JP2020087180A - Movable body tracking method and image processing device used therefor - Google Patents

Abstract

To provide a movable body tracking method capable of identifying a movable body accurately, in a relationship between only adjacent cameras without an integrated server, and tracking the same.SOLUTION: In an overlapping visual field region of visual field regions of adjacent two cameras, a common point which is handled as the same position in the visual field regions of respective cameras, is provided, when only one common point is provided, the one point is taken as a reference point, and when a plurality of common points are provided, one point out of the common points is taken as the reference point. Each of the cameras collates a movable body detected in a visual field thereof with a movable body model base and represents a movable body position by a representative point for tracking the movable body. With respect to the movable body detected by each of the adjacent two cameras in the overlapping visual field region, movable bodies having identity in an actual distance to a movable body position in a real space of the movable body represented by the reference point and the representative point and a direction are determined to be the same movable body to the detected movable body. With respect to the movable bodies which are determined to be the same movable body, movable body tracking information on one camera side is transmitted to the adjacent other camera side, for coordination of two adjacent cameras.SELECTED DRAWING: Figure 3

Description

The present invention relates to a moving body tracking method and an image processing apparatus for linking a plurality of cameras to track the behavior of the same moving body over a wide area. Here, the moving body includes various moving bodies such as a person, another animal, and a vehicle such as an automobile.

A person detection system for detecting a person as a moving body using a camera image is used, for example, for purposes such as counting the number of customers in a store, and is used for analyzing a visitor pattern in a store. As a person detection system, for example, when a person is detected by a camera, a method is known in which a person model is set in advance and a person is detected by detecting an image that matches the person model from a camera image. (For example, refer to Patent Document 1). Further, the technology for converting the position on the coordinate system of the image into the position on the three-dimensional coordinate system of the world coordinate system and the coordinate of the image by assuming the height of the three-dimensional coordinate system of the world coordinate system from the ground There is also known a technique of converting to a system position (for example, refer to Patent Document 2).

Japanese Patent No. 3406587 JP, 2006-338123, A JP, 2016-162306, A U.S. Pat. No. 7,319,479

When a person is tracked over the entire monitoring area using a plurality of cameras, the person position is identified based on the distance from the position directly below two adjacent cameras whose distances between the two cameras are known in advance ( If the same person is tried to be determined), the distance between the camera position and the person position becomes long, which may cause a large error due to factors such as the installation conditions of the camera, making it difficult to identify the person. Further, in order to detect the action record of a person by receiving the person position information from each camera and mapping it on a large one-dimensional map that represents the monitoring area, for example, as shown in Patent Document 4, a plurality of cameras are used. A centralized server was needed to coordinate.

Further, in the above method, it is necessary to measure the installation angle, orientation, and installation distance of two adjacent cameras in advance or to make them constant, and it is particularly difficult to measure the tilt of the camera in practice. Further, there is a problem that the work at the time of installing the camera is complicated.
If the camera installation conditions are completely measured, the error of the distance of the person estimated from the camera image is small, but in reality it is not possible to completely remove the error factors such as the installation angle of the camera and the inclination of the ground. It is difficult to estimate the person distance in the camera image based on the distance from the position directly below the camera. As the distance from the position directly below the camera to the person position increases, the error accumulates and a larger error occurs.

The above problems are not limited to the case of targeting a person, and are similar to the case of targeting various moving bodies such as an automobile.
The present invention has been made in view of the above circumstances, and it is possible to accurately identify and track a moving object by eliminating the need for an integrated server element and only between two adjacent cameras. Another object of the present invention is to provide a moving body tracking method and an image processing apparatus used for the same.

The moving body tracking method according to the present invention,
A method of tracking a moving object in a surveillance area in which the field of view of a plurality of cameras is set,
Has an overlapping field of view in which the field of view of two adjacent cameras partially overlaps,
Place a common point to be treated as the same position between the view areas of two adjacent cameras in the overlapping view area,
In the case of having only one common point, this one point is used as a reference point, and in the case of having a plurality of common points, one of these points is used as a reference point,
The moving body detected by each camera in each field of view is collated with the moving body model base, and the moving body is represented by a representative point and tracked,
With respect to the moving body detected by each of two adjacent cameras in the overlapping visual field area, the actual distance and the direction of the moving body position in the real space of the moving body represented by the reference point and the representative point are the same. It is determined that the moving bodies having the same are the same moving body,
With respect to the moving bodies which are determined to be the same moving body, the moving body tracking information on one camera side is transmitted to the other adjacent camera side so that the two adjacent cameras cooperate with each other.

Further, the image processing device according to the present invention,
An image processing device commonly used in the moving body tracking method,
Has a camera and a processor,
The processor is
In the overlapping visual field area where the visual field area of the own camera partially overlaps with the adjacent visual field areas of the other cameras, common points to be treated as the same position between the respective visual field areas are recorded, and further, only one common point is provided. In the case of using this one point as a reference point, and in the case of having a plurality of the common points, one of these points is recorded as the reference point,
An image processing process unit that represents the person as a representative point by collating the moving body detected by the own camera with the moving body model base,
A tracking processing unit that tracks the moving object captured by the image processing unit,
A communication process unit that communicates the moving body position information of the moving body being tracked with another image processing device on the other camera side,
For a moving body detected by each of the other cameras adjacent to the own camera in the overlapping field of view, the actual distance and direction between the reference point and the person's position in the real space of the moving body represented by the representative point are A coincidence determination process unit that determines that moving bodies having the same identity are the same moving body.

According to the present invention, since an integrated server element is not required and operation is performed only by registering a common point, setting is easy, and a moving object is accurately identified and tracked only by the relationship between two adjacent cameras. It becomes possible. In addition, the number of cameras for tracking a moving object can be linked virtually without any limitation. Therefore, it is possible to track a large-scale moving object over a wide area.

It is the whole block diagram which shows the person tracking system of execution form. It is a schematic diagram which shows the state which has arrange|positioned two common points and one middle point in the overlapping visual field area of two cameras. It is a schematic diagram which shows each visual field area|region showing the distance and direction between a reference point and a person position as a person identification method of the person recognized by the overlapping visual field area|region of two cameras. It is a schematic diagram which shows each visual field area in an overlapping visual field area in order to demonstrate the method of calculating the midpoint of ground height 0 from the common point which has ground height. It is a block diagram showing the composition of the image processing device in an embodiment. It is a schematic diagram which shows the state which has an angle difference in the coordinate system of two camera images.

Embodiments of the present invention will be described below with reference to the accompanying drawings.
In this embodiment, a person is used as the moving body. However, the present invention is not limited to humans, and can be applied to various moving bodies such as animals other than humans and vehicles such as automobiles.
As shown in FIGS. 1 and 2, a person tracking system 1 as a moving body tracking system has a plurality of image processing devices 2 (2a, 2b, 2c, etc.) including cameras 11 (11a, 11b, 11c, etc.). The image processing apparatuses 2 are communicatively connected by a network 3 such as a LAN. A totaling device 4 is connected to the network 3. The person tracking information of the person 5 tracked in cooperation with each image processing apparatus 2 is finally output to the totaling apparatus 4. Each image processing device 2 includes a camera 11 and a processor 12. These image processing devices 2 have the same structure, operate with the same program, and cooperate to perform person tracking.

It is assumed that the camera images are synchronized between the adjacent image processing apparatuses 2 and the simultaneity of the camera images is secured. At this time, the network 3 is preferably a network capable of high-speed communication such as Ethernet (registered trademark).

The camera 11 is an image pickup means such as a CMOS camera, and always picks up an image of the monitoring area 6 within the field of view 13 (13a, 13b, 13c, etc.), and continuously uses this camera image (field of view 13) as a video signal. To the processor 12 (12a, 12b, 12c, etc.). Each camera 11 is installed on a ceiling or a wall so as to capture an image of the floor surface or the like of the monitoring area 6. In addition, each camera 11 has an overlapping visual field area 30 (30ab, 30bc, etc.) in which the visual field areas 13 (13a, 13b, 13c, etc.) of two cameras 11 adjacent to the monitoring area 6 partially overlap. Is installed. In FIG. 1, the plurality of cameras 11 are installed so that the visual field regions 13 are adjacent to each other in a row in the lateral direction, but another visual field region 13 may be adjacent to any position in four directions around the visual field region 13. A plurality of cameras 11 may be installed in each.

The processor 12 captures and tracks the person 5 in the visual field region 13 captured by the camera 11, transmits the person tracking information of the same person to another adjacent image processing apparatus 2, and the two image processing apparatuses 2 adjacent to each other. Coordinate the tracking of the person 5. When the person 5 moves from one visual field area 13 to the other visual field area 13, the processor 12 performs person identification as to whether or not they are the same person. According to this person identification method, as shown in FIG. 3, a reference point 18 is applied to a person 5 imaged on each camera 11 side in an overlapping visual field area 30 in which the visual field areas 13 of two adjacent cameras 11 partially overlap. Persons 5a and 5b having the same actual distances dp1 and dp2 and the directions α1 and α2 from the person position 51 to the person position 51 are determined as the same person. Details of the person identifying method will be described later.

As illustrated in FIG. 5, the processor 12 includes a shared point recording unit 21, a person tracking history recording unit 22, an image processing process unit 23, a tracking processing process unit 24, a coordinate conversion process unit 25, a communication process unit 26, and a match determination process. The unit 27 and the like are provided. The processor 12 implements a person tracking method.

The common point recording unit 21 arranges two marks by markers in the overlapping visual field area 30 in which the visual field areas 13 of two adjacent cameras 11 partially overlap, and records these two points as the common points 15 and 16. .. The two shared points 15 and 16 are treated as the same position in the visual field regions 13 of the two adjacent cameras 11. Further, the shared point recording unit 21 records the midpoint 17 arranged at the intermediate position in the real space between the two shared points 15 and 16 as the shared point. The midpoint 17 is also treated as the same position in each visual field region 13 between the two adjacent cameras 11. That is, the shared points 15 and 16 are points determined using the marker as a mark, and the midpoint 17 is a point determined based on the point (shared point 15 and 16) of the marker, but any of the points 15 to Reference numeral 17 is also a common point that is treated as the same position in the visual field regions 13 of two adjacent cameras 11. When selecting a point closest to the person position 51 of the person 5 represented by the representative point 14 from the shared points 15 and 16 and the midpoint 17 registered as the shared point, and performing person identification in the overlapping visual field region 30. Is set as a reference point 18. At the time of person identification, the actual distances dp1 and dp2 from the reference point 18 to the person position 51 and the directions α1 and α2 are obtained. According to this, for example, the distance is shorter than that in the case of performing the person identification by obtaining the actual distance from the position directly below the camera position to the person position. By defining the common points 15 to 17 as the same point in the visual field areas 13 of the two adjacent cameras 11, the reference point 18 is calibrated and the error can be reduced. As a result, person identification can be performed accurately. Since the midpoint 17 registered as the common point is arranged near the center of the overlapping visual field region 30, the midpoint 17 may be used as the reference point 18. In this case, the distances dp1 and dp2 with respect to the person position 51 in the overlapping visual field region 30 are easily obtained as relatively short distances, and the determination process of the reference point 18 can be easily performed. Further, the two respective shared points 15 and 16 can be arranged at arbitrary positions in the overlapping visual field area 30, but in the rectangular overlapping visual field area 30, it is preferable to arrange them on the respective short sides, and The common points 15 and 16 are preferably arranged near the center of the short side of the overlapping visual field region 30. Thereby, the three points of the two shared points 15 and 16 and the one middle point 17 can be dispersed in the overlapping visual field region 30 as much as possible.

In FIG. 3, the markers for determining the common points 15 and 16 are described as being placed on the ground surface (ground height 0), but the ground height of the markers is actually arbitrary. be able to.

Further, although markers are arranged as targets for determining the shared points 15 and 16, if there is a feature that can be determined to be the same point or object in the overlapping visual field region 30, the shared points 15 and 16 are based on this feature. Can be placed and there is no need to place a marker as a marker.

In the common point recording unit 21, information on the ground height and position information on the camera image are recorded. Accordingly, it is possible to calculate the position on the camera image when the ground height of the shared points 15 and 16 is changed by using a known coordinate conversion technique (for example, paragraph 0002). With the coordinate conversion technique, for example, as shown in FIG. 4, the positions on the camera image when the shared points 15 and 16 have a ground height of 0 are registered as points 52 and 53. Then, the midpoint 17 of the ground clearance 0 can be calculated from the common points 52 and 53 of the ground clearance of 0. However, since the calculation becomes complicated and the marker that serves as a mark is often placed on the ground surface, it is preferable that the common points 15 and 16 are fixed at the ground height of 0.

When the overlapping visual field area 30 is wide, by disposing three or more common points, the common points can be dispersed in the overlapping visual field area 30 and the accuracy can be stabilized. In addition, by finding the midpoint to be placed at the intermediate position of each shared point in the real space and recording it as the shared point, it is possible to further increase the number of shared points and shorten the distance from the shared point to the person position. The accuracy of person identification can be stabilized.

Further, in FIG. 3, it is described that the midpoint 17 is placed at an intermediate position in the real space between the two shared points 15 and 16 and this is added as a shared point. It is not limited to point 17. For example, as a shared point to be further added based on the shared points 15 and 16, one point or two or more points are selected from an arbitrary number of points located on a line in the real space connecting the two shared points 15 and 16. It may be determined, or one or two or more points whose positions in the real space calculated from the shared points 15 and 16 are treated as the same may be selected and determined. Particularly, by arranging points whose positions in the real space are treated as the same so as to correspond to all the points on the image at the time of initialization, it is possible to reduce the calculation amount of the coincidence determination processing unit 27.

Although the shared points determined by the marks are described as two points 15 and 16 in FIG. 3, the coordinate systems 19a and 19b on the images of the visual field regions 13 of the two adjacent cameras 11 have the same coordinates. When it is known in advance that they have the same coordinate system, only one common point is arranged and recorded, and by handling this one common point as the reference point 18, the time and effort for recording the mark (marker) is reduced. You can

The person tracking history recording unit 22 records the person tracking information of the person 5 tracked by the image processing apparatus 2 of its own, and also records the person tracking information of the person 5 transmitted from another adjacent image processing apparatus 2. To do.

The image processing unit 23 performs a process of recognizing and capturing the person 5 in the camera image (viewing area 13) captured by the camera 11. As a person recognition method in the image processing unit 23, the person 5 in the camera image is vectorized by using the vector focus method (Japanese Patent No. 3406587), and a standardized person model base (for example, a person having a height of 1600 mm in the real space) is used. The person 5 is recognized and detected by collating with a model whose outer shape is modeled. The recognized person 5 is represented as a representative point 14 on the camera image. In addition, the representative point 14 of the person 5 on the image obtained from the person model base is subjected to coordinate conversion technology (for example, the coordinate conversion technology described in paragraph 0002) to obtain the ground height 0 of the person 5 in the real space. The point is calculated as the person position 51 in the real space. Thereby, the person position 51 can be shown in the visual field region 13 with a small error.

As the person recognition method, for example, when the method of indicating the barycentric position of the person image as the person position is used, the barycentric position of the person image changes depending on the shape shown in the camera image, and thus the person position changes, and the person position is detected. A large error will occur. In addition, it is not known what the height of the center of gravity of the person image itself is. As a result, there is a high risk of mistakenly identifying a person as another person. On the other hand, as in the vector focus method (Japanese Patent No. 3406587), the human position is collated with a standardized human model base, and the foot position of the human model base (position at ground height 0) is normalized in the form of the human position 51. Turn into. As a result, the position of the person captured in the camera image can be represented with a small error, and the person can be identified with high accuracy.

The tracking process unit 24 performs a process of tracking the person 5 captured by the image processing unit 23 in time series.

The coordinate conversion processing unit 25 performs a process of performing coordinate conversion on the position of the person 5 being tracked by the two-dimensional coordinate system of the camera image. At this time, as shown in FIG. 6, there is an angle difference θ between the two-dimensional coordinate system 19a of the camera image (13a) of its own camera 11 and the two-dimensional coordinate system 19b of the camera image (13b) of the adjacent camera 11. In this case, the angle difference θ can be obtained from the line connecting the two shared points 15 and 16, and the person position 51 which is coordinate-converted using the rotation matrix calculated from the angle difference θ is calculated by the following formula 1. .. As a result, even if there is an angle difference θ in the two-dimensional coordinate system of the camera images between the two cameras 11, the person positions 51 in the respective visual field regions 13a and 13b can be treated as the same two-dimensional coordinate system.

In the example of FIG. 6, the angle difference θ is obtained from the total of the angles θa and θb obtained from the difference between the coordinate axes (for example, the y-axis) of the visual field regions 13a and 13b and the straight line connecting the common points 15 and 16. That is, the angle difference θ can be obtained by the expression θa−θb in the image processing apparatus 2a on the visual field area 13a side and by the expression θb−θa in the image processing apparatus 2b on the visual field area 13b side. An angle difference θ is calculated between the processing devices 2a and 2b.
Alternatively, in the coordinate conversion processing unit 25 of each of the image processing devices 2a and 2b, the person position 51 in each of the visual field regions 13a and 13b is converted by performing conversion at the time of transmission and reception using the rotation matrix calculated from the angles θa and θb. It can be treated as the same two-dimensional coordinate system.

The communication process unit 26 transmits the position coordinates (person position information) of the person 5 output by the coordinate conversion process unit 25 to the adjacent image processing device 2, and the person 5 transmitted from the adjacent image processing device 2. Receives position coordinates (person position information). Further, the communication process unit 26 communicates the person tracking information about the person 5 determined to be the same person with the adjacent image processing device 2. Furthermore, if there is no image processing apparatus 2 on the camera 11 side to which the person tracking information should be finally transmitted, the communication process unit 26 transmits the person tracking information to the totaling apparatus 4.

The coincidence determination processing unit 27 performs a person identification process on whether or not the persons imaged by the two cameras 11 adjacent to the person 5 in the overlapping visual field region 30 are the same person. That is, with respect to the person 5 in the overlapping visual field region 30 of two adjacent cameras 11, the actual distance and direction in the real space between the reference point 18 and the person position 51 on one camera 11 side, and the other The actual distance and direction in the real space between the reference point 18 and the person position 51 on the camera 11 side are compared. As a result, the persons having the sameness in the distance and the direction are determined to be the same person. In this person identification, when a plurality of persons 5 are present in the overlapping visual field region 30, the same person is identified by individually comparing all of the plurality of persons.

As for the identity at the time of person identification, it is preferable to set the value within the model base width as the allowable value for the distance and the direction in the matching determination. For example, if the width of the person model base is about 42 cm, the allowable value is set to 30 cm, and if the difference in the distance and direction in the matching determination is within the allowable value, it can be determined that they are the same person. it can.

Next, a person tracking method by the person tracking system 1 will be described.
The plurality of cameras 11 are installed so that the visual field regions 13 of two adjacent cameras 11 have overlapping visual field regions 30 that overlap at the end regions (see FIG. 1 ). Then, as an initial setting, two markers are arranged in the overlapping visual field region 30 by a marker, and these two points are set as common points 15 and 16 which are treated as the same position between the visual field regions 13 of two adjacent cameras 11. (See Figure 2). By the common points 15 and 16, the visual field regions 13 of the two adjacent cameras 11 are associated and connected. When the shared points 15 and 16 are determined, the midpoint 17 is arranged at an intermediate position in the real space between the two shared points 15 and 16. A point selected from the shared points 15 and 16 or the midpoint 17 and used for identifying a person is used as a reference point 18. Here, for example, the midpoint 17 is set as the reference point 18 (see FIG. 3). A person in the monitoring area 6 is tracked by using the plurality of image processing apparatuses 2 having the above settings.

For example, as shown in FIG. 2, when a person 5 is newly detected in the visual field region 13a on the side of one of the cameras 11a and 11b between the two adjacent cameras 11a and 11b, a new person ID is assigned to this person 5. The assigned person 5 is indicated by the representative point 14 and is traced. When the person 5 enters the overlapping visual field area 30ab, the other camera 11b detects the person 5 as a new person 5 in the visual field area 13b. Is assigned and the person 5 is indicated by the representative point 14 for tracking. That is, in the overlapping visual field region 30ab, a person ID is given to the person 5 detected by each of the adjacent cameras 11a and 11b, and the person 5 is tracked in each of the visual field regions 13a and 13b for each of the cameras 11a and 11b. ..

Referring to FIG. 5, with respect to the person 5 in the overlapping visual field region 30ab, one camera 11a side obtains the position coordinates of the person position 51 of the person 5 detected by himself/herself, and uses this to obtain the person position coordinate data (person Position information) to the other camera 11b side, and the other camera 11b side also obtains the position coordinates of the person position 51 of the person 5 detected by himself/herself, and uses this as person position coordinate data (person position information). It is transmitted to one of the cameras 11a (the coordinate conversion process unit 25 and the communication process unit 26 shown in FIG. 5). At this time, as shown in FIG. 6, when there is an angular difference θ between the coordinate systems 19a and 19b on the images of the cameras 11a and 11b (view fields 13a and 13b), the two shared points 15 and The person position coordinate data (see the above equation 1), which has been subjected to coordinate conversion using the rotation matrix obtained from 16, is transmitted to the other cameras 11a and 11b.

Then, the adjacent image processing devices 2a and 2b perform person identification by comparing the person position coordinate data received from the other party with the person position coordinate data detected by itself (see FIG. 3 and FIG. 5). Judgment process unit 27). That is, the distance dp1 and the direction α1 between the person position 51 to be detected by itself and the reference point 18 are calculated based on the person position coordinate data in the own image processing apparatus 2a, and the distance dp1 and the direction α1 are received from the other party image processing apparatus 2b. The distance dp2 and the direction α2 between the person position 51 detected by the other party and the reference point 18 are calculated based on the person position coordinate data. The distances dp1 and dp2 and the directions α1 and α2 are compared to obtain the degree of coincidence, and the persons having the highest degree of coincidence and having the same identity are determined to be the same person. At this time, the distances dp1, dp2 and the directions α1, α2 between the person position 51 and the reference point 18 are expressed by the distance (actual distance) and the direction in the world coordinate system. The person identification in one image processing apparatus 2a and the person identification in the other image processing apparatus 2b are based on common person position coordinate data through bidirectional communication between these image processing apparatuses 2a and 2b. Since the processing is performed based on the reference point 18 that is the same position in both world coordinate systems in the real space, the same result can be obtained as the person identification result in both the image processing devices 2a and 2b.

Then, regarding the person 5 determined to be the same person, the person IDs assigned by the image processing devices 2a and 2b are integrated into one of the person IDs, and the person 5 exits from the overlapping visual field area 30ab and at the same time The person tracking information of the person 5 on the image processing apparatus 2a side, which also escapes from the area 13a, is passed to the image processing apparatus 2b of the other party and handed over. In this way, the person tracking information of the same person is transmitted between the two adjacent cameras 11a and 11b to cooperate with each other. Finally, when the person 5 moves out of the visual field region 13 without the image processing apparatus 2 on the camera 11 side to which the person tracking information should be transmitted, the image processing apparatus 2 on the last camera 11 side transmits the person 5 concerned. The person tracking information up to now is output to the aggregation device 4.

The image processing devices 2 described above have the same configuration and operate with the same program. Therefore, the setup of each camera 11 is easy, and the person tracking system 1 can be quickly and inexpensively constructed in the wide area monitoring region 6.

Further, two shared points 15 and 16 and a midpoint 17 between the shared points 15 and 16 are arranged, and one point selected from the shared points 15 and 16 or the midpoint 17 is used as a reference point 18 to the person position 51. Person identification is performed based on the distances dp1 and dp2 and the directions α1 and α2. As a result, the two cameras 11 are linked and connected only by the common points 15 and 16, so that stable and accurate person identification can be performed. If person identification is performed based on the distance from the camera position to the person position, the distance between one camera position and the person position, the distance between the other camera position and the person position, and the distance between the cameras are evaluated. It will be. In this case, the distance between the camera position and the person position and the distance between the cameras become long, and in addition, the error of each distance becomes large due to the installation conditions of the camera 11 and the like, and the person identification can be performed stably and accurately. Becomes difficult. On the other hand, by disposing the common points 15 and 16 in the overlapping visual field region 30 of each camera 11 and interlocking with each other and disposing the reference point 18, the person position 51 and the reference point 18 on one camera 11 side are By performing the person identification by evaluating the distance between the reference position 18 and the person position 51 on the other camera 11 side, each distance is short, so that even if there is a distance error, the camera position can be reduced. The person can be identified based on only the recognition of the tracking history of the person's position without recognition, and the person can be stably and accurately identified.

As described above, according to this person tracking system, the camera images of the plurality of image processing apparatuses 2 are associated and connected only at the common points 15 and 16, and the communication between the image processing apparatuses 2 for connection is performed by the person position. Since the image data is not communicated only with the coordinate data and the person tracking information, the communication amount is small. As a result, since the person 5 can be tracked only by the relationship between the two adjacent cameras 11, the unified map information over the entire monitoring area 6 for mapping the person position information from each camera 11 is obtained. Is unnecessary, and there is also no need for an overall server element that brings together a plurality of cameras 11, and the number of cameras 11 for tracking is virtually unlimited. Also, large-scale person tracking can be performed over a wide monitoring area 6.

It should be noted that the present invention is not limited to the above-mentioned embodiment, and it is possible to make necessary modifications within the scope of the claims.

In the above embodiment, tracking of a person is taken as an example, but by using a model base of a moving body other than a person (for example, a car), the model base is collated and the moving body position is calculated. The moving body can be identified with high accuracy by the same method as in the embodiment.

1 Person Tracking System 2 Image Processing Device 3 Network 4 Aggregation Device 5 Person (Mobile)
6 monitoring area 11 camera 12 processor 13 field of view area 14 representative points 15 and 16 common point 17 middle point (common point)
18 reference point 21 common point recording unit 22 person tracking history recording unit 23 image processing process unit 24 tracking processing process unit 25 coordinate conversion processing unit 26 communication processing unit 27 coincidence determination processing unit 30 overlapping visual field region 51 person position (moving body position)
52,53 Common point position (ground height 0)
dp1, dp2 distance α1, α2 direction θ angle difference

Claims (6)

A method of tracking a moving object in a surveillance area in which the field of view of a plurality of cameras is set,
Has an overlapping field of view in which the field of view of two adjacent cameras partially overlaps,
Place a common point to be treated as the same position between the view areas of two adjacent cameras in the overlapping view area,
In the case of having only one common point, this one point is used as a reference point, and in the case of having a plurality of common points, one of these points is used as a reference point,
The moving body detected by each camera in each field of view is collated with the moving body model base, and the moving body is represented by a representative point and tracked,
With respect to the moving body detected by each of two adjacent cameras in the overlapping visual field area, the actual distance and the direction of the moving body position in the real space of the moving body represented by the reference point and the representative point are the same. It is determined that the moving bodies having the same are the same moving body,
A moving body tracking method for transmitting moving body tracking information on one camera side to another adjacent camera side for moving bodies that are determined to be the same moving body so that two adjacent cameras cooperate with each other. The moving body tracking method according to claim 1,
Any two or more of these common points are set by arranging two or more landmarks in the overlapping visual field area, adding these two or more landmarks as common points, and adding any points that can be calculated from these landmarks as common points. A method for tracking a moving body, wherein the above is used as the reference point. The moving body tracking method according to claim 1,
When there is an angle difference between the coordinate systems on the images captured by the two adjacent cameras, at least two common points are arranged, and the moving body position information converted using the rotation matrix obtained from the two common points is used. A moving body tracking method for determining the same moving body using the same. The moving body tracking method according to claim 1,
A moving object tracking method in which when the coordinate systems on the images captured by two adjacent cameras do not have an angle difference, only one shared point is arranged and this one point is used as the reference point. The moving body tracking method according to any one of claims 1 to 4,
When the moving body detected by one of the cameras moves out of the field of view, if there is no other camera side to which the moving body tracking information should be transmitted, the moving body that outputs the moving body tracking information of this moving body to the tallying device Tracking method. An image processing apparatus commonly used in the moving body tracking method according to claim 1.
Has a camera and a processor,
The processor is
In the overlapping visual field area where the visual field area of the own camera partially overlaps with the adjacent visual field areas of the other cameras, common points to be treated as the same position between the respective visual field areas are recorded, and further, only one common point is provided. In the case of using this one point as a reference point, and in the case of having a plurality of the common points, one of these points is recorded as the reference point,
An image processing process unit that represents the moving body as a representative point by comparing the moving body detected by the own camera with the moving body model base,
A tracking processing unit that tracks the moving object captured by the image processing unit,
A communication process unit that communicates the moving body position information of the moving body being tracked with another image processing device on the other camera side,
The actual distance and direction of the moving body position in the real space of the moving body represented by the reference point and the representative point with respect to the moving body detected by each of the other cameras adjacent to the own camera in the overlapping visual field area. An image processing apparatus comprising: a matching determination processing unit that determines that moving objects having the same identity are the same moving object.

Images