Method, device and system for detecting abnormal pet behavior in public places

Abstract

The application discloses a detection method, device and system for abnormal pet behavior in public places. The method comprises the following steps: acquiring a real-time video stream, identifying the posture relationship between each pet and each person included in each video frame of the real-time video stream; extracting a target pet included in the image recognition result, at least one associated person meeting a distance threshold condition with the target pet, and a leash detection result matched with the target pet; if it is determined that the leash detection result is that there is no leash, acquiring the posture relationship between the target pet and each associated person in each identified posture relationship matched with a target time interval; and if it is determined that the posture relationship between the target pet and all associated persons is an independent motion relationship, generating abnormal behavior warning information for the target pet. By using the above technical scheme, the influence of factors such as leash shielding, too thin or poor light on the detection result can be avoided, and the detection accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video detection, in particular to a method, device and system for detecting abnormal pet behavior in public places. BACKGROUND

[0002] In order to improve the pet management efficiency in public places, at present, the real-time monitoring and automatic identification are usually adopted to supervise the abnormal pet behavior in public places.

[0003] When there is no connection relationship between the pet person and the pet, it can be regarded as abnormal pet behavior. The existing abnormal pet behavior identification method can only identify the clear and explicit traction rope, and in the case of light cross, thin traction rope or blocked traction rope, it is likely to produce false detection results. SUMMARY

[0004] The present application provides a method, device and system for detecting abnormal pet behavior in public places, which can avoid the influence of factors such as traction rope blocking, too thin or poor light on the detection results, and improve the detection accuracy.

[0005] According to an aspect of the present application, a method for detecting abnormal pet behavior in public places is provided, comprising:

[0006] Obtaining the real-time video stream sent by the monitoring camera in the target monitoring area, identifying the posture relationship between each pet and each person included in each video frame of the real-time video stream;

[0007] When the image recognition result matching the target time interval sent by the processor on the monitoring camera side is obtained, the target pet included in the image recognition result, at least one associated person meeting the distance threshold condition with the target pet and the traction rope detection result matching the target pet are extracted;

[0008] If it is determined that the traction rope detection result is that there is no traction rope, the posture relationship between the target pet and each associated person is obtained in the identified posture relationship matching the target time interval;

[0009] If it is determined that the posture relationship between the target pet and all associated persons is independent motion relationship, the abnormal behavior warning information for the target pet is generated.

[0010] Optionally, identifying the posture relationship between each pet and each person included in each video frame of the real-time video stream comprises:

[0011] Inputting the real-time video stream into a key point detection model, and obtaining the key point detection information of each pet and each person in each video frame output by the key point detection model;

[0012] input the key point detection information of each video frame into the continuous frame posture recognition model, and obtain posture relationships between each pet and each person included in each video frame output by the continuous frame posture recognition model.

[0013] Optionally, after extracting the target pet included in the image recognition result, at least one associated person meeting a distance threshold condition with the target pet, and a leash detection result matched with the target pet, the method further comprises:

[0014] If it is determined that the leash detection result is that there is a leash, it is determined that the target monitoring area is a normal pet behavior, and the detection of the target pet is ended.

[0015] Optionally, the posture relationship comprises a cooperative motion relationship and an independent motion relationship.

[0016] After if it is determined that the leash detection result is that there is no leash, and in the identified posture relationships matched with the target time interval, the posture relationship between the target pet and each associated person is obtained, the method further comprises:

[0017] If it is determined that the posture relationship between the target pet and at least one associated person is a cooperative motion relationship, it is determined that the target monitoring area is a normal pet behavior, and the detection of the target pet is ended.

[0018] Optionally, the continuous frame posture recognition model is an ST-GCN model based on a graph convolutional neural network.

[0019] According to another aspect of the present application, a device for detecting abnormal pet behavior in a public place is provided, comprising:

[0020] A posture relationship recognition module is configured to obtain a real-time video stream sent by a monitoring camera in a target monitoring area, and recognize posture relationships between each pet and each person included in each video frame of the real-time video stream.

[0021] An image recognition result processing module is configured to, when an image recognition result matched with a target time interval sent by a processor on the monitoring camera side is obtained, extract a target pet included in the image recognition result, at least one associated person meeting a distance threshold condition with the target pet, and a leash detection result matched with the target pet.

[0022] A posture relationship determination module is configured to, if it is determined that the leash detection result is that there is no leash, obtain a posture relationship between the target pet and each associated person in the identified posture relationships matched with the target time interval.

[0023] The abnormal behavior early warning module is configured to generate abnormal behavior early warning information for the target pet if it is determined that the posture relationship between the target pet and all the associated persons is an independent motion relationship.

[0024] According to another aspect of the present application, there is provided a system for detecting abnormal pet behavior in a public place, comprising a monitoring camera, an edge box terminal processor, and a monitoring camera side processor; wherein,

[0025] The monitoring camera is configured to acquire real-time video streams in a target monitoring area and send the real-time video streams to the edge box terminal processor and the monitoring camera side processor, respectively;

[0026] The edge box terminal processor is configured to execute the method for detecting abnormal pet behavior in a public place according to any embodiment of the present application;

[0027] The monitoring camera side processor is configured to generate image recognition results for each video frame of the real-time video streams according to the real-time video streams sent by the monitoring camera in the target monitoring area, and when a target pet is determined in the target monitoring area according to the image recognition results, filter image recognition results matching a target time interval and send the image recognition results to the edge box terminal processor.

[0028] Optionally, the monitoring camera side processor is specifically configured to:

[0029] identify the video frames of the real-time video streams using a biological recognition model, detect whether there is at least one pet in the target monitoring area in real time, and when it is detected that there is at least one pet in the target monitoring area, detect whether there is at least one person in the target monitoring area in a specified time interval;

[0030] If there is, a target pet is determined, and a distance estimation model is used to estimate the distances between each person and the target pet in the target monitoring area, and each person with a distance from the target pet less than a preset distance threshold is determined as an associated person of the target pet;

[0031] If there is not, it is determined that there is abnormal pet behavior in the target monitoring area.

[0032] Optionally, the monitoring camera side processor is further specifically configured to:

[0033] After the target pet and at least one associated person of the target pet in the target monitoring area are determined, a region of interest of the target pet and a region of interest of each associated person are acquired, and a leash detection model is used to detect whether there is a leash matching the target pet;

[0034] The region of interest of the target pet, the region of interest of each associated person, and the detection results of the leash that matches the target pet are used as image recognition results and sent to the edge box processor.

[0035] Optional, the detection system for abnormal pet behavior in public places also includes a monitoring terminal alarm device and a back-end alarm device;

[0036] The monitoring alarm device is used to issue a violation alarm in real time when abnormal pet behavior is detected in the target monitoring area, and to send video frames related to the abnormal pet behavior to the background alarm device.

[0037] The background alarm device is used to determine the relevant personnel information and monitoring area information when it receives video frames related to abnormal pet behavior sent by the monitoring terminal alarm device, and to issue an alarm reminder to the supervisor based on the relevant personnel information and monitoring area information.

[0038] The technical solution of this invention acquires real-time video streams within a target monitoring area, obtains the posture relationships between pets and people in each video frame based on the real-time video stream, and determines pet behavior within the target monitoring area based on image recognition results and posture relationships sent by the processor on the monitoring camera side. This method can avoid the impact of factors such as leash obstruction, excessively thin leash, or poor lighting on the detection results during the detection of abnormal pet behavior, thereby improving detection accuracy, reducing false positives, and better encouraging people to abide by established pet behavior rules in public places.

[0039] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of a method for detecting abnormal pet behavior in public places according to Embodiment 1 of the present invention;

[0042] Figure 2 This is a flowchart of another method for detecting abnormal pet behavior in public places according to Embodiment 2 of the present invention;

[0043] Figure 3 is a structural schematic diagram of a detection device for abnormal pet behavior in a public place according to Embodiment Three of the present application;

[0044] Figure 4 is a structural schematic diagram of a detection system for abnormal pet behavior in a public place according to Embodiment Four of the present application;

[0045] Figure 5 is a structural schematic diagram of another detection system for abnormal pet behavior in a public place according to Embodiment Four of the present application. DETAILED DESCRIPTION

[0046] In order to make the personnel in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0047] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0048] Embodiment One

[0049] Figure 1 A flowchart of a detection method for abnormal pet behavior in a public place according to Embodiment One of the present application is provided, and this embodiment can be applicable to the case of detecting whether a pet wears a leash in a public place. The method can be executed by a detection device for abnormal pet behavior in a public place, which can be realized in the form of hardware and / or software, and can generally be configured in a processor with image processing function in a monitoring camera. As shown in the figure, the method comprises: Figure 1

[0050] ​S110, acquire a real-time video stream sent by a monitoring camera in a target monitoring area, and identify a posture relationship between each pet and each person included in each video frame of the real-time video stream.

[0051] Optionally, the target monitoring area can refer to an area covered by a shooting range of the monitoring camera, and the real-time video stream is a video stream generated by the monitoring camera shooting the target monitoring area.

[0052] Optionally, the key points of each pet and each person in the video frame of the real-time video stream can be extracted and labeled by a key point detection model, and the labeled key points of each pet and each person can be sent to a continuous frame posture recognition model, and the continuous frame posture recognition model can generate a posture relationship between each pet and each person in the target monitoring area according to the key points of each pet and each person.

[0053] Optionally, the posture relationship between the pet and the person can include a cooperative motion relationship and an independent motion relationship. The cooperative motion relationship can be equivalent to the cooperative motion between the person and the pet through the leash, that is, the pet-leash-walking relationship between the person and the pet. The independent motion relationship can be understood as the relatively independent motion between the pet and the person.

[0054] Optionally, after the labeled key points of each pet and each person are sent to the continuous frame posture recognition model, the continuous frame posture recognition model can directly output the posture relationship between the pet and the person.

[0055] S120, when the image recognition result matching the target time interval sent by the processor on the monitoring camera side is acquired, extract the target pet, at least one associated person satisfying the distance threshold condition with the target pet, and the leash detection result matching the target pet included in the image recognition result.

[0056] Optionally, the processor on the monitoring camera side can acquire the image recognition result of each video frame according to the real-time video stream sent by the monitoring camera.

[0057] Optionally, when it is determined according to the real-time video stream that there is at least one pet in the target monitoring area, a specified pet can be selected as the target pet according to a preset pet determination rule. If there is one pet in the target monitoring area, the pet is the target pet. If there are multiple pets in the target monitoring area, each pet can be taken as the target pet, and the detection method of the abnormal pet behavior described in the present application can be executed.

[0058] Optionally, the image recognition result can include a ROI (region of interest) of the target pet, a ROI of each associated person in the video frame, and an identification result of whether the leash is identified in the video frame.

[0059] Optionally, the associated person of the target pet can be understood as a person in the video frame and the distance between the person and the target pet is less than a preset distance threshold.

[0060] Optionally, the target time interval can be a specified time period after the target pet is identified, and the image recognition result of the video frame in the time period is integrated, that is, the image recognition result matched with the target time interval is obtained.

[0061] S130, if it is determined that the pulling rope detection result is that there is no pulling rope, in the identified posture relationship of each of the target time interval, the posture relationship between the target pet and each associated person is obtained.

[0062] Optionally, if it is determined according to the image recognition result that there is no pulling rope in the target monitoring area, it can be that the target pet does not actually wear a pulling rope, or it can be that the pulling rope detection result is wrong due to video shooting reasons, light reasons, or the pulling rope itself is too thin, shielding, etc. In order to avoid misjudgment caused by pulling rope detection error, the ROI of the target pet and the ROI of each associated person are used to mark the target pet and the associated person in the posture relationship of each pet and each person identified by the continuous frame posture recognition model, and then it is determined whether the target pet and at least one associated person are in a cooperative motion relationship.

[0063] S140, if it is determined that the posture relationship between the target pet and all associated persons is an independent motion relationship, an abnormal behavior warning information for the target pet is generated.

[0064] Further, if the target pet does not have a pulling rope, but the target pet and at least one associated person are in a cooperative motion relationship, it can be determined that the pulling rope detection is wrong, and the pet is a normal pet behavior, if the target pet does not have a pulling rope, and the target pet and each associated person are in an independent motion relationship, it can be determined that the target pet is an abnormal pet behavior.

[0065] The technical scheme of the embodiment of the application, by obtaining the real-time video stream in the target monitoring area, obtaining the posture relationship between each pet and each person in each video frame according to the real-time video stream, and judging the pet behavior in the target monitoring area according to the image recognition result sent by the processor on the side of the monitoring camera and the posture relationship, can avoid the influence of factors such as pulling rope shielding, pulling rope being too thin, or poor light on the detection result in the detection process of abnormal pet behavior, can improve the detection accuracy, and can reduce the detection misjudgment, and can better supervise the personnel in public places to comply with the established pet behavior rules.

[0066] Embodiment two

[0067] Figure 2A flowchart of a detection method of an abnormal pet behavior in a public place is provided for the second embodiment of the present application. The second embodiment is based on the above-mentioned embodiments and specifically illustrates the detection method of the abnormal pet behavior in the public place. As shown in FIG. 17, the method comprises the following steps. Figure 2

[0068] In S210, real-time video streams sent by monitoring cameras in a target monitoring area are acquired, the real-time video streams are input into a key point detection model, and key point detection information of each pet and each person in each video frame output by the key point detection model is acquired.

[0069] Optionally, in order to better identify the motion state of the pet and the person, 17 key points can be set for the pet and the person. The key points of the pet can include a nose, left eyes, right eyes, a neck, a tail root, a left shoulder, a left elbow, a left front paw, a right shoulder, a right elbow, a right front paw, a left hip, a left knee, a left rear paw, a right hip, a right knee, and a right rear paw. The key points of the person can include a nose, left eyes, right eyes, left ears, right ears, a left shoulder, a right shoulder, a left elbow, a right elbow, a left wrist, a right wrist, a left crotch, a right crotch, a left knee, a right knee, a left ankle, and a right ankle.

[0070] In S220, the key point detection information of each video frame is input into a continuous frame posture recognition model, and posture relationships between each pet and each person included in each video frame output by the continuous frame posture recognition model are acquired.

[0071] Optionally, the continuous frame posture recognition model is an action classification algorithm ST-GCN model based on a graph convolutional neural network.

[0072] Optionally, in the ST-GCN model, a predefined coefficient matrix can be used to fuse features of different joints belonging to the same target. The coefficient matrix can be derived from a predefined joint topology, and each element in the coefficient matrix can be obtained by reweighting a set of learnable weights.

[0073] Optionally, the continuous frame posture recognition model can use a multi-branch time TCN (Temporal Convolutional Network) to replace a single-branch design. The multi-branch TCN is composed of six branches: one “1x1” Conv branch, one Max Pooling branch, and four time-domain 1D Conv branches with a kernel size of 3 and a dilation range of 1 to 4. First, the features are transformed by “1x1” convolution, and they are divided into 6 groups, each with equal channel width. Then, each feature group is processed using a single branch. The six outputs are connected together and processed by another “1x1” Conv to form the output of the multi-branch TCN.

[0074] ​The advantages of this design are that by improving the ST-GCN model, the time modeling capability can be effectively improved, while saving computational costs and parameters, and reducing the channel width of each branch.

[0075] S230. When the image recognition result matching the target time interval is obtained from the processor on the monitoring camera side, the target pet, at least one associated person who meets the distance threshold condition with the target pet, and the leash detection result matching the target pet are extracted from the image recognition result.

[0076] S240. If the leash detection result indicates that there is no leash, then among the identified posture relationships that match the target time interval, obtain the posture relationship between the target pet and each associated person; execute step S260 or step S270.

[0077] S250. If the leash detection result indicates the presence of a leash, then the behavior of the pet within the target monitoring area is determined to be normal, and the monitoring of the target pet is terminated.

[0078] It is understood that steps S240 and S250 are two parallel steps, and the execution of step S240 or step S250 can be determined based on the image recognition result obtained in S230.

[0079] S260. If it is determined that the posture relationship between the target pet and all associated personnel is an independent movement relationship, then generate an abnormal behavior warning message for the target pet.

[0080] S270. If it is determined that the posture relationship between the target pet and at least one associated person is a cooperative movement relationship, then the behavior of the target pet within the monitoring area is determined to be normal, and the detection of the target pet is terminated.

[0081] Optionally, steps S260 and S270 are two parallel steps. If step S240 is executed, step S260 or step S270 can be executed based on the posture relationship between the target pet and each associated person obtained in step S240.

[0082] The technical solution of this invention acquires real-time video streams within a target monitoring area, obtains the posture relationships between pets and people in each video frame based on the real-time video stream, and determines pet behavior within the target monitoring area based on image recognition results and posture relationships sent by the processor on the monitoring camera side. This method can avoid the impact of factors such as leash obstruction, excessively thin leash, or poor lighting on the detection results during the detection of abnormal pet behavior, thereby improving detection accuracy, reducing false positives, and better encouraging people to abide by established pet behavior rules in public places.

[0083] Embodiment three

[0084] Figure 3 A structural schematic diagram of an abnormal pet behavior detection device in a public place provided by Embodiment three of the present application is shown in FIG. 3. As shown in the figure, the device includes a posture relationship identification module 310, an image recognition result processing module 320, a posture relationship determination module 330, and an abnormal behavior early warning module 340. Figure 3

[0085] The posture relationship identification module 310 is configured to acquire real-time video streams sent by a monitoring camera in a target monitoring area, and identify posture relationships between each pet and each person included in each video frame of the real-time video streams.

[0086] The image recognition result processing module 320 is configured to, when acquiring an image recognition result sent by a processor on the monitoring camera side and matching a target time interval, extract a target pet, at least one associated person meeting a distance threshold condition with the target pet, and a leash detection result matching the target pet included in the image recognition result.

[0087] The posture relationship determination module 330 is configured to, if it is determined that the leash detection result is that there is no leash, acquire posture relationships between the target pet and each associated person in each of the identified posture relationships matching the target time interval.

[0088] The abnormal behavior early warning module 340 is configured to, if it is determined that the posture relationship between the target pet and all associated persons is an independent motion relationship, generate abnormal behavior early warning information for the target pet.

[0089] The technical solution of the embodiment of the present application acquires real-time video streams in a target monitoring area, acquires posture relationships between each pet and each person included in each video frame according to the real-time video streams, and judges the manner of pet behavior in the target monitoring area according to the image recognition result sent by the processor on the monitoring camera side and the posture relationship, which can avoid factors such as leash obstruction, too thin, or poor light affecting the detection result in the detection process of abnormal pet behavior, can improve the detection accuracy, can further reduce the situation of detection misjudgment, and can better supervise personnel to comply with the established pet behavior rules in public places.

[0090] On the basis of each of the above embodiments, the posture relationship identification module 310 can be specifically configured to:

[0091] The posture relationship between each pet and each person included in each video frame includes:

[0092] ​input the real-time video stream into the key point detection model, and obtain key point detection information of each pet and each person in each video frame output by the key point detection model;

[0093] input the key point detection information of each video frame into the continuous frame posture recognition model, and obtain a posture relationship between each pet and each person included in each video frame output by the continuous frame posture recognition model.

[0094] On the basis of each of the above embodiments, a traction rope processing module can also be included, specifically configured to:

[0095] If it is determined that the traction rope detection result is that the traction rope exists, it is determined that the target monitoring area is a normal pet behavior, and the detection of the target pet is ended.

[0096] On the basis of each of the above embodiments, a cooperative motion relationship processing module can also be included, specifically configured to:

[0097] If it is determined that the posture relationship between the target pet and the at least one associated person is a cooperative motion relationship, it is determined that the target monitoring area is a normal pet behavior, and the detection of the target pet is ended.

[0098] On the basis of each of the above embodiments, the continuous frame posture recognition model is an action classification algorithm ST-GCN model based on a graph convolutional neural network.

[0099] The detection device for abnormal pet behavior in a public place provided by the embodiments of the present application can execute the detection method for abnormal pet behavior in a public place provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0100] Embodiment Four

[0101] Figure 4 A structural schematic diagram of a detection system for abnormal pet behavior in a public place provided by Embodiment Four of the present application. As shown in Figure 4 The detection system for abnormal pet behavior in a public place includes a monitoring camera 410, an edge box end processor 420, and a monitoring camera side processor 430.

[0102] The monitoring camera 410 is used to obtain a real-time video stream in a target monitoring area, and send the real-time video stream to the edge box end processor 420 and the monitoring camera side processor 430, respectively.

[0103] The edge box end processor 420 is used to execute the detection method for abnormal pet behavior in a public place described in any of the embodiments of the present application.

[0104] The monitoring camera side processor 430 is configured to generate image recognition results of each video frame of the real-time video stream according to the real-time video stream sent by the monitoring camera 410 in the target monitoring area, and when a target pet is determined in the target monitoring area according to the image recognition results, screen image recognition results matching the target time interval and send the image recognition results to the edge box end processor 420.

[0105] Optionally, the monitoring camera side processor 430 can be configured inside the monitoring camera 410, and the edge box end processor 420 can be configured in the edge box near the monitoring camera 410. Because the configuration environment required by the processor computing power is different, the edge box end processor 420 can be configured with higher computing power, and the computing power of the monitoring camera side processor 430 is relatively lower than that of the edge box end processor 420. The monitoring camera side processor 430 and the edge box end processor 420 respectively perform different operation contents, for example, for the ROI extraction or the leash detection of a relatively mature model, the monitoring camera side processor 430 can be used for processing, and for the pet and personnel key point extraction and the posture relationship detection, the edge box end processor 420 can be used for execution, so as to improve the operation speed and reduce the detection time of the abnormal pet behavior.

[0106] Optionally, the monitoring camera side processor 430 can be specifically used for:

[0107] using a biological recognition model to recognize the video frames of the real-time video stream, detecting whether there is at least one pet in the target monitoring area in real time, and when it is detected that there is at least one pet in the target monitoring area, detecting whether there is at least one person in the target monitoring area in a specified time interval;

[0108] If yes, a target pet is determined, and a distance estimation model is used to estimate the distances between each person and the target pet in the target monitoring area, and each person with a distance less than a preset distance threshold from the target pet is regarded as an associated person of the target pet.

[0109] If no, it is determined that there is an abnormal pet behavior in the target monitoring area.

[0110] Optionally, the monitoring camera side processor 430 can be further specifically used for:

[0111] After the target pet and at least one associated person of the target pet in the target monitoring area are determined, a region of interest of the target pet and a region of interest of each associated person are obtained, and a leash detection model is used to detect whether there is a leash matching the target pet.

[0112] The target pet's region of interest, the region of interest of each associated person, and the leash matching result of the target pet are detected as image recognition results and sent to the edge box processor 420.

[0113] It can be understood that the monitoring camera side processor 430 is only used to generate image processing results of video frames, and does not generate the final detection result of the abnormal pet behavior, and the final detection result of the abnormal pet behavior is obtained by the edge box processor 420 according to the detected pet and person posture relationship and the image processing result sent by the monitoring camera side processor 430.

[0114] Figure 5 Another structure schematic diagram of the detection system of the abnormal pet behavior in the public place provided by the embodiment of the present application is provided. Figure 5 As shown in the figure, the detection system of the abnormal pet behavior in the public place further comprises a monitoring end alarm device 440 and a background alarm device 450.

[0115] The monitoring end alarm device 440 is used to send a violation alarm in real time when the abnormal pet behavior in the target monitoring area is detected, and send the video frame related to the abnormal pet behavior to the background alarm device 450.

[0116] The background alarm device 450 is used to determine the related person information and the monitoring area information of the abnormal pet behavior when receiving the video frame related to the abnormal pet behavior sent by the monitoring end alarm device 440, and send an alarm reminder to the supervisor according to the related person information and the monitoring area information of the abnormal pet behavior.

[0117] Optionally, the monitoring end alarm device 440 can include an audible and visual alarm device to remind the associated person of the target pet, and the background alarm device 450 can include a graphic alarm device and an audible and visual alarm device, the audible and visual alarm device is used to remind the supervisor to check the graphic alarm device, and the graphic alarm device can be used to display the abnormal pet behavior information in the target monitoring area.

[0118] The technical scheme of the embodiment of the present application can avoid the influence of factors such as a leash blocking, being too thin or poor light on the detection result in the detection process of the abnormal pet behavior, can improve the detection accuracy and detection efficiency, can reduce the detection misjudgment, and can timely notify the pet walking personnel and the supervision personnel, can better supervise the pet walking personnel to comply with the established pet behavior rules in public places.

Claims

1. A method for detecting abnormal pet behavior in public places, characterized in that, include: Acquire real-time video streams sent by surveillance cameras within the target monitoring area, and identify the posture relationships between each pet and each person included in each video frame of the real-time video stream; When the image recognition result matching the target time interval is obtained from the processor on the monitoring camera side, the target pet, at least one associated person who meets the distance threshold condition with the target pet, and the leash detection result matching the target pet are extracted from the image recognition result. If the leash detection result indicates that the leash is not present, then the posture relationship between the target pet and each associated person is obtained from the identified posture relationships that match the target time interval. If it is determined that the posture relationship between the target pet and all associated personnel is an independent movement relationship, then an abnormal behavior warning message for the target pet is generated; the independent movement relationship means that the movement between the pet and the personnel is relatively independent. The step of identifying the pose relationships between pets and people in each video frame of the real-time video stream includes: The real-time video stream is input into the key point detection model, and the key point detection information of each pet and each person in each video frame output by the key point detection model is obtained. The key point detection information of each video frame is input into the continuous frame pose recognition model, and the pose relationships between each pet and each person included in each video frame output by the continuous frame pose recognition model are obtained.

2. The method according to claim 1, characterized in that, After extracting the target pet, at least one associated person who meets the distance threshold condition with the target pet, and the leash detection result matching the target pet from the image recognition result, the method further includes: If the leash detection result confirms the presence of a leash, then the behavior within the target monitoring area is considered normal pet behavior, and the monitoring of the target pet is terminated.

3. The method according to claim 1, characterized in that, Postural relationships include cooperative motion relationships and independent motion relationships; If the leash detection result indicates that the leash is not present, after obtaining the posture relationships between the target pet and each associated person from the identified posture relationships matched with the target time interval, the process further includes: If the posture relationship between the target pet and at least one associated person is determined to be a cooperative movement relationship, then the behavior of the target pet within the monitoring area is determined to be normal, and the monitoring of the target pet is terminated.

4. The method according to claim 1, characterized in that, The continuous frame pose recognition model is based on the ST-GCN action classification algorithm, which is a graph convolutional neural network.

5. A device for detecting abnormal pet behavior in public places, characterized in that, include: The posture relationship recognition module is used to acquire real-time video streams sent by surveillance cameras within the target monitoring area and identify the posture relationships between each pet and each person included in each video frame of the real-time video stream. The image recognition result processing module is used to extract the target pet, at least one associated person who meets the distance threshold condition with the target pet, and the leash detection result that matches the target pet from the image recognition result when the processor on the monitoring camera side sends the image recognition result that matches the target time interval. The posture relationship determination module is used to obtain the posture relationship between the target pet and each associated person from the identified posture relationships that match the target time interval if the leash detection result is that the leash does not exist. The abnormal behavior early warning module is used to generate abnormal behavior early warning information for the target pet if it is determined that the posture relationship between the target pet and all associated personnel is an independent movement relationship; the independent movement relationship means that the movement between the pet and the personnel is relatively independent. Specifically, the posture relationship recognition module is used to: input the real-time video stream into the key point detection model and obtain the key point detection information of each pet and each person in each video frame output by the key point detection model; input the key point detection information of each video frame into the continuous frame posture recognition model and obtain the posture relationship between each pet and each person included in each video frame output by the continuous frame posture recognition model.

6. A detection system for abnormal pet behavior in public places, characterized in that, This includes surveillance cameras, edge box-side processors, and surveillance camera-side processors; among which, The surveillance camera is used to acquire real-time video streams within the target monitoring area and sends the real-time video streams to the edge box processor and the surveillance camera processor, respectively. The edge-box end processor is used to perform the method for detecting abnormal pet behavior in public places as described in any one of claims 1-4; The monitoring camera-side processor is used to generate image recognition results for each video frame of the real-time video stream based on the real-time video stream sent by the monitoring camera in the target monitoring area, and when the target pet is determined in the target monitoring area based on the image recognition results, it filters the image recognition results that match the target time interval and sends them to the edge box-side processor.

7. The system according to claim 6, characterized in that, The surveillance camera-side processor is specifically used for: Using a biometric model to identify video frames in a real-time video stream, it can detect in real time whether there is at least one pet in the target monitoring area, and when at least one pet is detected in the target monitoring area, it can detect whether there is at least one person in the target monitoring area within a specified time interval. If it exists, the target pet is identified, and the distance estimation model is used to estimate the distance between each person in the target monitoring area and the target pet. Each person whose distance from the target pet is less than the preset distance threshold is regarded as the associated person of the target pet. If not, then it is determined that there is abnormal pet behavior within the target monitoring area.

8. The system according to claim 7, characterized in that, The surveillance camera-side processor is further specifically used for: After identifying the target pet and at least one associated person within the target monitoring area, the region of interest (ROI) of the target pet and the ROI of each associated person is obtained, and a leash detection model is used to detect whether a leash matching the target pet exists. The region of interest of the target pet, the region of interest of each associated person, and the detection results of the leash that matches the target pet are used as image recognition results and sent to the edge box processor.

9. The system according to any one of claims 6-8, characterized in that, The detection system for abnormal pet behavior in public places also includes a monitoring terminal alarm device and a back-end alarm device; The monitoring alarm device is used to issue a violation alarm in real time when abnormal pet behavior is detected in the target monitoring area, and to send video frames related to the abnormal pet behavior to the background alarm device. The background alarm device is used to determine the relevant personnel information and monitoring area information when it receives video frames related to abnormal pet behavior sent by the monitoring terminal alarm device, and to issue an alarm reminder to the supervisor based on the relevant personnel information and monitoring area information.

Images