Human body abnormal behavior detection method based on YOLOv4 model

Through the human abnormal behavior detection method based on the YOLOv4 model and combined with the DS-YOLO algorithm, the problems of inefficient and missed detection of human abnormal behavior in the intelligent security monitoring system in public places are solved, and more efficient and accurate detection of human abnormal behavior is achieved.

CN118982865APending Publication Date: 2024-11-19SUZHOU LANGJIETONG INTELLIGENT TECH
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202411013304.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In the existing intelligent security monitoring system for public places, human abnormal behavior detection relies on manual inspection, resulting in inefficiency and missed inspection. Traditional visual algorithms have insufficient detection accuracy.

Method used

The human abnormal behavior detection method based on the YOLOv4 model is adopted to construct a human abnormal behavior model through deep learning, and the monitoring video is detected frame by frame, and supplementary detection is carried out in combination with the DS-YOLO algorithm, and the calculation results are fused to improve detection accuracy.

Benefits of technology

It effectively improves the efficiency and accuracy of abnormal behavior detection in humans, reduces the missed detection rate, improves the detection rate, and meets the detection requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982865A_ABST
    Figure CN118982865A_ABST
Patent Text Reader

Abstract

The invention discloses a human body abnormal behavior detection method based on a YOLOv4 model. The method comprises the following steps: S1, obtaining video data of a monitoring scene; s2, taking the human body behavior data set as input of a network model based on deep learning, completing human body target extraction and behavior state classification by training a deep neural network, constructing a relation between human body abnormal behaviors and states, and obtaining a YOLOv4 model of the human body abnormal behaviors; s3, performing target detection on the monitoring video frame by frame by using the YOLOv4 model obtained by training to obtain a human body abnormal behavior detection result; s4, performing supplementary detection by adopting a DS-YOLO human body abnormal behavior detection algorithm; and S5, human body abnormal behavior detection on a monitoring occasion is realized. Compared with the prior art, the problems that in an existing security and protection monitoring system, a visual algorithm is low in human body abnormal behavior detection efficiency of video data and missed in detection are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of security monitoring technology, and in particular to a method for detecting abnormal human behavior based on a YOL0v4 model. Background Art

[0002] Chinese patent CN108898717B discloses an intelligent security monitoring system, including: a personnel detection module, which is used to detect the position and identity of indoor personnel and determine whether there are unauthenticated personnel whose identities have not been verified; a simulation playback module, which is used to play a preset scene simulation sound after detecting the presence of an unauthenticated person in the room; a personnel analysis module, which is used to analyze the behavioral response of the unauthenticated person after hearing the scene simulation sound, and determine whether the response of the unauthenticated person is abnormal; an alarm reminder module, which is used to send an alarm reminder to a preset user device after detecting that the response of the unauthenticated person is abnormal.

[0003] In the existing intelligent security monitoring system for public places, the investigation and early warning of various human abnormal events in the monitoring video mostly rely on manual work, resulting in a small number of staff facing a large amount of investigation work, resulting in delayed problem discovery. The traditional visual algorithm first segments the target from the video sequence through contour information, then extracts features, compares the extracted human behavior features with abnormal behavior samples in the standard library, and finally determines whether there is abnormal behavior through classification. When processing information, detection, recognition and other tasks, there are problems such as low detection rate and missed detection, which cannot meet the detection requirements. In addition, the background of surveillance videos in different public areas is very complex, which further leads to low detection accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a method for detecting abnormal human behavior based on the YOL0v4 model to solve the problems of low efficiency and missed detection of abnormal human behavior in video data by visual algorithms in existing security monitoring systems.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme: a method for detecting abnormal human behavior based on the YOLOv4 model, comprising the following steps:

[0006] S1. Obtain video data of the monitoring scene, divide the monitoring scene screen into different grid areas, define abnormal human behavior and create a human behavior data set;

[0007] S2. Use the human behavior dataset as the input of the network model based on deep learning, complete human target extraction and behavior state classification by training the deep neural network, build the connection between abnormal human behavior and state, and obtain the YOLOv4 model of abnormal human behavior;

[0008] S3, using the trained YOLOv4 model to perform target detection on the surveillance video frame by frame to obtain the abnormal human behavior detection results;

[0009] S4, using DS-YOLO human abnormal behavior detection algorithm for supplementary detection, and merging the human abnormal behavior detection results calculated by YOLOv4 into a correlation matrix;

[0010] S5. Realize the detection of abnormal human behavior in the monitoring environment.

[0011] As a further description of the above technical solution:

[0012] In step S3, during the frame-by-frame detection of the surveillance video, the human behavior data set is scale-normalized, the candidate box coordinates of the human behavior, the confidence and category probability in the candidate box are tested, and the human behavior state category information is predicted for the different divided grid areas at the same time. According to the features in each pre-selected box, the pre-selected boxes with scores lower than the set threshold are filtered out, and the data is processed by non-maximum suppression of DIOU to obtain the detection results.

[0013] As a further description of the above technical solution:

[0014] In step S4, the DS-YOLO human abnormal behavior detection algorithm is used for supplementary detection, including the following steps:

[0015] S41, tracking the behavior trajectory of the human body in the surveillance video, monitoring the matching of the positioning results, and determining whether it is the same target human body;

[0016] S42, obtaining a matching result of monitoring and tracking by predicting, matching, and screening the human body trajectory of the next frame;

[0017] S43, calculate the tracking result and the detection result of the human body positioning, measure the matching degree between the two, calculate the human body feature value according to the detection algorithm, and calculate the minimum cosine distance between the feature values ​​to update the monitoring result.

[0018] As a further description of the above technical solution:

[0019] In step S43, the squared Mahalanobis distance is used to measure the matching degree between the tracking result and the human body positioning result.

[0020] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0021] 1. In the present invention, the abnormal human behavior detection based on the YOLOv4 model can effectively reduce the amount of repeated calculations in the process of extracting human behavior features and improve detection efficiency by filtering the feature values ​​in a branching manner.

[0022] 2. In the present invention, based on the DS-YOLO human abnormal behavior detection algorithm, DS is the DeepSort algorithm, and the DeepSort algorithm is used for detection, which is fused with the YOLOv4 calculation results into an association matrix to effectively reduce the missed detection rate. Combined with the algorithm matching tracking, the data is continuously updated to improve the detection rate when the human abnormal behavior detection task is heavy, improve the final human abnormal behavior detection rate, and avoid missed detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 This is a flow chart of a method for detecting abnormal human behavior based on the YOL0v4 model. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0028] In the description of the embodiments of the present invention, it should be noted that the terms "upper", "inner", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the inventive product is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0029] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Example 1

[0031] See also Figure 1 The present invention provides a technical solution: a method for detecting abnormal human behavior based on the YOLOv4 model, comprising the following steps:

[0032] S1. Obtain video data of the monitoring scene, divide the monitoring scene screen into different grid areas, define abnormal human behavior and create a human behavior data set;

[0033] S2. Use the human behavior dataset as the input of the network model based on deep learning, complete human target extraction and behavior state classification by training the deep neural network, build the connection between abnormal human behavior and state, and obtain the YOLOv4 model of abnormal human behavior;

[0034] S3, using the trained YOLOv4 model to perform target detection on the surveillance video frame by frame to obtain the abnormal human behavior detection results;

[0035] S4, using DS-YOLO human abnormal behavior detection algorithm for supplementary detection, and merging the human abnormal behavior detection results calculated by YOLOv4 into a correlation matrix;

[0036] S5. Realize the detection of abnormal human behavior in the monitoring environment.

[0037] In step S3, during the frame-by-frame detection of the surveillance video, the human behavior data set is scale-normalized, the candidate box coordinates of the human behavior, the confidence and category probability in the candidate box are tested, and the human behavior state category information is predicted for the different grid areas divided at the same time. According to the features in each pre-selected box, the pre-selected box with a score lower than the set threshold is filtered, and the data is processed by non-maximum suppression of DI0U to obtain the detection results. In the process of extracting human behavior features, convolution kernels are used for filtering to reduce the amount of repeated calculations in the process of extracting human behavior features without affecting performance.

[0038] In step S4, the DS-YOLO human abnormal behavior detection algorithm is used for supplementary detection, including the following steps:

[0039] S41, tracking the behavior trajectory of the human body in the surveillance video, monitoring the matching of the positioning results, and determining whether it is the same target human body;

[0040] S42, obtaining a matching result of monitoring and tracking by predicting, matching, and screening the human body trajectory of the next frame;

[0041] S43. Calculate the tracking results and the human body positioning results, use the square Mahalanobis distance to measure the degree of match between the two, find the human body feature values ​​according to the detection algorithm, and calculate the minimum cosine distance between the feature values ​​to update the monitoring results, thereby effectively improving the detection rate of abnormal human behavior detection tasks based on the YOLOv4 model in the surveillance video.

[0042] In step S1, after acquiring the video data of the monitoring scene, each frame of image data is processed, and each frame of image data is scaled, normalized, and channel sequence converted according to the model input requirements to achieve data preprocessing for subsequent processing and feature extraction.

[0043] Working principle: Based on the YOLOv4 model, abnormal human behavior detection can effectively reduce the amount of repeated calculations in the process of extracting human behavior features and improve detection efficiency by filtering the feature values ​​in a branching manner. Based on the DS-YOLO human abnormal behavior detection algorithm, DS stands for DeepSort algorithm, DS stands for DeepSort algorithm, and uses DeepSort algorithm for detection, which is fused with the YOLOv4 calculation results into an association matrix to effectively reduce the missed detection rate. Combined with algorithm matching tracking, the data is continuously updated to improve the detection rate of abnormal human behavior detection tasks, improve the final abnormal human behavior detection rate, and avoid missed detection.

[0044] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting abnormal human behavior based on the YOLOv4 model, characterized in that: The following steps are involved: S1. Obtain video data of the monitoring scene, divide the monitoring scene screen into different grid areas, define abnormal human behavior and create a human behavior data set; S2. Use the human behavior dataset as the input of the network model based on deep learning, complete human target extraction and behavior state classification by training the deep neural network, build the connection between abnormal human behavior and state, and obtain the YOLOv4 model of abnormal human behavior; S3, using the trained YOLOv4 model to perform target detection on the surveillance video frame by frame to obtain the abnormal human behavior detection results; S4, using DS-YOLO human abnormal behavior detection algorithm for supplementary detection, and merging the human abnormal behavior detection results calculated by YOLOv4 into a correlation matrix; S5. Realize the detection of abnormal human behavior in the monitoring environment.

2. According to a method for detecting abnormal human behavior based on the YOLOv4 model according to claim 1, it is characterized in that: In the step S3, during the frame-by-frame detection of the surveillance video, the human behavior data set is scale-normalized, the candidate box coordinates of the human behavior, the confidence and category probability in the candidate box are tested, and the human behavior state category information is predicted for the different divided grid areas at the same time. According to the features in each pre-selected box, the pre-selected boxes with a filter score lower than the set threshold are filtered, and the data is subjected to DI 0U non-maximum suppression processing to obtain the detection result.

3. According to a method for detecting abnormal human behavior based on the YOLOv4 model according to claim 2, it is characterized in that: In step S3, a convolution kernel is used for filtering in the process of extracting human behavior features.

4. A human abnormal behavior detection method based on the YOLOv4 model according to claim 1, characterized in that: In step S4, the DS-YOLO human abnormal behavior detection algorithm is used for supplementary detection, including the following steps: S41, tracking the behavior trajectory of the human body in the surveillance video, monitoring the matching of the positioning results, and determining whether it is the same target human body; S42, obtaining a matching result of monitoring and tracking by predicting, matching, and screening the human body trajectory of the next frame; S43, calculate the tracking result and the detection result of the human body positioning, measure the matching degree between the two, calculate the human body feature value according to the detection algorithm, and calculate the minimum cosine distance between the feature values ​​to update the monitoring result.

5. A human abnormal behavior detection method based on the YOLOv4 model according to claim 4, characterized in that: In step S43, the squared Mahalanobis distance is used to measure the matching degree between the tracking result and the human body positioning result.

Citation Information

Patent Citations

  • Human body abnormal behavior detection method based on T-TINY-YOLO network

    CN111898418A