JP4759988B2 - Surveillance system using multiple cameras - Google Patents

Description

  The present invention relates to a monitoring system for recognizing and collating or tracking a moving body, for example, a person, a car, or the like taken by a plurality of cameras installed in a target space.

  In order to cope with social unrest such as a decrease in crime clearance rate, the number of cameras for monitoring suspicious persons / suspicious vehicles is increasing for the purpose of maintaining and improving the security level. In the initial stage of introduction, the surveillance staff visually observes the video taken with the camera in the surveillance room, dispatches police and security guards to the scene in abnormal situations, or stores the video in the recorder for investigation after the crime occurs Using direct and indirect methods such as making use, it has contributed to crime prevention in advance and countermeasures. However, due to the rapid increase in the number of installations, not only cameras but also the number of monitoring personnel and recorders have been increased, resulting in increased costs and insufficient resources for security systems. As a countermeasure against this problem, there has been proposed a method for reducing the visual monitoring burden on a monitor by automatically recognizing a monitoring target such as a person or a car in a camera image using a computer. For example, Patent Document 1 discloses a technique in which, for each of a plurality of cameras, the whole person feature information and face feature information are extracted using an image recognition technique, and the extracted information is exchanged between the cameras.

JP 2003-324720 A

  In monitoring in which a large number of cameras are installed, monitoring support technology is required to enable the work to be efficiently performed with limited monitoring personnel resources in the monitoring target space. In order to ensure the security of the monitored area, in addition to the above-mentioned monitoring by the camera, if it is inside a building, take measures to allow only authorized persons to enter and exit each room. Is normal. As an unlocking method, not only conventional keys but also security code entry, IC card use, and personal identification by biometrics authentication such as fingerprints and veins are being introduced at a higher level. . As described above, in order to secure the security level, it is necessary to both increase the number of surveillance cameras and restrict access by locking. However, excessive system introduction will increase costs and inconvenience the user. It is necessary to be able to combine systems flexibly according to the security level to be achieved.

  Based on the requirements for the system described above, when a known technique is examined, as described in Patent Document 1, an image of a moving body collated with overall feature information is displayed on a monitoring monitor, or a moving body that is monitored by a monitoring person In this method, it is necessary to determine whether the monitoring person himself is suspicious or to estimate the movement path in the monitoring space of the target moving body, which is insufficient for reducing the monitoring burden. In addition, in cases where an unspecified number of people can access a specific area (for example, a corridor in a building), it is not practical to track everyone visually when the number of visitors is large, and applicable scenes are limited. The

  An object of the present invention is to improve the tracking accuracy of a moving object reflected on a surveillance camera.

  As means for solving the problem, first, a feature amount extraction means for installing a plurality of cameras in the monitoring target space and extracting the moving object and its feature amount information from the video using an image recognition technique is provided. In this feature amount extraction means, information extraction is repeatedly executed at predetermined time intervals or timings.

  The video captured by the camera and the extracted feature information of the moving body are sent to the moving body collation and tracking means via the communication network. The verification tracking means is composed of a monitoring space database, a moving object separation and aggregation means, a route calculation means, a moving object matching degree calculation means, a personal identification information matching means, a restricted area intrusion determination means, a behavior analysis / suspicious person detection means, which will be described later. .

  The verification tracking means expresses the entire monitoring target space as a three-dimensional model, and further has a monitoring space database that expresses the connection of the space in which the mobile body can move, centered on the network, and accumulates the transmitted feature information. In the surveillance space database, the position of each camera and its photographing range are correlated with the three-dimensional model and the network model, and the feature amount information is stored in association with the camera information.

  Here, since a plurality of moving bodies generally exist in one camera image, it is necessary to separate them as individual moving bodies. Therefore, the moving object separation and aggregation means that operates at a predetermined interval takes out the feature amount information for each camera stored in the database, and the distance from the feature amount set that has already been mapped to the feature amount space and grouped Using the information on the approximate proximity and the estimated moving direction / distance in the video space of the extracted moving object, it is determined which feature set belongs to. Here, the feature amount set refers to a collection of information obtained each time the feature amount extraction processing is performed on the same moving object, from one moving object entering one camera field of view until leaving. Using information, it is also possible to represent the feature value vector by its average value and its variance value.

  Next, in order to track the separated / aggregated moving body, a route calculation unit in the verification tracking unit obtains a movement route candidate between the camera photographing ranges in the monitoring target space. In the route calculation, the minimum value of the set link cost between the departure point and the destination point is searched for information modeled as a network in the database of the connection state of a person's movement space such as a passage or a room. If the configuration of the monitoring space is simple, it is possible to obtain all routes in all combinations between cameras in advance as a table.

  Then, the moving body matching degree calculation means in the verification tracking means calculates the matching degree between the feature amount sets of the two moving bodies. The degree of matching is defined as an index of closeness in the feature amount space of the two sets. At this time, the combination of the feature value sets extracted one after another each time the moving object passes through each camera field of view is the key to realizing appropriate tracking. Therefore, first, using the information on the moving direction when the moving body passes through the upstream camera field of view, the route calculation means explained above determines the route to the downstream camera that can be reached by the moving body, and the required time is calculated. presume. At this time, the route passing through the stairs and the elevator is obtained with a weight different from that of the passage. Then, the degree of matching with the moving body that has passed within a predetermined time from the required time is obtained. Whether or not the two moving objects match is determined by setting a determination value corresponding to the degree of matching. By changing this determination value, it is possible to narrow or widen the tracking target. In addition, in order to cope with the stay of the moving body, when there is no upstream side moving body corresponding to the moving body captured by the downstream camera by calculating the required time, the moving body goes back to the predetermined degree of matching. Ask for.

  It is necessary to show the processing described above to the monitoring staff using the display means. The display means includes a function of viewing the monitored space three-dimensionally and setting and displaying various viewpoints, and a function of displaying the position of the installed camera and its video. In the display of the camera video, the monitor selects a specific camera, or a plurality of camera videos are automatically switched and displayed on a single display unit with a time interval. By displaying the number of passing people in each moving direction on the video based on the feature amount information extracted for each camera, the monitor can grasp the congestion situation even when the video is not watched.

  In addition, when a monitor specifies a specific moving object in the video, the result of estimating the moving path of the moving object is displayed, and tracking is automatically performed by obtaining the degree of consistency between cameras. Is possible. In the case where the moving body becomes sparse, such as at night, it is possible to track all the moving bodies detected by the image recognition technology without specifying by the observer. Conversely, when there are many moving objects such as during the day, even if the tracking results are superimposed on the display, it becomes difficult to understand, so the tracking information of only the target designated by the observer is displayed. Here, by allowing the observer to arbitrarily change the judgment value corresponding to the degree of consistency, the tracking target can be expanded or narrowed, and the tracking failure can be prevented. become.

  On the other hand, the verification tracking means includes personal identification information matching means such as card authentication and biometric authentication installed on the door. Card authentication can be associated with specific information of the cardholder. Therefore, the vicinity of the door with the card authentication means is photographed by the monitoring camera, the feature value information of the appearance is extracted, and at the same time, the card information is associated with the owner information. By associating the apparent feature information with the personal information, it is possible to determine which area of the monitoring target the person can enter. This is determined by the restricted area intrusion determining means also in the verification tracking means. At this time, the restricted area intrusion determination means is connected to an open / close sensor attached to the door, and the open / close information is used for presence / absence determination of intrusion.

  At this time, if an unauthorized person enters the restricted area, warning information is displayed on the display means, and the monitor dispatches security personnel to the scene as needed, or leaves the area with speakers. Can be called.

  The verification tracking means analyzes the movement trajectory of the moving body to determine whether it is a first visitor, a normal user, or a suspicious person, passes through the monitoring space database, and enters the restricted area by the display means. A behavior analysis / suspicious person detection means for displaying an alarm similar to that of the intruder is provided.

  According to the present invention, it is possible to improve the tracking performance for a moving object in the monitoring target area, thereby reducing the burden on the monitoring staff and providing a security solution that is not troublesome for the user.

  An example of a monitoring system using the present invention will be described with reference to the drawings.

  An embodiment will be described with reference to FIG. In the monitoring system, first, a plurality of combinations of the camera 101 and the moving body feature amount extraction device 102 and the camera 103 and the moving body feature amount extraction device 104 are installed in the monitoring target space, and each is connected to the network 105. Although FIG. 1 shows a configuration in which two sets are connected, more sets can be connected. The verification tracking device 106 is connected to the camera and the moving object feature amount extraction device via the network, and receives video and feature amount information therefrom. The card authentication device 117 and the door opening / closing sensor 118 are connected to receive card information and door opening / closing information from the respective devices. The verification tracking device 106 is also connected to a monitoring display 107 that displays a camera image, a movement trajectory of a moving object to be monitored, an intrusion warning to a restricted area, and the like, and has a role of transmitting a situation to a monitoring person.

  The verification tracking device 106 includes a monitoring space database 108, a feature amount information reception storage function 109, an intra-camera moving object separation and aggregation function 110, an inter-camera moving path calculation function 111, an inter-camera moving object matching degree calculation function 112, and a monitoring status display function. 113, a card authentication matching function 114, a restricted area intrusion determination function 115, and a behavior analysis suspicious person detection function 116. Details of each function will be described later. Note that the function of the verification tracking device 106 described below can be implemented not only in one device but also distributed in a plurality of devices.

  Next, the moving body feature quantity extraction devices 102 and 104 will be described. This apparatus has a function of extracting feature amount information from an image captured by a camera by applying an image recognition method and outputting the feature amount information to the outside. The processing procedure flow will be described with reference to FIG. First, in step 201, an image captured by the camera is captured. The camera can be realized by any method such as an analog signal output such as NTSC format or a digital output such as JPEG format. In the case of the analog format, the digital information is used for image recognition processing. And stored in a memory in the apparatus.

  Next, in step 202, a moving body existing in the video is extracted from the stored digital video information. The background subtraction method is applied as a specific method of extraction. The background subtraction method saves video data in a state where there is no moving object as the background, and obtains the difference between the video data showing the moving object and the luminance value for each pixel. This is a method for obtaining a region of a moving object. The role of this step is to divide the moving body region into individual persons using information such as the size of the person.

  Next, in step 203, each feature amount is calculated from the area divided for each person in step 202. Here, a method for obtaining the most frequent luminance values of the lower body and the upper body as the feature amount will be described. First, since only the region where the whole body of a person is shown is processed, the processing is skipped when it falls on the upper, lower, left, and right boundaries of the image. Next, using the information such as the contour shape variation point and the luminance value variation point from the whole body region, the upper body part and the lower body part excluding the head are determined. Then, the luminance values of all the pixels in each partial area are aggregated, and the luminance value having the highest appearance frequency is set as the representative feature amount of the partial area. By this processing, two feature quantities (upper body mode luminance value and lower body mode luminance value) can be obtained. Further, by obtaining position information (for example, the position of the center of gravity of the upper body) in a person image and managing it together with the feature amount information, the movement speed and the movement direction can be obtained. As the feature amount, it is also possible to use a combination of the contour shape size and color information.

  Finally, in step 204, feature amount information (usually a plurality of information) obtained from one camera image is sent to the verification tracking device. Information to be transmitted includes information such as camera ID, video acquisition time, and local ID. The processing from step 201 to step 204 described above is repeatedly executed unless end designation is set in step 205. The value of the processing period is determined in consideration of the processing time in the apparatus to be mounted, the frame rate of the camera, and the moving speed of the moving object.

  Next, the monitoring space database 108 in the verification tracking device 106 will be described with reference to FIG. FIG. 3 shows the configuration data of the database. The monitoring space database 108 includes a building database 302, a user database 311, and a monitoring database 314, and the respective data are referred to each other.

  The building database 302 includes three-dimensional appearance shape data 303, floor space data 304, elevator data 305, stairs data 306, camera data 307, card authentication device data 308, door opening / closing sensor data 309, and network data 310. .

Here, the building database will be described with reference to FIG. The appearance shape three-dimensional data 303 is a representation of the monitoring target space by a three-dimensional model 401 in FIG. 4, and this data is used in order to make it easy to understand for a monitor or the like. The floor 402, the elevator 403, and the stairs 404 are shown as examples of the space-specific space data 304, the elevator data 305, and the stairs data 306, respectively. These three data indicate places where a person as a moving object may exist.

FIG. 5 shows an example of facility arrangement on the floor. On this floor, there are rooms 501 to 504, which are surrounded by a passage. Elevator 505 to the floor
People go in and out using 506 or stairs 507 and 508. A passage through which a person passes is photographed by cameras (and feature quantity extraction devices) 510 to 512. It is assumed that the doors of each room allow people to enter and exit from doors 517, 518, 519, and 520 with open / close sensors or doors 513, 514, 515, and 516 with card authentication devices. The camera data 307 in the building database 302 includes data on the position and shooting range attached to the floor in this way, and the card authentication device data 308 includes data on the location of the card authentication device and the room and passage connected as the entrance / exit. Similarly, the door opening / closing sensor data 309 includes data of a room and a passage connected as a sensor position and an entrance / exit.

  Next, network data 310 for use in route search will be described with reference to FIG. First, a skeleton representation of a space where people move, such as passages and rooms, is used as the base of the link. A point where a person's flow branches and joins, a point where a person enters and leaves the space by a door, an elevator, or a staircase is defined as a node, and a node is expressed as a link that allows movement of a person. FIG. 6 shows that this floor is expressed by a total of five networks, a network composed of passages and a network constructed for each room. As nodes, door nodes, elevator nodes, stair nodes, branch nodes, and terminal nodes are defined. A node is a node connecting a passage and a room, a passage and a passage, and a room and a room, and plays a role of connecting separate networks. As shown in FIG. 7, an elevator node is defined on each floor of the elevator, and the nodes are connected by links connecting the floors, and plays a role of connecting networks between the floors. The stairway node also plays a role of connecting networks between floors like the elevator node, but there is only one person in the link row connecting the elevator nodes (because there is only one car in the elevator). ), The difference between the links connecting the staircase nodes is that there is no such restriction. A branch node is a node defined at a branch point where no person enters or exits, and a terminal node is a node defined at the end of a link where no person enters or exits. By using the network data described above, it is possible to generate a route from a node link on an arbitrary floor to a node link on another arbitrary floor.

The user database 311 in FIG. 3 is a collection of information on specified persons who have the authority to access the space, and is composed of organization member data 312 and registered visitor data 313. Specifically, it includes information on the groups and companies occupying each floor, the organization structure, and the personnel list that constitutes the organization, information on access rights to each room at each organization and individual level, and each individual Card information and biometrics information. The same applies to registered visitor data, and information on visitors who temporarily access the floor is recorded in the same format as the organization member data. When an individual is identified by card authentication or other authentication (including biometric authentication such as face authentication, fingerprint authentication, vein authentication, voice authentication, etc.), the information in this database is referred to, and access permitted for the individual is permitted. An area can be specified. If a mobile object that is not linked to specific organization / individual information enters the restricted area, it is subject to alarm output.

  The monitoring database 314 in FIG. 3 accumulates data mainly sent from the camera and the feature quantity extraction device, and accumulates data such as the calculated moving body, moving path, and degree of matching. 315, extracted feature value data 316, in-camera verification tracking data 317, and inter-camera verification tracking data 318. Many functions in the verification tracking apparatus described below also use this database as a work area.

  Next, the feature amount information reception storage function 109 in FIG. 1 will be described below. This function receives video and feature amount information sent from a plurality of cameras and moving body feature amount extraction devices via a network, and extracts camera feature data 315 in the monitoring database 314 in the monitoring space database 108, respectively. It plays a role of storing in the quantity data 316. To store all video data, a large amount of storage capacity is required, so past data is deleted after the expiration date, or only scenes where people are shown are selectively recorded in the storage means. .

The in-camera moving object separation and aggregation function 110 in FIG. 1 will be described below. FIG. 8 exemplifies a graph in which snaps of images taken by each camera and feature amounts extracted within a predetermined time are plotted on a graph. Here, the snap image 801 and the extracted feature amount graph 802 of the camera 509 are illustrated in FIG. 8A for three cameras among the cameras arranged on the floor illustrated in FIG. 5, and the extracted feature amount graph 802 is illustrated in FIG. 8B. The snap image 811 and the extracted feature amount graph 812 of the camera 510 are illustrated, and the snap image 821 and the extracted feature amount graph 822 of the camera 512 are illustrated in FIG. 8C, respectively. In the snap image 801, three persons are extracted, the person 803 moves to the back side of the image, and the persons 804 and 805 move to the front of the image. The extracted feature amount graph 802 is obtained by plotting the extracted values on a two-dimensional graph representing the feature amount space with the upper body mode luminance value and the lower body mode luminance value as feature amounts as exemplified above. It is. Whereas the snap video 801 represents only one scene at a certain time, the extracted feature amount graph 802 plots and displays the feature amounts extracted during a predetermined time. As described with reference to the flowchart of FIG. 2, the moving object feature amount extraction apparatus extracts three persons 803, 804, and 805 from the snap image 801, and plots the images on the extracted feature amount graph 802. Three pieces of data are obtained and stored in the monitoring space database 108 via the feature amount information reception storage function 109. Due to differences in the movement time of the person in the camera field of view, fluctuations in the external environment such as lighting, fluctuations in the size and orientation due to the movement of the person, a time difference occurs in the extraction time of the feature quantity, and the feature quantity for the same person also varies. The in-camera moving body separation / aggregation function 110 plays a role of organizing feature quantities extracted with time lag and accompanying fluctuations as individual moving body information. That is, in the graph 802, feature quantities plotted apart are collected as respective data sets such as data sets 806, 807, and 808, and are associated with a moving object. In this example, the data set corresponding to the person 803 corresponds to the data set 808, the data set corresponding to the person 804 corresponds to the data set 807, and the data set corresponding to the person 805 corresponds to the data set 806, respectively. Expand the description.

The processing procedure of the in-camera moving object separation and aggregation function 110 will be described with reference to the flowchart of FIG. As for the information sent from the feature quantity extraction device, the video data is accumulated in the camera video data 315 of the monitoring database 314, and the feature quantity data is accumulated in the extracted feature quantity data 316. Stored as. In order to organize the sent data as moving body information, first, feature amount extraction information of an unprocessed camera is selected (step 901), and unprocessed time information that has not yet been organized is selected (step 901). 902). Next, by the camera / time-specific feature amount separation / aggregation process in step 903, the information is added to the existing mobile object information, registered as a new mobile object, or processed as an exiting mobile object. The processing result is stored in the in-camera verification tracking data 317 of the monitoring database 314. If information on another time that has not yet been processed remains (step 904), the processing from step 902 is repeated, and if extraction information on another camera that has not yet been processed remains (step 904). 905), the processing from step 901 is repeated.

  Next, details of the camera / time-specific feature amount separation / aggregation processing in step 903 will be described with reference to the flowchart of FIG. First, in step 1001, it is confirmed whether or not there is a data set (referred to as an in-camera existing moving body) already registered as a moving body in the camera for the camera information to be processed. If it does not exist, this is equivalent to a new moving object entering the camera field of view. Therefore, in step 1002, a data set is newly created as a moving object that newly appears in the camera with all unprocessed feature information. And register (this is called a new in-camera moving body). When 801, 811 and 812 in FIG. 8 meet this condition, three, two and one new moving bodies in the camera are registered, respectively.

  On the other hand, when there is an existing moving body in the camera, the feature amount information is either added as data as the same person as the existing moving body or registered as a new moving body. In order to determine this, two distance quantities, a spatial distance and a Mahalanobis distance, are used. First, unprocessed feature quantity information is selected (step 1003), and the spatial distance between the latest information in-camera position and the selected feature quantity in-camera position in the data set of all existing moving bodies in camera. Are calculated respectively (step 1004). As the position of the latest information, not only the immediately preceding data, but also, for example, a movement predicted position for the time using a Kalman filter may be used. This Kalman filter is a technique for obtaining a predicted value at a future time using measurement data observed as a time series, and improving the prediction accuracy using new measurement data. Further, the Mahalanobis distance between the feature quantity set and the selected feature quantity among the data sets in all the existing mobile bodies in the camera is calculated (step 1005). Mahalanobis distance is an index used in statistics, and is the value obtained by dividing the distance between the average value of the data set and the target value by the standard deviation of the set, and the proximity of the data set taking into account the distribution and one data. It represents. Here, it uses as a parameter | index which shows the similarity degree of a feature-value. This is repeated for all unprocessed feature values (step 1006).

  Next, in step 1007, a set whose magnitude is the minimum value is obtained from all the vector sets including the calculated spatial distance and Mahalanobis distance. Since the smallest vector size does not coincide with the existing moving object, a predetermined threshold value is set, and it is checked whether it is within the range of the value (step 1008). If it is within the range, as in step 1009, the feature amount information of the selected set is added to the existing moving body information in the camera, and this set is removed from the next selection candidate. Step 1007 and subsequent steps are repeated until there is a candidate existing mobile object (step 1010). In step 1008, if the vector size exceeds the predetermined range, the remaining existing moving body information is extracted from the camera field of view or the feature quantity that should be extracted by overlapping with other moving bodies is extracted. It is thought that it was not done. Therefore, when the latest position of the target existing mobile object is within the vicinity of the exit point with reference to, for example, the network data 310 of FIG. 3, it is considered that the mobile object has exited, and the existing mobile object information Is added to the information indicating that and deleted from the registered information as the existing moving body in the camera (step 1012). Finally, the unprocessed feature quantity that has not been associated with any in-camera existing moving body is registered as a new in-camera moving body as having newly entered the camera view (step 1011).

By performing the processing described above, information on the camera 509 can be aggregated as data sets 806 to 808 in the graph 802, and can be associated with individual persons in the video 801. In addition, regarding the information of the camera 510, the data sets 815 and 816 can be separated in the graph 812 and can be associated with the persons 814 and 813 of the video 811 respectively, and the information of the camera 512 is also the data set 824 in the graph 822. And the person 823 in the video 821 are associated with each other.

  Next, the inter-camera movement path calculation function 111 in FIG. 1 will be described below. The movement route calculation is targeted for network expression data composed of nodes and links as shown in FIG. 6, for example. The route search method is based on the Dijkstra method, which is a method for searching for a route with the lowest link cost. By applying the Dijkstra method, it is possible to obtain the minimum link cost path from the designated node to all remaining nodes. This Dijkstra method is an algorithm that determines the minimum value route to each node one by one from the starting point for the cost set for the link of the network, and spreads it throughout the network. It is widely used as a basic algorithm for shortest path search. In this case, it is necessary to select a link cost corresponding to the travel time. Therefore, in the floor, the distance between the nodes multiplied by the average moving speed of the person, between the staircase nodes multiplied by the staircase distance multiplied by the stair climbing average speed, and between the elevator nodes in the elevator hall Use the average waiting time plus the time required to move the elevator between floors (including the time to enter and exit the door). When using an index having no time dependency in the link cost as described above, the minimum weight path between the cameras and between the nodes can be obtained in advance offline, for example, the monitoring space database 108. The network data 310 can be stored so that a route can be obtained simply by referring to it.

  However, in consideration of the following explanation items, it is necessary to execute a process for obtaining a movement route when it is actually necessary. First, the moving speed within the camera field of view can be obtained from the time transition data at the position of the feature amount information of the extracted moving object. The link cost is changed by applying the calculated moving speed instead of the average moving speed to the link cost setting. In addition, the travel time at the elevator is not a statistical average value, but the link cost is changed by using the actual value obtained from the elevator operation data at that time (which floor at which time). Since movement by an elevator is greatly influenced by the call status and operation management mode on other floors, it is necessary to model the operation of the elevator and estimate its behavior by simulation for high-precision estimation. On the other hand, the moving direction of the moving body from the camera field of view can be obtained from the feature amount information of the moving body, but the path greatly depends on the direction. Therefore, it is necessary to first obtain a route on the network from the starting point of the route search to an intermediate branch point in the moving direction, and then obtain the minimum cost route from the branch point by applying the Dijkstra method.

Based on the above description, the inter-camera movement path calculation function 111 will be described with reference to FIG. First, unprocessed camera data for which route calculation has not yet been completed is selected (step 1101), and moving body information in the camera data is selected (step 1102). Then, the moving body speed / direction is calculated, and the link cost of the network is set using the speed information (step 1103). Next, the link cost between the elevator nodes using the elevator operation data is set (step 1104), and the estimated route to the branch point in consideration of the moving direction is calculated (step 1105). Finally, the minimum cost path by the Dijkstra method is calculated from the branch point (step 1106), and the path is stored in the inter-camera collation tracking data 318 in the monitoring space database 108 (step 1107). The above processing is repeated for all moving objects (step 1108) and for all camera data (step 1109).

The method described above is applied to the floor in FIG. 5 and the network representation in FIG. 6, and a moving body captured by the camera 509 in FIG. 5 will be described as an example. As illustrated in FIG. 8A, the feature amount, moving speed, and moving direction can be obtained from the moving body information of the three persons extracted from the camera image, and the link cost is set using the moving speed. To do. Since the person 803 moves to the back of the video, the route to the branch point can be obtained as the estimated route 1202 in FIG. On the other hand, since the persons 804 and 805 move to the front of the video, the route to the branch point can be obtained as the estimated route 1201 in FIG. Then, the minimum link cost route to all nodes not set as a route is obtained by applying the Dijkstra method from each branch point. Similarly, the estimated route 1203 and the estimated route 1204 can be obtained as estimated routes for the camera 510 and the estimated route 1205 as estimated routes for the camera 512, respectively. In this case, a camera is installed at the branch point, but in general, the branch point and the camera installation point do not match, and it is clearly necessary to obtain a route automatically by route search processing when the network scale increases. Become.

  Here, in the example shown in FIG. 5, since the camera is installed only in the passage, only the network composed only of the links on the passage is the search target, but the camera is installed in the room. In this case, it is necessary to search across multiple networks. In this case, a two-stage search is applied in which a route from the room to the passage is first obtained and then the passage network is searched.

  Next, the inter-camera moving body matching degree calculation function 112 in FIG. 1 will be described below. The purpose of this function will be described with reference to FIG. Here, it is assumed that three persons appear, two persons move from the camera 509 to the camera 510, and the remaining one person moves from the camera 509 to the camera 512. At this time, it is estimated that the person 803 of the video 801 collates with the person 823 of the video 821, and the person 804 and the person 805 of the video 801 collate with the person 813 and the person 814 of the video 811, respectively. This collation requires the above-described movement path between cameras and aggregate data of moving bodies within the camera. When the three pieces of video data in FIG. 8 are viewed over time, first, three people appear in the video 801, move through the camera, and then leave the camera view. The person appears in the video 811 and the video 821 and then exits with a delay of the time the person moves from that time. If there is a person's doorway on the way out of the camera's field of view, the person captured by the upstream camera will not appear in the downstream camera, or the person who is not reflected in the upstream camera will appear in the downstream camera. It may appear. In the present embodiment, only three persons to be monitored are described, but it is usually necessary to assume that people move from one to the next in each camera.

Based on the above, in order to accurately realize the person verification that is the purpose of this function, it is first necessary to narrow down the verification data on the time axis in consideration of the movement time. For this reason, the inter-camera movement path calculation function 111 is required. The route information obtained in step 1 is used. When it is determined that the mobile body extracted from each camera has left the field of view (determined in step 1012 of the flowchart of FIG. 10), the path to the camera installation point on the destination path of the mobile body and Refer to the travel time. And as mobile object information that may reach the camera, a search time window is set before and after the estimated arrival time (exit time plus movement time), and with reference to the mobile object information that has exited, Remember. As an example, FIG. 13 shows on the time axis the case where the persons 804 and 805 in 801 in FIG. 8 move from the camera 509 in FIG. 5 that is the upstream camera to the camera 510 that is the downstream camera. The persons 804 and 805 in the camera 509 are reflected on the camera at the same time, and there is an overlap in time from entry to exit. In addition, the search time window set in the camera 510 also shows that the ranges for two persons overlap.

Next, a procedure for estimating the corresponding moving body between the cameras will be described. FIG. 14 shows the search time window from the camera 509 set on the time axis of the camera 510, and the entry and exit of the person 813 and the person 814 as shown in the video 811 of FIG. It shows a state that is extracted. Here, it is assumed that when the camera leaves the field of view, a process of searching for a corresponding moving body between the cameras is executed. First, when the person 814 leaves, the degree of matching between the moving bodies is calculated using all the upstream moving bodies included in the range of the search time window as candidates. In this case, the degree of matching between the feature value set 815 corresponding to the person 814, the feature value set 807 corresponding to the person 804 and the person 805, and the feature value set 806 is calculated.

Here, in order to define the degree of matching between two data sets, first, a value obtained by a linear discriminant function used in discriminant analysis in statistics is defined as the proximity. A linear discriminant function is an optimal straight line (generally a hyperplane) that separates two groups, and a value obtained by substituting the value of each sample point in each group is called a discriminant score. The discrimination score is an index indicating the degree of separation between the sample point and the linear discrimination function. It is known that the deviation sum of squares of the discrimination scores for all the samples in two groups can be expressed by adding two indices, an index indicating the distance between the two groups and an index indicating the sum of the fluctuations in each group. Since the value obtained by dividing the former index by (total number of samples-1) corresponds to the variance, the square root of the value is defined as the proximity here. Specifically, _assuming that the average of individual groups of discrimination scores is Z1 _ave and Z2 _ave , the overall average is Z _ave , and the number of samples is N1 and N2, respectively, the proximity D12 is defined by the following formula 1.

D12 = √ ((N1 * ( Z1 ave -Z ave) 2 + N2 * (Z2 ave -Z ave) 2) / (N1
+ N2-1)) (Formula 1)
In addition to the definition of proximity, a method using the Mahalanobis distance is also applicable. If the proximity is small, the two moving bodies are likely to be the same, and conversely if the proximity is large, the possibility is high that they are different. Therefore, a predetermined range of proximity is normalized from 0 to 1, and a value subtracted from 1 is used as the degree of matching defined between the two moving objects, so that the index becomes easier to understand. When the lower limit value of the proximity is Dmin and the upper limit is Dmax, the matching degree M12 corresponding to the proximity D12 is defined by the following equation 2.

M12 = 1− (D12−Dmin) / (Dmax−Dmin) (Expression 2)
However, M12 = 0 when D12 is larger than Dmax, and M12 = 1 when D12 is smaller than Dmin.

When the degree of matching defined above is obtained between the feature quantity set 815 and the feature quantity set 806 and between the feature quantity set 815 and the feature quantity set 807 in FIG. 8, the combination of the feature quantity set 815 and the feature quantity set 806 is obtained. The matching degree is a value close to 1. If the value is larger than the set determination value, the two sets are determined as matching. If they match, the corresponding search time window information for 805 is deleted, and the reference to the moving body information of the corresponding person 805 on the upstream side is recorded in the exited 814 moving body information. As a result, the tracking including the moving route becomes possible. Similarly, when the person 813 leaves, since the corresponding moving object in the search time window is only for the person 804, the degree of matching between the feature quantity set 816 and the feature quantity set 807 is calculated. , Whether it becomes a value larger than the judgment value. In this case, since the value is close to 1, it is assumed that the two sets match, the corresponding search time window information for 804 is deleted, and the moving body information of the person 813 who has left is changed to the moving body information of the person 804. Record the references.

  Basically, by performing the above processing for each piece of information corresponding to each camera, it is possible to calculate the degree of matching between moving bodies and determine whether there is a match. This process can be executed at any time from entry to exit, such as when entering the camera field of view or when a predetermined time has elapsed since entering. Here, when an actual scene is assumed, it is necessary to assume a case where a person enters or exits in the middle of a route between two cameras, stays for a long time, or turns back in the middle.

  That is, when the moving object tries to calculate the degree of matching with the moving object information of the upstream camera at the timing when the moving object leaves the camera field of view, when the moving object arrives as expected, the search time window as described above. You can select moving body information within the range, but if you have stayed for a long time, the search time window that should be collated has been exceeded, so search past the past time in the window. There is a need. A search time window that is not associated with a past mobile object is searched, and the search is continued until the degree of consistency with the mobile object information exceeds a determination value (= has stayed) or a predetermined search limit time is reached. . Alternatively, if the camera is turned back in the middle, it searches backwards for the camera's own past moving body information, until the matching degree exceeds the judgment value (returned) or until the predetermined search limit time is reached. Continue searching. If a matching moving body is searched for in both cases of staying and turning back, the moving object information of the upstream camera to be tracked is the one closer to the current time. If no corresponding moving body is found in any of the determinations described above, it is assumed that a person has entered from the middle of the route, and the moving body is traced toward the downstream camera with this camera as a base point. . Note that if the search time window in which the corresponding moving object is not found exceeds the search time limit, it is considered that the person of the upstream camera has exited in the middle of the route.

A summary of the above description is shown in the flowchart of FIG. When a moving object exits from the field of view of a certain camera, the information on the moving object is referred to (step 1501), and a search time window that includes the exit time within the range is searched (step 1502). The degree of matching between the moving body information of the upstream camera corresponding to the target search time window and the leaving moving body information is calculated (step 1503), and when the value exceeds the determination value (step 1504), the exiting is performed. A reference to the moving body information of the upstream camera that matches the moving body information is recorded, the corresponding search time window is deleted (1505), and the process returns to step 1502. If it is determined in step 1504 that the matching level is equal to or less than the determination value, the process returns to step 1502. This is executed for all search time windows including the exit time. When the determination value is low, or when similar mobile objects arrive at the same time, the information may match a plurality of upstream mobile object information. That is, in the verification tracking here, not only a single route but also a network including a branch may be formed.

As a result of checking all the search time windows, if an upstream mobile body corresponding to the leaving mobile body is found (step 1506), the process is terminated. If the search is not searched, the degree of consistency between the moving object corresponding to the search time window that has not been deleted or the moving object that has left the camera and the actual moving object is calculated by going back in time. (Step 1507), and when the degree of matching exceeds the determination value (Step 1508), record the reference to the upstream mobile body information matched with the leaving mobile body or the mobile body information exited in the past from the same camera, If it corresponds to the search time window, the information is deleted (step 1511) and the process ends. If the matching degree is equal to or smaller than the determination value in step 1508, the process returns to step 1507 and searches past information as long as the search limit time is not exceeded (step 1509). If the search limit time is exceeded, the information is recorded with the exit mobile body as a reference tracking base point (step 1510), and the process ends.

  Next, the monitoring status display function 113 in FIG. 1 will be described below. This function refers to the information in the monitoring space database 108 and displays the current status or the past status to the monitor through the monitoring display 107. An example of a screen displayed on the monitoring display 107 will be described with reference to FIG. FIG. 16 schematically shows a monitoring console and a display. An overall window 1601 is a main window 1602 showing the whole or a part of the building layout, a status window 1603 showing the monitoring status, and images of each monitoring camera. Is composed of video windows 1604 to 1609 for projecting images.

  Here, the main window 1602 shows a schematic diagram of each floor, a camera arrangement, its number, and a movement display when a person is being tracked and a warning display for entering a restricted area. For the display of this window, the information of the building database 302 (illustrated in FIGS. 4 and 5) and the monitoring database 314 in FIG. 3 is used, but more realistic display can be achieved by using three-dimensional display technology and computer graphics technology. Can also be provided to the observer. On the other hand, it is possible to display using a deformed form that simplifies the building information and displays the camera and other related equipment in an emphasized manner.

  The status window 1603 has a function of displaying warning information obtained by the behavior analysis suspicious person detection function 116 and the restricted area intrusion determination function 118 in FIG.

  Six screens from a screen 1604 to a screen 1609 display the video together with the surveillance camera number. Either the video information sent via the network 105 is directly displayed on each video window, or the video data stored in the monitoring database 314 is read and displayed. In general, the number of video windows for display is often smaller than the number of installed cameras due to the limitation of the display area. At this time, the camera to be displayed in each video window is individually selected and switched for monitoring, or a plurality of camera videos are registered in advance for each video window, and automatically switched and displayed at predetermined time intervals. Use the method to do. In addition, by referring to the in-camera verification tracking data 317, the number of moving bodies that have left the office is counted, so that the number of people passing from a predetermined time can be obtained and displayed in an overlapping manner. It is also possible to display the display separately for each direction of movement.

  An example of a window for setting monitoring conditions including this video display will be described with reference to FIG. Here, for each of the six video windows, either automatic switching display or fixed display can be selected, and in the case of automatic switching, the time interval can be input. Select the camera to be selected from the pull-down menu, and set the camera number and order. On the other hand, the setting related to the display of the tracking status as exemplified in the main window 1602 is exemplified in the lower part of the same monitoring condition setting window. The verification tracking processing is performed at any time by the intra-camera moving object separation and aggregation function 110, the inter-camera path calculation function 111, and the inter-camera moving object matching degree calculation function 112 of FIG. 1 described so far, and the result is stored in the monitoring database 314. Saved. When this data is displayed on the screen, if the number of people increases, a large number of overlapping routes are displayed on the screen, making it difficult for the observer to understand. Therefore, here, the display of the tracking status is basically based on a request from the supervisor. On the other hand, when the number of people is small but the importance of security monitoring becomes high, such as at night, it is necessary to efficiently support the monitoring work for the monitoring staff. Therefore, a function is provided to display information obtained by the tracking process on the main window even if there is no request from the monitoring person for a predetermined number of people or less and for a predetermined set time. This is called automatic tracking display. Settings required for automatic tracking display include ON / OFF of the function, upper limit number of tracking targets, setting of display target time (here, three examples are possible), and selection of a video window dedicated to automatic tracking.

  FIG. 18 shows a screen display example when the automatic tracking display setting is set to ON. In this example, three persons are to be tracked, and three tracking paths 1801 to 1803 are displayed in the main window. The tracking path is displayed with reference to data stored in the inter-camera verification tracking data 318 of the monitoring database 314. In the automatic tracking dedicated video window 1804, when a person to be tracked enters the camera view, the video is displayed. If it is necessary to display multiple camera images at the same time, the requested video is stored separately, and the accumulated video is played back and displayed when the tracked object leaves the camera field of view. To do. Multiple automatic tracking video windows can be set.

  FIG. 19 shows an example of a screen when a tracking target is designated by a request from a supervisor. When a tracking target in an arbitrary video window is selected by a pointing device (1901), a moving path is displayed in the main window (1902), and a camera through which the tracking target passes is highlighted (1903). ). When a video window dedicated to tracking is specified, the image obtained when passing through each camera and the value of the matching degree obtained by the inter-camera moving body matching degree calculation function 112 are displayed in an overlapping manner. This is automatically switched for each passing camera. As a method of displaying the state of consistency, a numerical value of consistency may be displayed as shown in this embodiment, a graph corresponding to this numerical value, or a display window indicating the degree of consistency or a display window indicating the degree of consistency. It is possible to notify by changing the display color.

  Here, FIG. 20 shows an example of a determination value setting window for determining that two moving bodies match with the inter-camera moving body matching degree calculation function. The judgment value can be set for each individual camera. This is because environmental conditions differ depending on the installation location of the camera. FIG. 20 shows an example of setting using a slide bar. In addition, a determination value increase / decrease button is provided to increase or decrease the determination value for all cameras at once, and a function for resetting the determination value of all cameras to 1 by a reset button is executed. This window can be operated directly by the supervisor, but is also used as an adjustment tool for the system setter.

  In addition, when a monitoring and tracking operation is assumed, it may be necessary to determine whether or not the action is suspicious by playing back the previous video in addition to the real-time video information at that time. FIG. 21 shows an example of a monitoring video playback controller for reproducing, fast-forwarding, and rewinding the stored video in synchronization. Select the camera to be displayed in each video window and set the start time. The default is the current time. For playback control, buttons of normal playback, fast forward playback, playback playback, fast forward playback playback, and stop are used.

Next, the card authentication device 117 in FIG. 1 will be described. Here, the card authentication device refers to a device equipped with an IC card information reading function and a password input function installed to unlock the door. In FIG. 5, it is illustrated as 513, 514, 515, 516, and in FIG. 22, the installation position is indicated by an IC card symbol 2202. The card authentication matching function 114 in FIG. 1 connects to the card authentication device 117, searches the user database 311 in the monitoring space database 108, and searches for information that matches the card ID. Then, from the retrieved information, it is confirmed whether or not there is an authority to enter the room that the person is going to enter, and if there is an authority, unlocking is performed. When entry / exit using this apparatus occurs, the ID information is displayed near the symbol shown in FIG. If there is no authority to enter the room, the card symbol is blinked or the color is changed, and an alarm is issued to the supervisor. The access information is stored in the card authentication device data 308 of the monitoring space database 108 as an entry / exit record.

  The card authentication matching function 114 also plays a role of coordinating with the verification tracking processing of the mobile object. For example, by combining with a surveillance camera attached to a position where a person who operates the card authentication device enters the field of view, a feature amount set viewed from the appearance of the human part extracted by the surveillance camera and a specific individual of the IC card owner It becomes possible to connect information. This makes it possible to collate and track what has been tracked as an anonymous person so far in combination with personal information, so for example, it is possible to set a consistency judgment value according to the work location of the individual who is collated and tracked Thus, it is considered that the tracking accuracy is further improved. Also, even when the card authentication device is not in the camera field of view, it is possible to associate personal information that can be referred to from the card ID as a person who moves between cameras or who has entered from the middle of the path between cameras. Become.

  In the present embodiment, the card authentication apparatus is used for the description. However, this apparatus can be used instead of a biometric authentication apparatus such as fingerprint authentication or vein authentication. That is, any device can be used as long as it can refer to information for identifying an individual.

  Next, the door opening / closing sensor 118 in FIG. 1 will be described. The open / close sensor detects opening / closing of the door, and it can be assumed that a person has entered or exited. In FIG. 5, the open / close sensors 517, 518, 519, and 520 are illustrated, and in FIG. The restricted area intrusion determination function 115 in FIG. 1 is connected to the door opening / closing sensor 118, and if the person being tracked is associated with specific personal information when the door opening / closing is detected during the tracking of the person between the cameras. Check whether the room has authority to enter the room, and issue an alarm if there is no authority to enter the room. Alarms are communicated to the supervisor through status window display and door symbol blinking. Similarly, if the person being tracked is not connected to specific personal information, it is checked whether or not an unspecified person can enter the room, and an alarm is issued if it is not possible. Furthermore, an alarm can be transmitted to the target person by installing a speaker in the passage.

  As shown in FIG. 5, by attaching sensors or card authentication devices to all the doors in the room, it becomes possible to estimate the entrance / exit of a person on the way of the movement path between cameras, and as a result, the tracking accuracy is improved.

  Finally, the behavior analysis suspicious person detection function 116 in FIG. 1 will be described. This function generates an alarm when analyzing the movement trajectory of a person and when the possibility that the person is a suspicious person is high. When a person knows the monitoring space well, it is considered that he or she draws a movement trajectory that is straight toward the destination with little fluctuation in movement speed. On the contrary, when the place is visited for the first time, it moves slowly while confirming the moving direction, and it is considered that the fluctuation of the moving locus becomes large. On the other hand, it is considered that a suspicious person with a purpose travels back and forth many times in the same place or takes a long time to move between cameras. Therefore, when the movement trajectory fluctuation exceeds a predetermined value, it is determined as suspicious behavior. For example, the alarm display 2203 in FIG. 22 is output on the screen, or the fact is displayed in the status window. Call attention.

  In the above-described embodiment, the present invention has been described as a monitoring system for a person in a building, but this can also be applied to a vehicle in a road space as a monitoring target.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to provide the monitoring system which recognizes and collates or tracks the mobile body image | photographed with the some camera installed in object space, for example, a person, a motor vehicle, etc. with high precision.

It is a figure which shows the whole structure of the monitoring system using this invention. It is a flowchart which shows the process sequence of a moving body feature-value extraction apparatus. It is a figure which shows the breakdown of the monitoring space database. It is a figure explaining the building composition used as an illustration by this invention. It is the figure which showed the floor structure used as an illustration by this invention. It is a figure which shows the network expression of the floor structure used as an illustration. It is a figure explaining the link of the staircase node and elevator node which connect between two floors. It is a figure which shows the example which plotted the example of the image | video snap image | photographed with the monitoring camera, and the extracted feature-value on a graph. It is a flowchart which shows the process sequence of the moving body isolation | separation collection function in a camera. It is a flowchart of a camera / time-specific feature amount separation and aggregation process. It is a flowchart which shows the process sequence of the movement path | route calculation function between cameras. It is a figure explaining the example of the calculated route. It is explanatory drawing which sets the search time window for collating the mobile body information extracted with the upstream camera with a downstream camera. It is a figure explaining the relationship on the time axis between the feature-value extracted with a certain camera, and the set search time window. It is a flowchart which shows the process sequence of the moving body matching degree calculation function between cameras. It is a figure explaining the display content on the monitor display. It is a figure which shows the example of the window for setting monitoring conditions. It is a figure which shows the example of a display on the monitoring display at the time of automatic tracking display setting. It is a figure which shows the example of a display of the information of the tracking object which the monitoring person selected. It is a figure which shows the example of the window which sets the determination value with respect to a matching degree. It is a figure which shows the example of the controller for reproducing | regenerating and displaying the accumulate | stored video data. It is a figure which shows the example which displayed the information of the card | curd authentication, the door opening / closing sensor, and the behavior analysis suspicious person detection function on the monitoring display.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101,103,509-512 ... Monitoring camera, 102,104 ... Mobile body feature-value extraction apparatus, 105 ... Network, 106 ... Collation tracking apparatus, 107 ... Monitoring display, 108 ... Monitoring space database, 109 ... Reception of feature-value information Storage function 110 ... Camera moving object separation and aggregation function 111 ... Camera moving path calculation function 112 ... Camera moving object matching degree calculation function 113 ... Monitoring status display function 114 ... Card authentication matching function 115 ... Restriction Area intrusion determination function, 116 ... behavior analysis suspicious person detection function, 117 ... card authentication device, 118 ... door open / close sensor, 201 to 205, 901 to 905, 1001 to 1012, 1101 to 1109, 1501 to 1511 ... processing steps of the flowchart 302 ... Building database,
303: Three-dimensional appearance shape data, 304: Space data by floor, 305 ... Elevator data, 306 ... Stair data, 307 ... Camera data, 308 ... Card authentication device data, 309 ... Door open / close sensor data, 310 ... Network data, 311 ... User database, 312 ... Organization / member data, 313 ... Registered visitor data, 314 ... Monitoring database, 315 ... Camera video data, 316 ... Extracted feature data, 317 ... In-camera verification tracking data, 318 ... Camera Inter-tracking tracking data, 401 ... Building exterior image, 402 ... Floor composition image, 403 ... Elevator image, 404 ... Stair image, 501-504 ... Room, 505, 506 ... Elevator, 507, 508 ... Stair, 513-516 ... Card Door with authentication device, 517-52 ... door with open / close sensor, 801 ... camera 509 snap video, 802 ... camera 509 feature plot graph, 803-805, 813, 814, 823 ... person, 806-808, 824, 815, 816 ... feature set, 811 ... Camera 510 snap image, 812 ... Camera 510 feature plot graph, 821 ... Camera 512 snap image, 822 ... Camera 512 feature plot graph, 1201-1205 ... Estimated path, 1601 ... Entire display, 1602 ... Main window, 1603 ... Status Window, 1604 to 1609 ... video window, 1801 to 1803 ... tracking path, 1804 ... video window dedicated to automatic tracking, 1901 ... tracking target selection, 1902 ... moving path display, 1903 ... passing camera display, 1904 ... passing point camera Image display, 2201 ... door opening and closing sensor symbol, 2202 ... card authentication device symbol, 2203 ... the alarm display.

Claims (5)

Multiple cameras are installed in the monitoring target space, and information indicating the feature amount of the moving object is extracted from the video imaged by each camera, and the extracted feature value is used to shoot with each camera. In a monitoring system for determining whether the same moving body is the same,
Intra-camera moving body separation and aggregation means for separating and aggregating feature quantities extracted for each individual camera by moving body, inter-camera movement path calculation means for calculating a movement path between cameras of the moving body, and using the movement path The inter-camera moving body that estimates the moving body that has moved between the cameras, calculates the degree of consistency of the feature set before and after the movement of the moving body that is estimated to have moved between the cameras, and tracks the matching in correspondence A matching and tracking device including a matching degree calculation means and a database for managing the entire monitoring space information;
A display device for displaying the state of the moving body,
A personal authentication device and a door opening / closing sensor are connected to the verification tracking device,
The verification tracking device includes an authentication matching unit that links the mobile body and personal information acquired from the personal authentication device when the mobile body performs personal authentication by the personal authentication device.
The authentication matching means and the inter-camera moving body matching degree calculation means track the moving body in combination with the personal information, and the moving body is only the door opening / closing sensor of the personal authentication device and the door opening / closing sensor. A plurality of cameras are used, wherein when the door provided with is opened, it is determined from the personal information whether the mobile body has the right to enter the room, and an alarm is issued if there is no right. Had a surveillance system. The monitoring system according to claim 1,
A monitoring system using a plurality of cameras, wherein an alarm is issued when the moving body enters an entry restricted area. The monitoring system according to claim 1 or 2,
In the in-camera moving body separating and aggregating means , the spatial distance between the in-camera position of the latest information of the existing moving body in the camera and the extracted in-camera position of the extracted feature amount, and the feature amount set and extraction of the existing moving body in the camera moving the two feature amounts extracted by using an index to separate aggregated by mobile, detected by the camera at the inter-camera movement path calculation means and Mahalanobis distance calculated feature amount space distance between the feature quantity The path cost using the moving speed / direction information of the body is set, the movement path between the cameras of the moving body is calculated in consideration of the path cost, and the inter- camera moving body matching degree calculating means A monitoring system using a plurality of cameras, characterized in that the degree of matching is calculated by using the closeness of distances between feature quantity sets. The monitoring system according to any one of claims 1 to 3,
You can select the means for the monitor to individually specify the tracking target to be displayed on the display device, and the means for automatically selecting the tracking target according to the number of people passing through the system. A monitoring system using a plurality of cameras, which has a function that can be arbitrarily set for each camera. The monitoring system according to any one of claims 1 to 4,
A monitoring system using a plurality of cameras, wherein a camera image stored in a database and moving object verification tracking information are reproduced and displayed in synchronization with time.

Images