Personal abnormal behavior detection method based on CSRT target tracking algorithm optimization

Abstract

The invention discloses an individual abnormal behavior detection method based on CSRT target tracking algorithm optimization, and belongs to the technical field of abnormal behavior detection, and the method comprises the steps: configuring an OpenCV library and required environment variables, initializing required parameters, and obtaining an initial tracking target; after an initial tracking target is determined, entering a main loop tracking process, and analyzing and processing any frame of video data; and outputting a real-time tracking result, and displaying the information of the tracking target and the annotation of the abnormal behavior on a video interface in real time. According to the individual abnormal behavior detection method based on CSRT target tracking algorithm optimization provided by the invention, high-precision and real-time tracking of a single target in a video is realized, similarity measurement with a space-time constrained individual behavior rule base is carried out, and rapid identification and early warning are carried out on a potential abnormal behavior mode.

Description

Technical Field

[0001] The present invention relates to the technical fields of video metrology and abnormal behavior detection technology, and in particular to a personal abnormal behavior detection method based on the optimization of the CSRT target tracking algorithm. Background Art

[0002] With the rapid development of monitoring technology and computer vision technology, video-based abnormal behavior detection technology has become an indispensable key tool in the field of public safety. By capturing video streams in real time through surveillance cameras and using computer vision algorithms to deeply analyze the video content, abnormal behaviors can be identified and monitored in real time. By analyzing features such as the behavior trajectories, postures, speeds, and residence times of individuals or crowds in the video, abnormal behaviors that do not conform to normal behavior patterns, such as crowd gathering, running, pushing, and crowding, can be effectively identified. The real-time detection of these abnormal behaviors is of great significance for preventing crimes, avoiding traffic accidents, reducing crowd stampede incidents, and improving the level of public safety.

[0003] However, the existing video-based abnormal behavior detection methods face a series of technical challenges and limitations. First of all, traditional monitoring systems usually require manual intervention and long-term manual observation, which not only wastes a large amount of human resources, but also due to the huge amount of video data, the efficiency of manual analysis far cannot meet the requirements. In this process, abnormal behaviors are often easily overlooked or misjudged due to their scarcity, resulting in missed reports and false alarms. Secondly, the existing abnormal behavior detection methods usually rely on appearance feature-based detection, but for complex and dynamic scenarios, especially in crowded or dynamic environments, appearance features are not sufficient to effectively distinguish abnormal behaviors. Although some methods have begun to introduce motion features (such as trajectories, speeds, gaits, etc.) for auxiliary judgment, they often lack sufficient mining of the temporal information of motion features and fail to fully utilize the change patterns of these dynamic features, resulting in low detection accuracy. In addition, the existing detection systems usually face the problem of sample imbalance when processing video data. The number of samples of normal behaviors is much larger than that of abnormal behavior samples, and the types of abnormal behaviors are numerous and complex, posing challenges to the training and accuracy of the model. In order to improve the detection accuracy, the existing methods often require a large amount of labeled data, but the labeling cost of video data is high and time-consuming, greatly limiting the popularization and application of this technology. More importantly, some of the existing abnormal behavior detection methods, especially those based on trajectory analysis, mostly focus on macro-level behavior patterns, such as aggregation, crowding and other pattern recognition, lacking in-depth analysis of fine-grained individual behaviors. For example, the trajectory of an individual may exhibit certain special patterns in a specific scenario, such as zigzag wandering, S-shaped running, etc., but the existing technology still lacks the ability to recognize these complex trajectory patterns. Most of the existing methods fail to effectively utilize the dynamic temporal information of the trajectory, resulting in the inability to accurately match and identify the details of the trajectory behavior.

[0004] Therefore, although the existing technology has achieved initial results in some areas, there are still the following major problems in intelligent video surveillance and abnormal behavior detection: (1) High cost of data labeling: The amount of surveillance video data is huge, and obtaining high-quality data labels requires a lot of manpower and time.

[0005] (2) Insufficient utilization of motion features: Existing methods for analyzing motion features are mainly limited to simple statistics and splicing, and fail to deeply explore their spatiotemporal dynamic information, which limits the accurate identification of abnormal behaviors.

[0006] (3) Lack of fine-grained analysis of trajectory patterns: Current methods for detecting abnormal trajectory behavior mostly focus on macro-behavior patterns and fail to explore in depth the identification of individual behavior characteristics and complex trajectory patterns.

[0007] Therefore, existing abnormal behavior detection methods still have great limitations in practical applications, especially in the fields of smart cities, intelligent transportation and public safety, where the application effect is far from expected. How to improve the accuracy, real-time and robustness of video data analysis, especially the detection capability in complex dynamic scenes, is still a key problem that needs to be solved in this technology. Summary of the invention

[0008] The purpose of the present invention is to provide a method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization to solve the problems existing in the background technology.

[0009] To achieve the above object, the present invention provides a method for detecting abnormal behavior of an individual based on optimization of a CSRT target tracking algorithm, comprising the following steps: S1. Configure the OpenCV library and required environment variables and build a personal behavior rule library with spatiotemporal constraints, initialize the required parameters and obtain the initial tracking target, and use the homography matrix to convert the video frame pixel coordinates of the initial tracking target into the actual geographic space coordinates; S2. After determining the initial tracking target, enter the main loop tracking process, use the constructed time-space constraint personal behavior rule library, and use the similarity measurement method to compare and analyze the detection data of the initial tracking target with the threshold set in the personal behavior rule library to determine whether there is abnormal behavior; S3. Output real-time tracking results, and display the information of the tracking target and the annotation of abnormal behavior in real time on the video interface.

[0010] Preferably, initializing the required parameters in S1 includes: Create a CSRT-based target tracker for dynamic tracking of individual moving targets; Read video files and parse video frames; Set a homography matrix for mapping the pixel coordinates of video frames to actual geographical coordinates.

[0011] Preferably, the initial tracking target is obtained in S1 as follows: Read the first frame of the video; Prompt the user to select an initial tracking region ROI in the monitoring interface, determine whether the ROI is valid and define the initial position of the tracking target in this region; Initialize the CSRT tracker with the selected ROI and store the initial geometric information of the region.

[0012] Preferably, the processing steps in S2 include: S21. Use the CSRT tracker to update the position of the target in the video in real time; S22. Calculate the moving speed of the target according to the real-time trajectory information of the target; S23. Visualize the trajectory and real-time speed of the target in the video; S24. Compare the real-time trajectory information, moving speed information, sojourn time of the target with the personal behavior rules defined in the rule library to determine whether there is suspicious behavior; S25. User interaction management. After one round of judgment, the user re-selects the tracking region and initializes according to the needs.

[0013] Preferably, the content of S21 is as follows: Call the CSRT tracker to update the position of the target in the current frame, and obtain the center point coordinates and bounding box of the target; Record the pixel coordinates of the center point of the target, and at the same time use the homography matrix to convert the center point pixel coordinates into geographical coordinates.

[0014] Preferably, the content of S22 is as follows: Record the historical trajectory of the tracking target, including the position of the center point of the tracking target in each frame; Store the consecutive positions of the center point of the tracking target as a sequence of trajectory points for generating a complete tracking target trajectory. The trajectory data includes the sequence of trajectory points, displacement, and speed value, and is continuously updated and stored for subsequent analysis; Calculate the moving speed (unit: m / s) according to the displacement of the center point of the tracking target in adjacent frames.

[0015] Preferably, the content of S23 is as follows: Draw the bounding box of the individual tracking target in the current frame; Display the historical trajectory of the tracking target to form a visual representation of the continuous trajectory; Dynamically annotate the real-time speed of the tracking target on the video interface.

[0016] Preferably, the content of S24 is as follows: Perform similarity measurement on the generated tracking target trajectory and the behavior patterns in the spatio-temporal constrained personal behavior rule library, and determine whether the trajectory conforms to the abnormal pattern through feature matching; The content of the personal behavior rule library includes the defined individual target behavior patterns (such as Z-shaped, S-shaped, O-shaped trajectories, etc.) and their corresponding threshold conditions (speed range, residence time, etc.); Similarity measurement method: Combine features such as trajectory length, shape, and relative distance, and use the trajectory similarity analysis method (Euclidean distance matching) to calculate the similarity between the trajectory and the patterns in the personal behavior rule library; At the same time, combine the speed characteristics of the tracking target. When the speed is low and the trajectory pattern conforms to abnormal behavior, it is determined that there is suspicious behavior of the tracking target. Mark the tracking target as "suspected illegal behavior" or "abnormal trajectory", trigger a warning signal and lock the tracking target.

[0017] Preferably, collect the already obtained trajectory data and preprocess the trajectory data; Input the processed trajectory data into the personal behavior rule library, compare it with the time threshold, morphological similarity, and speed threshold in the personal behavior rule library, and determine whether the feature comparison result exceeds the threshold in the personal behavior rule library. If it does not exceed the threshold, output the compliance effect. If it exceeds the threshold, record the violation features, trigger a warning signal and track the features; Feed back the output result to the user terminal.

[0018] Preferably, the content of S3 is as follows: The video playback interface displays the target tracking frame, historical trajectory, and speed information; The abnormal behavior detection result is prompted in real time by a pop-up window, marking the behavior type, occurrence time, and specific location of the individual abnormal target; Output the specific features of the abnormal behavior, including trajectory type, similarity, speed range, stay (wander) time, and trigger rule conditions, etc.

[0019] Therefore, the present invention adopts the above-mentioned personal abnormal behavior detection method optimized based on the CSRT target tracking algorithm, and has the following beneficial effects: (1) Through the multi-dimensional analysis method of trajectory similarity measurement and speed range verification, compare the detection data of the target with the threshold set in the personal behavior rule library for comparative analysis, and achieve accurate identification of individual abnormal behavior. The algorithm can effectively detect complex trajectory patterns (such as Z-shaped, S-shaped, etc.) and abnormal behaviors such as low-speed wandering, providing technical guarantee for the rapid warning of abnormal individuals; (2) By introducing the homography matrix, the system realizes the accurate mapping between the pixel coordinates of video frames and the actual geographical coordinates, improving the accuracy of tracking target positions and trajectory analysis. At the same time, the optimized CSRT algorithm has strong anti-occlusion and high robustness in target tracking, ensuring stable operation in complex environments; (3) Only requires the user to select the region of interest (ROI) in the monitoring screen by mouse in the initial stage, and then the system can automatically complete the tasks of target tracking and abnormal behavior monitoring. Without manual annotation or complex settings, it significantly reduces the operation complexity and labor cost, and is applicable to the real-time monitoring requirements in various scenarios; (4) The introduced rule library supports dynamically adjusting or customizing the behavior pattern according to the actual scenario, enabling the system to adapt to the diverse requirements of different application scenarios (such as public safety, industrial monitoring, etc.), thus enhancing the scalability and generality of the algorithm; (5) Compared with traditional manual monitoring or annotation training methods relying on deep learning, this algorithm realizes efficient target behavior monitoring and early warning functions with lower computing resources and operation costs, and is applicable to scenarios with limited resources (few sample quantities).

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0021] Figure 1 It is a schematic diagram of the overall process of the embodiment of the present invention; Figure 2 It is a schematic diagram of the platform interface of the embodiment of the present invention; Figure 3 It is a schematic diagram of the early warning of the embodiment of the present invention; Figure 4 It is the real-time tracking trajectory of Embodiment 1 of the present invention, where (a) is a schematic diagram of the target speed of 0.36 m / s; (b) is a schematic diagram of the target speed of 0.79 m / s; (c) is a schematic diagram of the target speed of 0.50 m / s; (d) is a schematic diagram of the target speed of 0.25 m / s; (e) is a schematic diagram of the target speed of 0.75 m / s; (f) is a schematic diagram of the target trajectory early warning; Figure 5 It is the real-time tracking trajectory of Embodiment 2 of the present invention, where (a) is a schematic diagram of the target speed of 0.84 m / s; (b) is a schematic diagram of the target speed of 1.19 m / s; (c) is a schematic diagram of the target speed of 1.33 m / s; (d) is a schematic diagram of the target speed of 0.60 m / s; Figure 6This is the real-time tracking trajectory of Embodiment 3 of the present invention. Among them, (a) is a schematic diagram with the target speed of 0.85 m / s; (b) is a schematic diagram with the target speed of 0.60 m / s; (c) is a schematic diagram with the target speed of 1.34 m / s; (d) is a schematic diagram with the target speed of 1.20 m / s. Figure 7 This is the real-time tracking trajectory of Embodiment 4 of the present invention. Among them, (a) is a schematic diagram with the target speed of 0.85 m / s; (b) is a schematic diagram with the target speed of 1.34 m / s; (c) is a schematic diagram with the target speed of 1.20 m / s. Figure 8 This is the real-time tracking trajectory of Embodiment 5 of the present invention. Among them, (a) is a schematic diagram with the target speed of 0.60 m / s; (b) is a schematic diagram with the target speed of 0.84 m / s; (c) is a schematic diagram with the target speed of 1.33 m / s; (d) is a schematic diagram with the target speed of 0.84 m / s. Detailed implementation manners

[0022] The following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0023] Please refer to Figure 1 - Figure 2 , a personal abnormal behavior detection method optimized based on the CSRT target tracking algorithm, comprising the following steps: S1. Configure the OpenCV library and required environment variables, construct a personal behavior rule library with spatio-temporal constraints, initialize the required parameters, obtain the initial tracking target, and use the homography matrix to convert the video frame pixel coordinates of the initial tracking target into actual geospatial coordinates. The content of the personal behavior rule library with spatio-temporal constraints is shown in Table 1.

[0024] (1) Initializing the required parameters includes: Create a CSRT-based target tracker for dynamically tracking individual moving targets; Read the video file and parse the video frames; Set the homography matrix for mapping the video frame pixel coordinates to the actual geographical coordinates.

[0025] (2) The content of obtaining the initial tracking target is as follows: Read the first frame of the video; Prompt the user to select the initial tracking region ROI through the mouse in the monitoring interface, determine whether the ROI is valid, and define the initial position of the tracking target in this region; Initialize the CSRT tracker using the selected ROI and store the initial geometric information of the region.

[0026] Table 1 Content of the Personal Behavior Rule Base with Spatiotemporal Constraints ;

[0027] S2. After determining the initial tracking target, enter the main loop tracking process. Use the constructed personal behavior rule base with spatiotemporal constraints and compare and analyze the detection data of the initial tracking target with the thresholds set in the personal behavior rule base using a similarity measurement method to determine whether there are abnormal behaviors. For example Figure 3 , the real-time trajectory and real-time speed of the tracking target are displayed.

[0028] S21. Use the CSRT tracker to update the position of the target in the video in real time.

[0029] Call the CSRT tracker to update the position of the target in the current frame, and obtain the center point coordinates and bounding box of the target; Record the pixel coordinates of the center point of the target, and at the same time use the homography matrix to convert the pixel coordinates of the center point into geographic coordinates.

[0030] S22. Calculate the moving speed of the target according to the real-time trajectory information of the target.

[0031] Record the real-time trajectory of the target, including the center point position of each frame; store the consecutive positions of the center point of the target as a sequence of trajectory points for generating a complete trajectory record. The trajectory data includes the sequence of trajectory points, displacement, speed value, etc., and is continuously updated and stored for subsequent analysis; Calculate the moving speed (unit: m / s) according to the displacement of the center point of the target in adjacent frames.

[0032] S23. Visualize the trajectory and real-time speed of the target in the video.

[0033] Draw the bounding box of the individual tracking target in the current frame; Display the historical trajectory of the target to form a visual representation of the continuous trajectory; Dynamically annotate the real-time speed of the target on the interface.

[0034] S24. Compare the historical trajectory information, moving speed information, staying time of the target with the personal behavior rules defined in the rule base to determine whether there are suspicious behaviors.

[0035] After generating the trajectory features of the output target during the tracking process, load the preset personal behavior rule base with spatiotemporal constraints, perform similarity measurement between the generated target trajectory and the behavior patterns in the personal behavior rule base with spatiotemporal constraints, and judge whether the trajectory conforms to the abnormal pattern through feature matching; The content of the rule library includes the defined individual target behavior patterns (such as Z-shaped, S-shaped, O-shaped trajectories, etc.) and their corresponding threshold conditions (such as speed range, target stay time, etc.); Similarity measurement method: Combining features such as trajectory length, shape, relative distance, etc., the trajectory similarity analysis method (Euclidean distance matching) is used to calculate the similarity between the trajectory and the patterns in the rule library; At the same time, combining the speed characteristics of the target, when the speed is low and the trajectory pattern conforms to abnormal behavior, it is determined that the target has suspicious behavior. Mark the target as "suspected illegal behavior" or "abnormal trajectory", trigger a warning signal and lock the target.

[0036] The method for judging abnormal behavior using the personal behavior rule library is as follows: Collect the obtained trajectory data and preprocess the trajectory data; Input the processed trajectory data into the personal behavior rule library, compare it with the time threshold, morphological similarity, and speed threshold in the personal behavior rule library, and judge whether the feature comparison result exceeds the threshold in the personal behavior rule library. If it does not exceed the threshold, output a compliance effect. If it exceeds the threshold, record the violation features, trigger a warning signal and track the features; Feed the output result back to the user end.

[0037] S25. User interaction management. After a round of judgment, the user can reselect the tracking area and initialize according to needs. Press the ESC key to exit the program; press the R key to reselect the tracking area (ROI) and re-initialize the tracker.

[0038] S3. Output real-time tracking results, and display the information of the tracking target and the annotation of abnormal behavior in real time on the video interface, such as Figure 3 。

[0039] The video playback interface displays the target tracking frame, historical trajectory, and speed information; The detection result of abnormal behavior is prompted in real time through a pop-up window, annotating the behavior type, occurrence time, and specific location of the individual abnormal target; Output the specific features of abnormal behavior, including trajectory type, similarity, speed interval, and trigger rule conditions.

[0040] The trajectories of different targets have been measured in different regions, such as Figure 4 - Figure 8 They are the real-time tracking trajectory diagrams of Embodiment 1 to Embodiment 5 respectively, showing the effect of the method of the present invention for real-time reporting of target speed, trajectory, etc.

[0041] Therefore, the present invention adopts the above-mentioned method for detecting personal abnormal behaviors optimized based on the CSRT target tracking algorithm. This method realizes high-precision and real-time tracking of a single target in a video by introducing a homography matrix for target correction, and performs similarity measurement with a personal behavior rule base with spatio-temporal constraints to quickly identify and warn of potential abnormal behavior patterns. This technology is widely applied in the fields of intelligent video surveillance, traffic monitoring, public security, behavior analysis, etc., and particularly demonstrates important value in applications such as illegal behavior detection, abnormal behavior recognition and warning, and intelligent traffic management in intelligent security systems. This technology can also be extended and applied to scenarios such as smart city construction and autonomous driving to provide support for the intelligent development of multiple fields.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization, characterized in that: The following steps are involved: S1. Configure the OpenCV library and required environment variables and build a personal behavior rule library with spatiotemporal constraints, initialize the required parameters and obtain the initial tracking target, and use the homography matrix to convert the video frame pixel coordinates of the initial tracking target into the actual geographic space coordinates; S2. After determining the initial tracking target, enter the main loop tracking process, use the constructed time-space constraint personal behavior rule library, and use the similarity measurement method to compare and analyze the detection data of the initial tracking target with the threshold set in the personal behavior rule library to determine whether there is abnormal behavior; S3. Output real-time tracking results, and display the information of the tracking target and the annotation of abnormal behavior in real time on the video interface.

2. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 1 is characterized in that: Initialization of required parameters in S1 includes: Create a CSRT-based target tracker for dynamic tracking of individual targets; Read video files and parse video frames; Set the homography matrix to map the pixel coordinates of the video frame to the actual geographic coordinates.

3. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 2 is characterized in that: The initial tracking target content obtained in S1 is as follows: Read the first frame of the video; Prompt the user to select the initial tracking area ROI with the mouse in the monitoring interface, determine whether the ROI is valid and define the initial position of the tracking target in the area; Initialize the CSRT tracker with the selected ROI and store the initial geometry of the region.

4. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 3 is characterized in that: The processing steps in S2 include: S21, using the CSRT tracker to update the position of the target in the video in real time; S22, calculating the moving speed of the target according to the real-time trajectory information of the target; S23, visualizing the trajectory and real-time speed of the target in the video; S24, comparing the historical trajectory information, movement speed information, and stay time of the target with the personal behavior rules defined in the rule library to determine whether there is any suspicious behavior; S25, user interaction management, after a round of judgment, the user reselects the tracking area and initializes it according to needs.

5. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 4 is characterized in that: The contents of S21 are as follows: Call the CSRT tracker, update the tracking target position in the current frame, and obtain the center point coordinates and bounding box of the target; The pixel coordinates of the center point of the tracking target are recorded, and the homography matrix is ​​used to convert the pixel coordinates of the center point into geographic coordinates.

6. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 5, characterized in that: The content of S22 is as follows: Record the historical trajectory of the tracking target, including the center point position of the tracking target in each frame; The continuous position of the center point of the tracking target is stored as a trajectory point sequence to generate a complete tracking target trajectory. The trajectory data includes the trajectory point sequence, displacement, and velocity value, which are continuously updated and stored for subsequent analysis; The moving speed is calculated based on the displacement of the center point of the tracking target in adjacent frames.

7. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 4 is characterized in that: The content of S23 is as follows: Draw the bounding box of the individual tracked target in the current frame; Display the historical trajectory of the tracked target to form a visual representation of the continuous trajectory; Dynamically mark the real-time speed of the tracking target on the video interface.

8. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 5, characterized in that: The content of S24 is as follows: The generated tracking target trajectory is measured for similarity with the behavior pattern in the personal behavior rule library with time and space constraints, and feature matching is used to determine whether the trajectory conforms to the abnormal pattern; The content of the personal behavior rule library includes the defined individual target behavior patterns and their corresponding threshold conditions; Combined with the length, shape and relative distance characteristics of the tracking target trajectory, the trajectory similarity analysis method is used to calculate the similarity between the tracking target trajectory and the individual behavior rule library pattern; At the same time, combined with the speed characteristics of the tracking target, when the speed is low and the trajectory pattern conforms to abnormal behavior, it is determined that the tracking target has suspicious behavior, and the tracking target is marked as suspected illegal behavior or abnormal trajectory, triggering a warning signal and locking the tracking target.

9. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 8, characterized in that: The content of abnormal behavior judged by using personal behavior rule library is as follows: Collect the acquired trajectory data and pre-process the trajectory data; The processed trajectory data is input into the personal behavior rule library, and compared with the time threshold, morphological similarity and speed threshold in the personal behavior rule library to determine whether the feature comparison result exceeds the threshold in the personal behavior rule library. If it does not exceed the threshold, the compliance effect is output; if it exceeds the threshold, the violation feature is recorded, the warning signal is triggered and the feature is tracked; Feed the output results back to the user.

10. The method for detecting abnormal personal behavior based on CSRT target tracking algorithm optimization according to claim 9, characterized in that: The content of S3 is as follows: The video playback interface displays the target tracking frame, historical trajectory and speed information; The abnormal behavior detection results are displayed in a real-time pop-up window, marking the behavior type, occurrence time and specific location of individual abnormal targets; Output the specific characteristics of abnormal behavior, including trajectory type, similarity, speed range, stay time and trigger rule conditions.

Images