A Method and System for Detecting and Identifying Abnormal Behavior in Communities

Abstract

This invention discloses a method and system for detecting and identifying abnormal behavior in communities. The method includes the following steps: acquiring and preprocessing videos of pedestrian behavior within the community; detecting whether pedestrians are carrying garbage bags based on the preprocessed videos; and identifying abnormal pedestrian behavior based on the results of the garbage bag detection. This invention proposes using the OpenPose human pose estimation algorithm to detect skeletal points of people and tracking them by calculating the Euclidean distance between these skeletal points. It uses the YOLOv4 model to detect garbage bags and employs a matching algorithm between people and garbage bags to detect whether people are carrying garbage bags. This invention can effectively identify four abnormal behaviors in real time within a community: fighting, littering, falling, and running.

Description

Technical Field

[0001] This invention belongs to the field of image action recognition and processing technology, specifically relating to a method and system for detecting and recognizing abnormal behavior in communities. Background Technology

[0002] With the development of China's social economy and the deepening of urbanization, people's demands for public safety are gradually increasing. Abnormal behavior recognition technology plays a crucial role in social governance, public safety, and smart communities, and has become a popular research area in recent years. In the past, community services relied on security personnel constantly watching surveillance videos to detect unusual situations. While this manual method of anomaly identification was highly accurate, the sheer volume of video data and the continuous operation of surveillance equipment made security personnel prone to fatigue, negligence, and overlooking abnormal behaviors, potentially leading to irreparable losses. Furthermore, after an incident, security management personnel often need to review surveillance footage for investigation and evidence collection. However, most segments in the video show normal situations, with the actual occurrence of abnormal behavior typically occupying only a small portion of the video. In such cases, security personnel must review the entire recording to conduct the necessary investigation and evidence collection, which is inefficient and requires significant manpower and resources. Therefore, this manual monitoring method struggles to detect abnormal events in surveillance videos in real time, failing to effectively guarantee the safety of community residents and the quality of community services.

[0003] Techniques such as object detection, human pose estimation, and personnel tracking can reliably analyze and locate surveillance videos, enabling the identification of human behavior. Current research primarily focuses on understanding skeletal point motion information, but has not yet deeply explored and utilized the interaction information between the human body and related objects, as well as the inherent characteristics of the objects. However, this information is crucial for the accurate identification of human-object interaction behaviors. For example, two important characteristics of littering are the tendency of the garbage bag to leave the hand and the trajectory of the garbage bag as it falls. Without considering the interaction information of the garbage bag leaving the body, littering can be confused with behaviors such as arm swinging, leading to false detections. Simultaneously, the trajectory of the garbage bag can assist in the identification of littering, preventing missed detections. Therefore, incorporating object-related information into the abnormal behavior system is significant for improving the effectiveness of abnormal behavior identification. To address these needs, this invention proposes an abnormal behavior detection and identification method for communities, enabling real-time identification of abnormal behaviors such as littering, falling, and running within the community. First, an object detection algorithm is used to detect garbage bags in the video. Then, the garbage bag is used as an external skeletal point to connect with human skeletal points, and the trajectory of the garbage bag is calculated. This method incorporates object-related information into the system to improve the effectiveness of abnormal behavior recognition. Summary of the Invention

[0004] This invention aims to address the shortcomings of existing technologies by proposing a method and system for detecting and recognizing abnormal behavior in communities, thereby solving the problem that the interaction information between the human body and related objects, as well as the inherent feature information of the objects, are not fully utilized in action recognition.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for detecting and identifying abnormal behavior in communities includes the following steps:

[0007] Acquire videos of pedestrian behavior within the community and preprocess them;

[0008] Based on the preprocessed behavioral video, detect whether pedestrians are carrying garbage bags;

[0009] Based on the results of the detection of whether or not a person is carrying a garbage bag, abnormal pedestrian behavior is detected and identified.

[0010] Preferably, the preprocessing includes: skeletal point extraction, pedestrian tracking, and skeletal point storage;

[0011] The method for extracting the skeletal points is as follows:

[0012] Based on frame-by-frame images of the behavioral video, the OpenPose human pose estimation algorithm is used to detect skeletal points of pedestrians in the images, obtain the two-dimensional coordinates of human skeletal points, and realize the extraction of the skeletal points.

[0013] The pedestrian tracking method is as follows:

[0014] Based on the frame-by-frame images, the first Euclidean distance between a pedestrian in a preset number of frames and all pedestrians in the previous frame is calculated. The person with the shortest first Euclidean distance is identified as the same person in different frames, thus achieving pedestrian tracking.

[0015] The method for saving the bone points is as follows:

[0016] Obtain a time window with a preset number of frames, and save the skeletal points of the same person within the time window into a queue to obtain the skeletal point queue.

[0017] The preferred method for detecting whether pedestrians are carrying garbage bags is as follows:

[0018] Obtain the dataset of garbage bag images and divide it into training and test sets;

[0019] Based on the training set and the test set, train the YOLOv4 object detection model;

[0020] Based on the trained YOLOv4 object detection model, garbage bags are detected frame by frame in the behavior video to obtain the position coordinates of the garbage bags in each frame.

[0021] Based on the location coordinates of the garbage bag, calculate the second Euclidean distance between the preset garbage bag and all pedestrian neck bone points in the frame-by-frame image;

[0022] The pedestrian with the smallest second Euclidean distance from the preset garbage bag is identified as a possible carrier of the preset garbage bag;

[0023] When the minimum value of the second Euclidean distance is lower than a preset threshold, the potential carrier of the garbage bag is identified as the garbage bag carrier; otherwise, the preset garbage bag is identified as uncarried.

[0024] Preferably, the abnormal pedestrian behavior includes: abnormal movement behavior in the community and littering behavior in the community.

[0025] Preferably, the method for detecting and identifying abnormal movement behavior in the community is as follows:

[0026] The extracted skeletal points are optimized by removing skeletal points from the pedestrian's head, discarding skeletal point data that OpenPose did not estimate, and filling in the positions of missing skeletal points to achieve preprocessing of the skeletal points.

[0027] Based on the preprocessed skeletal points and the skeletal point queue, a skeleton spatiotemporal graph is constructed.

[0028] A graph convolutional network is used to extract human motion information from the spatiotemporal graph of the skeleton, thereby enabling the detection and identification of abnormal movement behaviors in the community.

[0029] Preferably, the method for detecting and identifying littering behavior in the community is as follows:

[0030] Based on the preprocessed skeletal points, obtain the skeletal points of the human hand.

[0031] Using the garbage bag as an exoskeleton point, it is connected to the human hand bone point to construct a human skeleton diagram associated with the garbage bag;

[0032] Connect the skeletal points of the human skeleton diagram used to construct the associated garbage bag in different frames to obtain the spatiotemporal diagram of the object-related skeleton.

[0033] Based on the spatiotemporal graph of the object's related skeleton, the motion trajectory of the garbage bag is modeled to obtain the motion trajectory feature information of the garbage bag.

[0034] The graph convolutional network is used to extract the spatiotemporal information of human motion from the spatiotemporal graph of the object's related skeleton;

[0035] The spatiotemporal information of human movement is spliced ​​and fused with the movement trajectory feature information of the garbage bag to obtain fused features;

[0036] Based on the fusion features, the detection and identification of littering behavior in the community can be achieved.

[0037] The preferred method for obtaining the movement trajectory feature information of the garbage bag is as follows:

[0038] The two-dimensional position coordinates of the garbage bag in the preset number of frames and the two-dimensional position coordinates of the garbage bag in the previous frame of the preset number of frames are obtained and calculated to obtain the garbage bag offset vector.

[0039] Based on the offset vector of the garbage bag, the magnitude of the offset vector is calculated to obtain the motion trajectory feature information of the garbage bag.

[0040] This application also provides an abnormal behavior detection and identification system for communities, including: a preprocessing module, a garbage bag detection module, and an abnormal behavior detection module;

[0041] The preprocessing module is used to acquire and preprocess videos of pedestrian behavior within the community.

[0042] The garbage bag detection module is used to detect whether a pedestrian is carrying a garbage bag based on the preprocessed behavioral video.

[0043] The abnormal behavior detection module is used to detect and identify abnormal pedestrian behavior based on the result of the detection of whether or not a person is carrying a garbage bag.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] This invention acquires videos of pedestrian behavior within a community and preprocesses them to extract skeletal points from the pedestrians in the videos, track the pedestrians, and save the skeletal points. Based on the preprocessed video, it detects whether the pedestrians are carrying garbage bags. This invention first detects whether they are carrying garbage bags and then performs behavior detection, making the system more accurate and efficient. This invention fuses the acquired garbage bag and human motion information as the basis for recognition, resulting in higher recognition accuracy. Attached Figure Description

[0046] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the 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.

[0047] Figure 1This is a flowchart of the overall implementation scheme of Embodiment 1 of the present invention;

[0048] Figure 2 This is a schematic diagram of the skeleton point preprocessing operation in Embodiment 1 of the present invention; wherein, Figure a is the skeleton map detected by OpenPose, and Figure b is the skeleton map after the skeleton point preprocessing operation;

[0049] Figure 3 This is a schematic diagram of the skeleton spatiotemporal diagram of Embodiment 1 of the present invention;

[0050] Figure 4 This is a schematic diagram of the object-related skeleton spatiotemporal diagram of Embodiment 1 of the present invention;

[0051] Figure 5 This is a schematic diagram of the movement trajectory of the garbage bag according to Embodiment 1 of the present invention;

[0052] Figure 6 This is a schematic diagram of the overall structure of Embodiment 1 of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] like Figure 1 As shown, a method for detecting and identifying abnormal behavior in a community includes the following steps:

[0057] Acquire videos of pedestrian behavior within the community and preprocess them;

[0058] Based on preprocessed behavioral videos, detect whether pedestrians are carrying garbage bags;

[0059] Based on the results of detecting whether or not a pedestrian is carrying a garbage bag, abnormal pedestrian behavior is detected and identified.

[0060] Preprocessing includes: skeleton point extraction, pedestrian tracking, and skeleton point saving;

[0061] The method for extracting skeletal points is as follows:

[0062] Based on frame-by-frame images of behavioral videos, the OpenPose human pose estimation algorithm is used to detect skeletal points of pedestrians in the images and obtain the two-dimensional coordinates of human skeletal points, thereby realizing the extraction of skeletal points; specifically, this embodiment obtains the two-dimensional coordinates of 18 human skeletal points.

[0063] The pedestrian tracking method is as follows:

[0064] Based on frame-by-frame images, the first Euclidean distance between a pedestrian in a preset number of frames and all pedestrians in the previous frame is calculated. The person with the shortest first Euclidean distance is identified as the same person in different frames, thus achieving pedestrian tracking. Specifically, this embodiment uses 5 frames as a time window.

[0065] The method for saving bone points is as follows:

[0066] Obtain a time window with a preset number of frames, save the skeleton points of the same person within the time window in a queue, and obtain the skeleton point queue.

[0067] The method for detecting whether pedestrians are carrying garbage bags is as follows:

[0068] Obtain the dataset of garbage bag images and divide it into training and test sets;

[0069] Based on training and testing sets, a YOLOv4 object detection model is trained. Specifically, for garbage bag identification, before detecting whether a person is carrying a garbage bag, a large number of garbage bag images need to be collected for training the YOLOv4 object detection model. First, a web crawler tool is written in Python to find and download publicly available free image resources on the Internet. Then, the downloaded images are manually screened, and images that do not meet the criteria are deleted. Ultimately, this invention collects more than 2,000 relevant images covering multiple scenarios. This invention uses 70% of the collected image data as the training set for training, and the remaining 30% as the testing set. The trained YOLOv4 model can accurately detect the location of garbage bags in the images.

[0070] Based on the trained YOLOv4 object detection model, garbage bags are detected frame by frame in the behavior video to obtain the position coordinates of the garbage bags in each frame.

[0071] Based on the location coordinates of the garbage bag, calculate the second Euclidean distance between the preset garbage bag and all pedestrian neck bone points in each frame of the image;

[0072] The pedestrian who is closest to the second European distance from the pre-set garbage bag is identified as a potential carrier of the pre-set garbage bag;

[0073] When the minimum value of the second Euclidean distance is lower than a preset threshold, a potential carrier of the garbage bag is identified as a garbage bag carrier; otherwise, the garbage bag is assumed to be unclaimed. Specifically, after garbage bag detection and matching are completed, the skeletal point queue of individuals not carrying garbage bags is used to identify abnormal movement behaviors in the community. If an individual is carrying a garbage bag, the skeletal point queue of the individual and the location information of the garbage bag are used to identify garbage disposal behavior in the community.

[0074] Abnormal pedestrian behavior includes: abnormal movement behavior in the community (normal behavior) and littering behavior in the community (littering behavior).

[0075] Abnormal movement behaviors in the community specifically include falls and running, and the detection and identification methods are as follows:

[0076] The extracted skeletal points are optimized by removing skeletal points from the pedestrian's head, discarding skeletal point data that OpenPose did not estimate, and filling in the positions of missing skeletal points, thus achieving skeletal point preprocessing. Specifically, the skeletal point preprocessing operations are as follows: Figure 2 As shown in Figure a. Figure a is the skeleton detected by OpenPose, and Figure b is the skeleton after skeletal point preprocessing. For abnormal behaviors such as running and falling, the motion information of head skeletal points is not important. Therefore, in the normal behavior recognition process, five head skeletal points are removed: two for the eyes, two for the ears, and one for the head. When OpenPose fails to estimate human skeletal points or when the estimated skeletal points do not include neck skeletal points, the skeletal data is discarded. Furthermore, when OpenPose generates incomplete human skeletal data containing neck skeletal points, it uses the distance vector of the missing skeletal points relative to the neck in the previous frame to fill in the positions of the missing skeletal points.

[0077] Based on the preprocessed skeletal points and the skeletal point queue, a skeleton spatiotemporal graph is constructed. Specifically, after completing the skeletal point preprocessing, the skeleton spatiotemporal graph is constructed using the skeletal point queue within the time window. First, skeletal points in the same frame are connected according to the natural structure of the human body to form the skeleton graph. Then, identical skeletal points in adjacent frames within the time window are connected to form the skeleton spatiotemporal graph, referring to... Figure 3 .

[0078] A graph convolutional network (GCNN) is used to extract human motion information from the spatiotemporal skeleton map, enabling the detection and identification of abnormal movement behaviors in the community. Specifically, after obtaining the spatiotemporal skeleton map, a GCNN is used to extract the human motion information contained within it. The GCNN consists of nine graph convolutional layers, with the following output feature channels for each layer: 64, 64, 64, 128, 128, 128, 256, 256, and 256, respectively. At the end of the network, a global average pooling layer is used to generate feature vectors. Finally, the feature vectors are input into a fully connected layer classifier for identifying abnormal movement behaviors in the community.

[0079] The method for detecting and identifying littering behavior in the community is as follows:

[0080] Based on the preprocessed skeletal points, obtain the skeletal points of the human hand.

[0081] Using garbage bags as exoskeleton points, and connecting them with the skeletal points of the human hand, a human skeleton diagram related to garbage bags is constructed.

[0082] By connecting the skeletal points of the human skeleton map used to construct the associated garbage bag from different frames, a spatiotemporal map of the object-related skeleton is obtained; specifically, the construction of the object-related skeleton spatiotemporal map is implemented. First, reliable human skeleton points are obtained through the same skeletal point preprocessing operation as in the recognition of abnormal movement behavior in the community. Then, the garbage bag is connected as an extraoskeleton point to the two hand skeleton points of the human body to construct a human skeleton map associated with the garbage bag. If the garbage bag is missing, its position in the adjacent frame is used to fill in the gap. Next, the same skeletal points from different frames are connected to form the object-related skeleton spatiotemporal map, referring to... Figure 4 .

[0083] Based on the object-related skeleton spatiotemporal graph, the motion trajectory of the garbage bag is modeled to obtain the motion trajectory feature information of the garbage bag.

[0084] Using graph convolutional networks, spatiotemporal information of human motion is extracted from spatiotemporal graphs of human-related skeletons;

[0085] The spatiotemporal information of human movement is spliced ​​and fused with the trajectory feature information of garbage bags to obtain fused features;

[0086] Based on fusion features, the detection and identification of littering behavior in the community can be achieved.

[0087] The method for obtaining the trajectory feature information of garbage bags is as follows:

[0088] Obtain the two-dimensional position coordinates of the garbage bag in the image of the preset number of frames and the two-dimensional position coordinates of the garbage bag in the previous frame image of the preset number of frames, and calculate to obtain the garbage bag offset vector.

[0089] Based on the offset vector of the garbage bag, the magnitude length of the offset vector is calculated to obtain the motion trajectory feature information of the garbage bag.

[0090] Specifically, the process involves modeling the trajectory of the garbage bag. After constructing the spatiotemporal graph of the object's related skeleton, the garbage bag's trajectory will be further modeled during the garbage disposal behavior recognition process to obtain inherent object motion information and reduce the false negative rate of garbage disposal behavior. (Refer to...) Figure 5 When littering occurs, the trajectory of the garbage bag exhibits a characteristic descent offset. The 2D position coordinates of the garbage bag are subtracted from the 2D position coordinates of the same garbage bag in the previous frame to obtain the garbage bag offset vector. The magnitude of this offset vector is then calculated and used as the feature information of the garbage bag's trajectory. A spatiotemporal graph of the object's skeleton is input into a graph convolutional network (GCNN) to extract the spatiotemporal information of human motion. The GCNN structure is the same as that used in ordinary behavior recognition. Then, the spatiotemporal information of human motion is concatenated and fused with the trajectory feature information of the garbage bag. Finally, the fused features are input into a fully connected layer classifier to detect and recognize littering behavior in the community.

[0091] The above methods are combined to form a process for detecting and identifying abnormal behavior in the community, referring to... Figure 6 .

[0092] The test video is input into the community abnormal behavior detection and recognition process, and the action category in each frame of the test video is output.

[0093] Example 2

[0094] This application also provides an abnormal behavior detection and identification system for communities, including: a preprocessing module, a garbage bag detection module, and an abnormal behavior detection module;

[0095] The preprocessing module is used to acquire and preprocess videos of pedestrian behavior within the community.

[0096] The garbage bag detection module is used to detect whether pedestrians are carrying garbage bags based on pre-processed behavioral videos.

[0097] The abnormal behavior detection module is used to detect and identify abnormal pedestrian behavior based on the results of detecting whether or not a person is carrying a garbage bag.

[0098] Preprocessing includes: skeleton point extraction, pedestrian tracking, and skeleton point saving;

[0099] The method for extracting skeletal points is as follows:

[0100] Based on frame-by-frame images of behavioral videos, the OpenPose human pose estimation algorithm is used to detect skeletal points of pedestrians in the images and obtain the two-dimensional coordinates of human skeletal points, thereby realizing the extraction of skeletal points; specifically, this embodiment obtains the two-dimensional coordinates of 18 human skeletal points.

[0101] The pedestrian tracking method is as follows:

[0102] Based on frame-by-frame images, the first Euclidean distance between a pedestrian in a preset number of frames and all pedestrians in the previous frame is calculated. The person with the shortest first Euclidean distance is identified as the same person in different frames, thus achieving pedestrian tracking. Specifically, this embodiment uses 5 frames as a time window.

[0103] The method for saving bone points is as follows:

[0104] Obtain a time window with a preset number of frames, save the skeleton points of the same person within the time window in a queue, and obtain the skeleton point queue.

[0105] The method for detecting whether pedestrians are carrying garbage bags is as follows:

[0106] Obtain the dataset of garbage bag images and divide it into training and test sets;

[0107] Based on training and testing sets, a YOLOv4 object detection model is trained. Specifically, for garbage bag identification, before detecting whether a person is carrying a garbage bag, a large number of garbage bag images need to be collected for training the YOLOv4 object detection model. First, a web crawler tool is written in Python to find and download publicly available free image resources on the Internet. Then, the downloaded images are manually screened, and images that do not meet the criteria are deleted. Ultimately, this invention collects more than 2,000 relevant images covering multiple scenarios. This invention uses 70% of the collected image data as the training set for training, and the remaining 30% as the testing set. The trained YOLOv4 model can accurately detect the location of garbage bags in the images.

[0108] Based on the trained YOLOv4 object detection model, garbage bags are detected frame by frame in the behavior video to obtain the position coordinates of the garbage bags in each frame.

[0109] Based on the location coordinates of the garbage bag, calculate the second Euclidean distance between the preset garbage bag and all pedestrian neck bone points in each frame of the image;

[0110] The pedestrian who is closest to the second European distance from the pre-set garbage bag is identified as a potential carrier of the pre-set garbage bag;

[0111] When the minimum value of the second Euclidean distance is lower than a preset threshold, a potential carrier of the garbage bag is identified as a garbage bag carrier; otherwise, the garbage bag is assumed to be unclaimed. Specifically, after garbage bag detection and matching are completed, the skeletal point queue of individuals not carrying garbage bags is used to identify abnormal movement behaviors in the community. If an individual is carrying a garbage bag, the skeletal point queue of the individual and the location information of the garbage bag are used to identify garbage disposal behavior in the community.

[0112] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A community-oriented abnormal behavior detection and recognition method, characterized in that, The method comprises the following steps: Obtaining a pedestrian behavior video in a community and pre-processing the video; Based on the pre-processed behavior video, detecting whether a pedestrian carries a garbage bag; Based on the result of the garbage bag carrying detection, detecting and identifying the abnormal behavior of the pedestrian; The pre-processing comprises bone point extraction, pedestrian tracking and bone point preservation; The method for extracting the bone points is as follows: Based on the frame-by-frame image of the behavior video, the OpenPose human posture estimation algorithm is used to detect the bone points of the pedestrians in the image, obtain the two-dimensional coordinates of the human bone points, and realize the extraction of the bone points; The realization method of the pedestrian tracking is as follows: Based on the frame-by-frame image, the first Euclidean distance between the pedestrians in the preset frame number of images and all pedestrians in the previous frame image is calculated, the person with the shortest first Euclidean distance is identified as the same person in different frame images, and the pedestrian tracking is realized; The method for preserving the bone points is as follows: A time window of a preset frame number is obtained, the bone points of the same person in the time window are preserved in a queue, and a bone point queue is obtained; The method for detecting whether a pedestrian carries a garbage bag is as follows: Obtaining garbage bag picture data set and dividing the training set and the test set; Based on the training set and the test set, training a YOLOv4 target detection model; Based on the trained YOLOv4 target detection model, detecting the garbage bag in the frame-by-frame image of the behavior video, and obtaining the position coordinates of the garbage bag in the frame-by-frame image; Based on the position coordinates of the garbage bag, calculating the second Euclidean distance between the preset garbage bag and the neck bone points of all pedestrians in the frame-by-frame image; The pedestrian with the minimum second Euclidean distance from the preset garbage bag is identified as the possible carrier of the preset garbage bag; When the minimum value of the second Euclidean distance is lower than a preset threshold, the possible carrier of the garbage bag is identified as a garbage bag carrier, otherwise, it is identified that the preset garbage bag is not carried by anyone; The abnormal behavior of the pedestrian includes the abnormal movement behavior in the community and the garbage throwing behavior in the community; The detection and identification method of the abnormal movement behavior in the community is as follows: Optimizing the extracted bone points, removing the bone points of the head of the pedestrian, discarding the bone point data not estimated by the OpenPose, filling the positions of the missing bone points, and realizing the pre-processing of the bone points; Based on the pre-processed bone points and the bone point queue, a skeleton space-time graph is constructed; Using a graph convolution network, the human movement information in the skeleton space-time graph is extracted, and the detection and identification of the abnormal movement behavior in the community are realized; The detection and identification method of the garbage throwing behavior in the community is as follows: Based on the pre-processed bone points, obtaining the palm bone points of the human body; Connecting the garbage bag as an extracorporeal bone point with the palm bone points of the human body to construct a human skeleton graph associated with the garbage bag; Connecting the bone points used to construct the human skeleton graph associated with the garbage bag in different frames to obtain an object-related skeleton space-time graph; Based on the object-related skeleton space-time graph, modeling the movement trajectory of the garbage bag to obtain the movement trajectory feature information of the garbage bag; Adopt the graph convolution network, extract the human motion space-time information in the object related skeleton space-time graph; The human motion space-time information is spliced and fused with the motion trajectory feature information of the garbage bag to obtain fused features. Based on the fused features, the detection and identification of the garbage throwing behavior in the community are realized.

2. The community-oriented anomalous behavior detection identification method of claim 1, wherein, The method for obtaining the motion trajectory feature information of the garbage bag is: Obtain the two-dimensional position coordinates of the garbage bag in the preset number of images and the two-dimensional position coordinates of the garbage bag in the last frame image of the preset number of images, and calculate to obtain the garbage bag offset vector; Based on the garbage bag offset vector, the module length of the offset vector is calculated to obtain the motion trajectory feature information of the garbage bag.

3. A community-oriented abnormal behavior detection and recognition system for implementing the method of any one of claims 1-2, characterized by, It includes: A pretreatment module, a garbage bag detection module and an abnormal behavior detection module; The pretreatment module is used for acquiring the pedestrian behavior video in the community and performing pretreatment; The garbage bag detection module is used for detecting whether the pedestrian carries a garbage bag based on the pretreated behavior video; The abnormal behavior detection module is used for detecting and identifying the abnormal behavior of the pedestrian based on the result of the garbage bag carrying detection.

Images