JP7285536B2 - Classifier generation device, target person estimation device and their programs, and target person estimation system - Google Patents

Description

The present invention relates to a classifier generation device and the like for estimating a target person's position by distributing a classifier for classifying the target person to an arbitrary surveillance camera system.

As a technique for searching for a suspicious person or the like using deep learning, for example, the technique disclosed in Patent Document 1 is disclosed. In the technique disclosed in Patent Document 1, an information processing apparatus receives an input of feature information other than a face image indicating a feature of a person to be identified, A first determination unit that determines the degree of matching between the features and the features of the person appearing in the image; and a second determination unit that determines whether or not there is.

JP 2018-142137 A

However, the technique disclosed in Patent Document 1 determines whether a person matches based on feature information other than a face image. A comparison needs to be made. In other words, since the image information of the target person is shared, there is a problem that the personal information is leaked to the monitor or the like. In particular, when searching for a wandering elderly person, a runaway, or the like, it is desirable to investigate the target while keeping personal information as confidential as possible.

The present invention provides a classifier generation device and the like that can estimate the position of a target person to be searched while keeping the personal information of the target person hidden.

A classifier generation apparatus according to the present invention generates a classifier for identifying a target person for each target person based on first image capturing means for capturing an image of the target person and image information captured by the first image capturing means. and a classifier transmitting means for transmitting the classifier so that a control device associated with the second imaging means installed at an arbitrary place can be used.

As described above, in the classifier generating apparatus according to the present invention, an image of a subject is captured, and based on the captured image information, a classifier for specifying the subject is generated for each subject, and an arbitrary classifier is generated. In order to transmit the classifier so that the control device associated with the second imaging means installed in the place can be used, information such as an image that identifies the subject's own personal information is first sent. By distributing only the classifier without disclosing it to three parties, it is possible to specify the person imaged by the second imaging means.

A target person estimation apparatus according to the present invention includes: detection result information receiving means for receiving, as detection result information, a result of using the classifier generated by the classifier generation device in the control device; and the received detection result. The apparatus comprises position specifying means for specifying the position of the subject based on the information, and output control means for outputting the result of the specified position of the subject.

Thus, in the subject estimation apparatus according to the present invention, the detection result information obtained by the classifier generated by the classifier generation apparatus is received, and the position of the subject is specified based on the received detection result information. Since the result of the target person's position is output, it is possible to easily confirm the target person's position imaged by the second image pickup means installed at an arbitrary location and search.

A target person estimation device according to the present invention comprises range specifying means for specifying a range in which the target person exists based on the moving speed of the target person.

As described above, in the target person estimation device according to the present invention, since the range in which the target person exists is specified based on the moving speed of the target person, the search range can be narrowed down to simplify the search work. It has the effect of

In the subject estimating device according to the present invention, the output control means may be the location of the second imaging means installed at an arbitrary location, or the location of the subject estimated from the imaging information of the second imaging means. is displayed on a map, and the detection result information classified by the classifier by the control device associated with each second imaging means is displayed as the location of the second imaging means or the estimated location of the subject are displayed in association with

As described above, in the target person estimation device according to the present invention, the output control means detects the location of the second imaging means installed at an arbitrary location, or the target estimated from the imaging information of the second imaging means. The location of the person is displayed on a map, and the detection result information classified by the classifier by the control device associated with each second imaging means is displayed as the location of the second imaging means or the estimated location of the subject. Since the images are displayed in association with each other, it is possible to instantaneously confirm which second imaging means has imaged what kind of object on the map, and to confirm whether or not there is a person who seems to be the object.

In the target person estimation device according to the present invention, the output control means calculates the detection result information by a ratio indicating the possibility that the person imaged by the second imaging means is the target person, and the ratio is When it is equal to or greater than a predetermined value, it is displayed in a manner different from other result information.

As described above, in the target person estimation device according to the present invention, the output control means calculates the result information at a ratio indicating the possibility that the person imaged by the second image pickup means is the target person, and the ratio is a predetermined value. is greater than or equal to the value of , it is displayed in a manner different from that of other result information, so there is an effect that the position information of a person who seems to be the target person can be easily visually recognized on the map.

1 is a system schematic diagram of a subject estimation system according to a first embodiment; FIG. It is a block diagram which shows the hardware constitutions of the candidate estimation system which concerns on 1st Embodiment. It is a hardware block diagram of computers, such as a classifier generation apparatus, a monitoring control apparatus, and a management apparatus. 4 is a flowchart showing processing of the subject estimation system according to the first embodiment; 1 is a functional block diagram of a classifier generation device according to a first embodiment; FIG. 4 is a flowchart showing processing of the classifier generation device according to the first embodiment; 1 is a functional block diagram of a monitoring control device according to a first embodiment; FIG. 4 is a flowchart showing processing of the monitoring control device according to the first embodiment; 3 is a functional block diagram of a management device according to the first embodiment; FIG. 6 is a flowchart showing processing of the management device according to the first embodiment; 10 is a flowchart showing an example of specific processing from acquisition of classification result information to map display. It is a figure which shows an example of the displayed map.

Embodiments of the present invention will be described below. Also, the same reference numerals are given to the same elements throughout the present embodiment.

(First embodiment of the present invention)
A target person estimation system including a classifier generation device and a target person estimation device according to the present embodiment will be described with reference to FIGS. 1 to 12. FIG. For example, when a caregiver searches for a wanderer, the target person estimation system according to the present embodiment uses a surveillance camera or the like installed at an arbitrary location without disclosing the personal information of the wanderer. By identifying a person, it becomes possible to efficiently search for a wanderer while ensuring privacy. In the following embodiments, searching for a wanderer in a care facility is taken as a specific example, and a person to be searched, such as a wanderer, is defined as a search target, and a caregiver, etc., is defined as a searcher. In addition to searching for wanderers, this technology can also be applied to runaways, kidnappings, missing persons, searching for people (including familiar people and criminals), and the like.

FIG. 1 is a diagram showing a system overview of a subject estimation system according to this embodiment. The target person estimation system 1 includes a classifier generation system 10 installed in a care facility where a search target person 2 usually resides, and a surveillance camera system 20 installed in convenience stores, supermarkets, utility poles on roads, etc. in the city. , a management device 30 that analyzes information obtained from the monitoring camera system 20 and presents it to a searcher, and a cloud server 100 that stores acquired data and the like.

The classifier generation system 10 is installed, for example, in the vicinity of the entrance of a nursing home or the like. A classifier generating device 12 for generating a classifier for classifying the search target person 2 based on image data captured by the camera 11 . The classifier generated by the classifier generation device 12 is created by deep learning, and is a processing unit that detects a specific person and classifies (classifies) the detected person. Weight information is generated according to the hair style, clothes, external shape, face, etc. at the time of leaving. The generated classifier weight information is uploaded to and stored in the cloud server 100 . Details of the configuration and processing of the classifier generation device 12 will be described later.

The monitoring camera system 20 can be applied to the present embodiment based on the monitoring camera system 20 that has already been widely used. The surveillance camera system 20 is installed, for example, in a convenience store, and has a second camera 21 that constantly monitors the inside of the store and the entrance/exit to capture a moving image, and saves image data captured by the second camera 21. and a monitor control device 22 that reproduces the data. In general, one monitor control device 22 manages image information of a plurality of second cameras 21 . For example, a plurality of second cameras 21 installed in a store may be managed by a single monitoring control device 22 installed in the same store, or as shown in FIG. 21 imaging information and the like are collectively managed at a place such as a monitoring center.

In this embodiment, the monitoring control device 22 downloads the weight information of the classifier generated by the classifier generation device 12 from the cloud server 100, and the second camera 21 uses the classifier using the downloaded weight information. Then, an analysis is made as to whether or not the person imaged is the search target person 2 . The classifier weight information generated by the classifier generation device 12 is information obtained by weighting the characteristics of the search target person 2, and does not include information that identifies an individual. In other words, the analysis as to whether or not the search target person 2 is the search target person 2 is performed without disclosing the personal information of the search target person 2 to an unspecified number of people. The analyzed result is uploaded to the cloud server 100 and stored. The configuration and processing of the monitoring control device 22 will be detailed later.

The management device 30 is used when a searcher confirms the position of the search target person 2 and when performing system management. The management device 30 downloads the detection result information detected and classified by the monitoring control device 22 from the cloud server 100, specifies the position of the person to be searched 2, and outputs it. The detection result information includes information that the person imaged by the second camera 21 seems to be the search target person 2, and includes, for example, the probability that the person is the search target person 2. If the probability of being the person to be searched 2 is high (for example, if the probability is higher than a preset threshold), the position information of the second camera 21 that captured the person, or the installation of the second camera 21 It is possible to specify the position of the person to be searched 2 from the orientation and the captured imaging information. When the position is specified, information is displayed to the effect that a person who seems to be the search target person 2 has been imaged at that position on the map. The searcher can determine whether the person imaged by the second camera 21 is the search target person 2 from the information displayed on the map and the image information actually imaged by the second camera 21. becomes possible.

At this time, it is also possible to calculate and specify the range in which the search target person 2 exists in consideration of the moving speed of the search target person 2 . For example, characteristics of the search target person 2, such as when they go out on foot, when they go out by bicycle, when they go out by car, when they frequently use public transportation such as trains, etc. By considering the above, the maximum range in which the search target person 2 exists is estimated by the time elapsed since going out, and the search target person 2 is mainly searched from the imaging information of the second camera 21 installed within that range. becomes possible. In addition, when the position of the person to be searched 2 at a certain time is specified or highly probable, the maximum range in which the person to be searched 2 exists at the next time is defined as the time elapsed from the time to the next time. From the moving speed of the search target person 2, it is possible to successively estimate the maximum range in which the search target person 2 exists. The configuration and processing of the management device 30 will be detailed later.

The management device 30 may have both a function of confirming the position of the person to be searched 2 and a management function of managing the entire system, or each function may be performed by separate management devices. may In addition, the information of the person imaged by the second camera 21 may be transmitted to the management device 30 together with the probability information only when the probability of being the search target person 2 is high. 30. Furthermore, the position of the person to be searched 2 may be determined by calculation on the management device 30 side from the position information of the second camera 21, or the position information calculated on the monitoring control device 22 side that controls the second camera 21. may be transmitted to the management device 30 . Furthermore, the calculation for identifying the position of the person to be searched 2 may be limited to the second camera 21 existing in the maximum range where the person to be searched 2 exists, estimated by the management device 30 .

FIG. 2 is a block diagram showing the hardware configuration of the subject estimation system according to this embodiment. The target person estimation system 1 includes a first camera 11 that captures an image of a person who can later become a search target person 2, and a classifier that can detect and classify the imaged person based on image information captured by the first camera 11. , a second camera 21 as a monitoring camera installed in an arbitrary place such as in the city, and managing information captured by the second camera 21 and driving the second camera 21. a monitoring control device 22 that controls and receives the classifier generated by the classifier generating device 12 to detect and classify the person to be imaged as the search target person 2; A management device 30 is provided for receiving information to specify the position of the person to be searched 2 and displaying the information on a display 31 or the like. In addition, a cloud having a data server that stores imaging information captured by the first camera 11 and the second camera 21, classifier weight information generated by the classifier generation device 12, and/or analysis results of the monitoring control device 22, etc. A server 100 is provided.

Note that the cloud server 100 may be provided within the system as part of the subject estimation system, or may exist as an external server not included in the subject estimation system.

FIG. 3 is a hardware configuration diagram of computers such as the classifier generation device 12, the monitoring control device 22, and the management device 30. As shown in FIG. A computer of each device includes a CPU 41 , a RAM 42 , a ROM 43 , a hard disk (HD) 44 , a communication I/F 45 and an input/output I/F 46 . An operating system and various programs are stored in the ROM 43 and the HD 44 , read out to the RAM 42 as necessary, and the CPU 41 executes each program. The communication I/F 45 is an interface for communicating between apparatuses and with various devices. The input/output I/F 46 is an interface for receiving inputs from camera devices such as the first camera 11 and the second camera 21 and input devices such as a touch panel, keyboard, and mouse, and for outputting information to the display 31 . Note that the hardware configuration described above is merely an example, and can be changed as necessary.

FIG. 4 is a flow chart showing processing of the target person estimation system according to the present embodiment. First, the first camera 11 installed near the entrance of a facility or the like where the person to be imaged who can later become the person to be searched 2 resides takes images of the person to be imaged leaving from a plurality of angles (S1). The classifier generation device 12 detects and classifies the imaged person based on the characteristics of the imaged person (for example, clothes, hairstyle, etc. when leaving) included in the image information captured by the first camera 11. Weight information of a classifier that can be used is generated (S2). When the whereabouts of a person to be imaged becomes unknown and a search is made as a search target person 2, the weight information of the classifier corresponding to the search target person 2 is distributed to the monitoring control devices 22 of the monitoring camera systems 20 installed in various places. (S3).

The monitoring control device 22 that has received the weight information of the classifier applies the person imaged by the second camera 21 associated with the monitoring control device 22 to the classifier to determine whether the person is the search target person 2. is analyzed (S4). The analysis result is transmitted to the cloud server 100 (S5). The management device 30, which has received the analysis result from the cloud server 100, performs estimation processing considering the moving speed of the search target person 2 (S6), and displays the position of the search target person 2 on the map along with the probability that the image was captured. Display (S7). The processes of S4 to S7 are repeated until the corresponding search target person 2 is found.

Next, the configuration and processing of each device will be described in detail. FIG. 5 is a functional block diagram of the classifier generation device according to this embodiment. The classifier generation device 12 includes an imaging information input unit 121 that inputs imaging information obtained by imaging the person to be imaged from a plurality of directions by the first camera 11, and an imaging information storage unit 122 that stores the input imaging information. , a teacher data creation unit 123 that creates teacher data for creating a classifier that identifies a person to be imaged based on imaging information, and a classification that generates weight information for the classifier by performing deep learning using the teacher data. A weight information storage unit 125 that stores the weight information of the generated classifier, and a data transmission unit 126 that uploads the generated weight information of the classifier to the cloud server 100 . Since the information uploaded to the cloud server 100 at this time is the weight information of the classifier, the personal information of the person to be imaged will not be passed on to a third party, and confidentiality can be ensured.

The first camera 11 is arranged so as to capture images of the person to be imaged from a plurality of angles in order to obtain the feature data of the person to be imaged more accurately. It is preferable that the imaging location is the entrance when exiting the building in which the person to be imaged resides. In addition, the imaging process may be always performed using a moving image, or may be performed as a still image or a moving image only when the person to be imaged passes, depending on the sensor that reacts when the person passes. .

The teacher data creation unit 123 cuts out each frame from the imaging information (here, moving image) of the person to be imaged captured by the first camera 11, and if a person is imaged in the frame, the area of the person is extracted. Cut out only to create training data. The classifier generation unit 124 generates a classifier using deep learning from the created teacher data. The data transmission unit 126 uploads the generated classifier weight information to the cloud server 100 .

Note that the imaging information input by the imaging information input 121 as shown in FIG. 5 is stored in the imaging information storage unit 122 . This imaged information is raw data imaged by the first camera 11 and contains a lot of information for identifying an individual such as the face of the person being imaged. Alternatively, the image pickup information may be deleted sequentially according to the capacity, and the latest image pickup information may be always stored. Also, the weight information storage unit 125 does not necessarily have to be provided as a configuration, but it is desirable to store the weight information in the weight information storage unit 125 as backup data just in case.

FIG. 6 is a flow chart showing the processing of the classifier generation device according to this embodiment. First, the imaging information input unit 121 receives input of imaging information captured by the first camera 11 (S1). The teacher data creation unit 122 cuts out a person from the imaging information and creates teacher data (S2). The classifier generation unit 124 generates a classifier using deep learning (S3). The data transmission unit 126 uploads the weight information of the generated classifier to the cloud server 100 (S4), and ends the classifier generation process. It should be noted that a correct label may be given to the teacher data in the classifier generation process. )), face recognition, or the like may be used.

FIG. 7 is a functional block diagram of the monitoring control device according to this embodiment. The monitoring control device 22 acquires the weight information of the corresponding classifier when the person to be imaged corresponding to the classifier uploaded to the cloud server 100 goes missing or becomes the search target person 2. An acquisition unit 221, an imaging information input unit 222 for inputting imaging information captured by the second camera 21, an image processing unit 223 for performing image preprocessing on the input imaging information, and an image processing unit 223. A classification processing unit 224 that detects and classifies the search target person 2 by applying the obtained image to a classifier created based on the weight information acquired by the weight information acquisition unit 221, and uploads the results of the classification processing to the cloud server 100. and a data transmission unit 225 .

The weight information acquisition unit 221 acquires weight information in response to a classifier acquisition instruction from the cloud server 100 . For example, a download link of a classifier on the cloud server 100 is received from the cloud server 100 using two-way communication, a request is made to the received download link, and the classifier is downloaded. The downloaded information is classifier weight information and does not contain personal information. Therefore, personal information that identifies the person to be searched 2 will not be leaked to a third party. Note that the classifier acquisition method is not limited to the above method, and any method that can receive the corresponding weight information from the cloud server 100 may be used.

The image processing unit 223 extracts each frame from the imaging information (here, moving image) captured by the second camera 21, and if a person is captured in the frame, extracts only the area of the person. conduct. Then, in the classification processing unit 224, the image data processed by the image processing unit 223 is classified by a classifier created based on the weight information acquired by the weight information acquisition unit 221, and the images are captured. Then, it is classified whether the person found is the search target person 2 or not. The data transmission unit 225 uploads the classified result to the cloud server 100 .

FIG. 8 is a flowchart showing processing of the monitoring control device according to this embodiment. First, the weight information acquisition unit 221 receives a classifier weight information acquisition instruction from the cloud server 100 (S1). Weight information is acquired from the cloud server 100 according to the received acquisition instruction (S2). The imaging information input unit 222 receives input of imaging information captured by the second camera 21 (S3). The image processing unit 223 cuts out a person from the imaging information (S4). The classification processing unit 224 performs classification processing using a classifier on the person in the image cut out in S4 (S5). The classification process referred to here classifies whether or not the person in the image cut out in S4 is the search target person 2 using a classifier created based on the weight information acquired from the cloud server 100, and calculates the matching degree. . The data transmission unit 225 uploads the classification result to the cloud server 100 (S6), and the processes of S3 to S6 are repeated until the search target person 2 is searched.

In the classification process of S5, the probability that the person in the image cut out in S4 is the search target person 2 may be obtained as the classification result using a classifier created based on the weight information acquired from the cloud server 100. In addition, information on the position where the second camera 21 is installed, or information on the position of the search target person 2 estimated based on the installation orientation of the second camera 21 and the captured image information is also stored in the cloud server. You can upload to 100. Furthermore, the current maximum real-time search target person 2 range is calculated according to the time when the person was imaged by the second camera 21 and the movement state at that time (for example, walking, jogging, bicycling, etc.), and classified. You may upload to the cloud server 100 with a result.

FIG. 9 is a functional block diagram of the management device according to this embodiment. The management device 30 includes a classification result acquisition unit 301 that acquires the classification results uploaded to the cloud server 100, and based on the exit time and moving speed of the search target person 2, or based on the acquired classification result. A target range specifying unit 302 for estimating and calculating the range in which 2 exists, and a map that shows the probability that the search target person 2 exists is generated based on the information of the classification result and the maximum range in which the search target person 2 exists. A map generation unit 303 and a display control unit 304 for displaying the generated map on the display 31 are provided.

Note that the target range identification unit 302 may obtain the maximum range estimated from the time when the search target person 2 left the building and the moving speed, or the search target person 2 included in any of the classification result information. is imaged by the second camera 21, the time and moving speed, and the position where the second camera 21 is installed, the range in which the search target person 2 can currently exist can be estimated and obtained. good.

FIG. 10 is a flowchart showing processing of the management device according to this embodiment. First, the classification result acquisition unit 301 receives an instruction to acquire classification result information from the cloud server 100 (S1). Classification result information is acquired from the cloud server 100 according to the received acquisition instruction (S2). The target range identification unit 302 estimates the maximum range in which the search target person 2 exists (S3). The map generator 303 generates a map based on the acquired classification result information and the estimated maximum range (S4). The display control unit 304 displays the generated map on the display 31 (S5). The processes of S1 to S5 are repeated until the search target person 2 is found.

Here, an example of map generation processing will be described in detail. FIG. 11 is a flowchart showing an example of specific processing from acquisition of classification result information to map display, and FIG. 12 is a diagram showing an example of the displayed map. First, target classification result information is extracted from the time when the search target person 2 goes out (S1). That is, since the classification result information before the search target person 2 leaves the house is not helpful, only the classification result information from the time when the search target person 2 leaves until the current time is processed. The camera ID for individually identifying the second camera 21 included in the extracted classification result information, and the pre-registered camera ID and position information (for example, latitude and longitude information) of the second camera 21 are stored. The obtained second camera information is linked (S2). The position information of the camera ID included in the classification result information is acquired from the second camera information, and a marker is placed on the map (S3). The matching degree included in the classification result information is calculated, and if the matching degree is equal to or greater than a predetermined value, the marker is set to be highlighted (S4). Based on the position information of the second camera 21 corresponding to the classification result information and the maximum range estimated by the target range identification unit 302, if the second camera 21 is within the maximum range, A line is drawn on the map to the position of the second camera 21 (S5). Display the generated map (S6). Note that the position where the marker is placed is not limited to the installation position of the second camera 21, and may be, for example, the position of the search target estimated based on the installation orientation of the second camera 21 and the imaged information. .

The searcher refers to the classification result information and the imaging information of the second camera 21 in the map as shown in FIG. 2 is easier to identify. In FIG. 12 , with the exit point (black circle) as the starting point, the degree of matching with the search target person 2 is calculated and displayed as the classification result of the person imaged by each second camera 21 . In the case of FIG. 12, persons are imaged by cameras (1), (2), (4) and (6), and the probability that each person is the search target person 2 is displayed. Among them, the person captured by camera (2) has a probability of 95% being search target person 2, and the line (chain line) connecting camera (2) and the starting point and the marker (star mark) are emphasized. is displayed. The searcher can confirm the display screen as shown in FIG. 12 and take action to secure the person imaged by the camera (2) with the highest priority.

As described above, by checking the map in FIG. 12, it is possible to instantly visually recognize the degree of matching between the person imaged by the camera and the person to be searched 2. In addition to this, a display function is also provided. can be added. For example, by analyzing the moving image captured by the camera (2), it is possible to calculate in what direction and at what speed the captured person moved with respect to the camera (2). . That is, it is possible to estimate the position of the person at the present time, and in the case of FIG. The probability is calculated and displayed as 70%.

In addition, it is possible to predict the moving route of the search target person 2 (indicated by the dotted line in FIG. 12) from the proportion of the person imaged by each camera being the search target person 2 and the imaging time. It is also possible to calculate a predicted future travel route (indicated by a thick line in FIG. 12) and an estimated current location from the travel route and the analysis of the moving image described above.

Furthermore, the display mode of markers and lines in each camera may be changed according to the elapsed time. For example, image information obtained by imaging a person who has a 100% probability of being search target person 2 five hours ago and image information obtained by imaging a person who has an 80% probability of being search target person 2 one minute ago. Considering that the search target person 2 has moved, it can be determined that the image information captured in the latter is more credible as information indicating the current position of the search target person 2 . In such a case, for example, even if the probability is 100%, the marker or line of this camera may not be highlighted, but rather the marker or line with the probability of 80% may be highlighted. .

As described above, in the target person estimation system according to the present embodiment, a classifier for specifying the search target person 2 is generated for each search target person 2, and the second camera 21 installed at an arbitrary location Since the generated classifier is transmitted to the associated monitor control device 22, only the classifier is sent without disclosing information such as an image that identifies the personal information of the search target person 2 to a third party. By distributing it, the person photographed by the second camera 21 can be specified.

Further, it receives classification result information by the classifier generated by the classifier generation device 12, identifies the position of the search target person 2 based on the received classification result information, and obtains the result of the identified position of the search target person 2. By outputting, it becomes possible to easily confirm the position of the person to be searched 2 imaged by the second camera 21 installed at an arbitrary location, and to search.

Furthermore, by specifying the range in which the search target person 2 exists based on the moving speed of the search target person 2, the search range can be narrowed down and the search work can be simplified.

Furthermore, the display control unit 304 displays the location of the second camera 21 installed at an arbitrary location or the location of the search target person 2 estimated from the imaging information of the second camera 21 on the map. , the classification result information classified by the classifier by the monitoring control device 22 associated with each second camera 21 is displayed in association with the location of the second camera 21 or the estimated location of the subject, It is possible to instantly confirm which second camera 21 has captured what kind of imaging target on the map, and to confirm the presence or absence of a person who seems to be the search target person 2 .

Furthermore, the display control unit 304 calculates the classification result information with a ratio indicating the possibility that the person imaged by the second camera 21 is the search target person 2, and if the ratio is equal to or greater than a predetermined value, the other By displaying in a manner different from the classification result information of 1, it is possible to easily visually recognize the position information of a person who seems to be the search target person 2 on the map.

1 Target person estimation system 2 Search target person 10 Classifier generation system 11 First camera 12 Classifier generation device 20 Surveillance camera system 21 Second camera 22 Monitoring control device 30 Management device 31 Display 41 CPU
42 RAM
43 ROMs
44 hard disk 45 communication I/F
46 input/output I/F
100 cloud server 121 image information input unit 122 image information storage unit 123 teacher data creation unit 124 classifier generation unit 125 weight information storage unit 126 data transmission unit 221 weight information acquisition unit 222 image information input unit 223 image processing unit 224 classification processing unit 225 data transmission unit 301 classification result acquisition unit 302 target range identification unit 303 map generation unit 304 display control unit

Claims (8)

one image capturing means installed near the entrance of the facility where the subject is staying and capturing the features including the hairstyle and clothes of the subject when leaving;
a classifier generating means for performing deep learning based on the imaging information captured by the one imaging means and generating a classifier for classifying the subject for each subject;
The person staying in the facility captured the image of the control device associated with the other imaging means installed at a plurality of different locations outside the facility by the other imaging means associated with the control device. Only the weight information of the classifier obtained as a result of deep learning the characteristics of the subject is selected so that it can be used as the classifier that can classify whether the subject is a subject or not, and the subject individual is specified. a classifier transmitting means for transmitting to the control device outside the facility after excluding the characteristic information, the information regarding the individual privacy of the subject, and the information indicating the appearance of the subject imaged by the one imaging means; A classifier generation device comprising: A detection result information receiving means for receiving, as detection result information, a result of using the classifier generated by the classifier generation device according to claim 1 in the control device;
Position specifying means for specifying the position of the subject based on the received detection result information;
and output control means for outputting a position result of the identified target person. In the subject estimation device according to claim 2,
A subject estimating device comprising range identifying means for identifying a range in which the subject exists based on the moving speed of the subject. In the subject estimation device according to claim 2 or 3,
The output control means is
displaying on a map the locations of the other imaging means respectively installed at a plurality of different locations outside the facility , or the location of the subject estimated from the imaging information of the other imaging means; Subject estimation for displaying the detection result information classified by the classifier by the control device associated with each other imaging means in association with the location of the other imaging means or the estimated location of the subject. Device. In the subject estimation device according to claim 4,
The output control means is
Based on the detection result information, a ratio indicating the possibility that the person imaged by the other imaging means is the target person is calculated, and when the ratio is a predetermined value or more, it is different from the other detection result information A subject estimation device displayed in a form. one image capturing means installed near the entrance of the facility where the subject is staying and capturing the features including the hairstyle and clothes of the subject when leaving;
a classifier generating means for performing deep learning based on the imaging information captured by the one imaging means and generating a classifier for classifying the subject for each subject;
The person staying in the facility captured the image of the control device associated with the other imaging means installed at a plurality of different locations outside the facility by the other imaging means associated with the control device. Only the weight information of the classifier obtained as a result of deep learning the characteristics of the subject is selected so that it can be used as the classifier that can classify whether the subject is a subject or not, and the subject individual is specified. a classifier transmitting means for transmitting to the control device outside the facility after excluding the characteristic information, the information regarding the individual privacy of the subject, and the information indicating the appearance of the subject imaged by the one imaging means;
detection result information receiving means for receiving detection result information based on the transmitted weight information of the classifier;
computing means for computing the position of the subject based on the received detection result information;
and output control means for outputting the result of calculation. Deep learning is performed based on image information captured by one imaging means installed near the entrance of the facility where the subject is staying and capturing features including the hairstyle and clothing of the subject when leaving the subject, and performing deep learning. classifier generation means for generating a classifier for classifying the subject for each subject;
The person staying in the facility captured the image of the control device associated with the other imaging means installed at a plurality of different locations outside the facility by the other imaging means associated with the control device. Only the weight information of the classifier obtained as a result of deep learning the characteristics of the subject is selected so that it can be used as the classifier that can classify whether the subject is a subject or not, and the subject individual is specified. a computer as a classifier transmission means for transmitting to the control device outside the facility after excluding the characteristic information, the information regarding the individual privacy of the subject, and the information indicating the appearance of the subject imaged by the one imaging means A classifier generation program characterized by functioning.
Detection result information receiving means for receiving, as detection result information, a result of the use of the classifier generated by the classifier generation program according to claim 7 operating a computer in the control device;
Position specifying means for specifying the position of the subject based on the received detection result information;
A target person estimation program that causes a computer to function as output control means for outputting the position result of the specified target person.

Images