画像中の人をカウントするための方法及び装置 |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2019144997A (ja) * | 2018-02-23 | 2019-08-29 | エイアイビューライフ株式会社 | 情報処理装置 |
JP7033305B2 (ja) | 2018-02-23 | 2022-03-10 | エイアイビューライフ株式会社 | 情報処理装置 |
CN111666786A (zh) * | 2019-03-06 | 2020-09-15 | 杭州海康威视数字技术股份有限公司 | 图像处理方法、装置、电子设备及存储介质 |
WO2022123919A1 (ja) * | 2020-12-11 | 2022-06-16 | ソニーグループ株式会社 | 情報処理装置、情報処理方法、情報処理プログラム |
Also Published As
Publication number | Publication date |
---|---|
JPWO2015133159A1 (ja) | 2017-04-06 |
JP6406345B2 (ja) | 2018-10-17 |
US10445566B2 (en) | 2019-10-15 |
US20170076148A1 (en) | 2017-03-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6406345B2 (ja) | 画像処理装置、画像処理方法、および、画像処理プログラム | |
JP6125188B2 (ja) | 映像処理方法及び装置 | |
JP5529660B2 (ja) | 瞳孔検出装置及び瞳孔検出方法 | |
JP5771413B2 (ja) | 姿勢推定装置、姿勢推定システム、および姿勢推定方法 | |
JP4642128B2 (ja) | 画像処理方法、画像処理装置及びシステム | |
TWI701609B (zh) | 影像物件追蹤方法及其系統與電腦可讀取儲存媒體 | |
US20160217326A1 (en) | Fall detection device, fall detection method, fall detection camera and computer program | |
JP6007682B2 (ja) | 画像処理装置、画像処理方法及びプログラム | |
JP6587435B2 (ja) | 画像処理装置、情報処理方法及びプログラム | |
JP6217635B2 (ja) | 転倒検知装置および転倒検知方法、転倒検知カメラ、並びにコンピュータ・プログラム | |
JP2008197904A (ja) | 人物検索装置および人物検索方法 | |
JP2004192378A (ja) | 顔画像処理装置およびその方法 | |
KR20150032630A (ko) | 촬상 시스템에 있어서의 제어방법, 제어장치 및 컴퓨터 판독 가능한 기억매체 | |
JP7192582B2 (ja) | 物体追跡装置および物体追跡方法 | |
CN106937532B (zh) | 用于检测真正用户的系统和方法 | |
JP2006343859A (ja) | 画像処理装置及び画像処理方法 | |
JP2018088049A (ja) | 画像処理装置、画像処理方法、及びプログラム | |
US20180191951A1 (en) | Imaging apparatus and imaging condition setting method and program | |
JP5877725B2 (ja) | 画像監視装置 | |
WO2019016879A1 (ja) | 物体検出装置、及び、物体検出手法 | |
KR101353860B1 (ko) | 병렬 영상 필터링 모듈을 구비한 다시점 카메라를 이용한 시각 장애인 가이드 방법, 시스템 및 컴퓨터 판독 가능한 기록 매체 | |
JP6767788B2 (ja) | 情報処理装置、情報処理装置の制御方法およびプログラム | |
Pathak et al. | Fall detection for elderly people in homes using Kinect sensor | |
KR20100060622A (ko) | 인체 자세 판별 시스템 및 그 방법 | |
JP2017194798A (ja) | 情報処理装置、情報処理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: 15757697 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2016506153 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15123235 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15757697 Country of ref document: EP Kind code of ref document: A1 |