Falling abnormal behavior identification method based on law enforcement and case handling area scene

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0039] This technical solution proposes a method for detecting falling down based on key points of the human body. The overall algorithm includes three parts: human body detection, key point positioning of the human body, and classification model. According to the overall recognition steps, it can be divided into four parts: two-dimensional image data acquisition, human body detection based on deep learning, human body key point positioning, and falling behavior recognition. (1) First obtain the video image data in the law enforcement case through the camera, (2) Then obtain the coordinate position of the human body detection frame through the target detection network, (3) Then use the human body key point detection network for each human body frame separately to obtain The position of each key point of the human body, (4) Finally, input each human body detection frame into the classification network for two classifications to determine whether the person is down, such as figure 1 The four overall steps are shown.

[0040] The detailed description of each step is as follows:

[0041] Step 1. Image data acquisition: Through the indoor camera installed in the law enforcement and case handling area, the activity video of the personnel in the camera area is acquired, and the image data containing the human body based on the time series is obtained.

[0042] Step 2. Human detection: Use Faster R-CNN algorithm with high real-time and accuracy in the target detection algorithm. By inputting the image obtained in step 1 to the detection network, the human body detection block diagram is finally obtained, and then through NMS (non- Maximum supression (non-maximum suppression) removes redundant human body detection block diagrams so that each human body has only a unique frame.

[0043] Step 3. Human body key point positioning: input the obtained human body detection frame image into SPPE (single-personpose estimator) network to obtain a human body key point distribution map. Such as figure 2 Shown is a schematic diagram of 14 key points of the human body. The white solid circles are the positions of the key points, including the head, neck, left and right shoulders, left and right elbow pads, left and right wrists, left and right hip joints, left and right knees, and left and right ankles.

[0044] Step 4. Recognition of falling behavior: such as image 3 Shown is the flow chart for judging each human body frame. When judging multiple human body frames, just repeat input. Input each human body detection frame and its key point coordinates in each image, first judge the spatial relationship between the human body detection frame and the detection area, and determine whether the human body is in the detection area, if it is to execute the next step, otherwise end this recognition; then pass The key point distribution of the human body determines whether the detected human body is a complete human body. If it is, it is input into the two-classification network, otherwise the recognition is ended. This step effectively solves the false detection caused by the inaccurate human body detection algorithm and only a partial human body is detected; Input the coordinate distribution map of the key points of the human body into the two-classification network to obtain the classification result; finally, judge whether it is a fall-down behavior according to the classification result, if it is, a warning is issued, otherwise the recognition is ended.

[0045] In the two-classification network in step 4, ResNet-18 (Residual Network-18, residual network-18 layer) composed of a residual structure added to a classification network with an 18-layer network is used, such as Figure 4 Schematic diagram of the residual structure. By adding the residual structure, the overall performance of the network is effectively improved and the recognition effect is improved. The residual structure function formula is:

[0046] Y=F(x)+x

[0047] In the formula, X is the convolution feature of the input residual structure; F(x) is the shortcut connections, the output of the convolutional layer surrounded by the residual structure; Y is the output of the residual structure.

[0048] The overall structure of the ResNet18 network is as Figure 5 As shown, the crossing arrow is the residual structure, the solid line is the direct transfer, and the dashed arrow is to keep the output dimension consistent and upgrade; the image (image) input length and width are 224×224; 3×3conv, 64 represents the convolution kernel It is 3×3, the feature dimension is 64, and the other layers are the same; the averge pool (average pooling) layer normalizes the dimensions of the image; 2-d fc represents the full convolutional layer of two dimensions, and the data of 512 dimensions Perform full convolution into two dimensions; finally pass the softmax (normalization operation) layer to get the probability value of yes or no falling down behavior.

[0049] The present invention also provides a computer-readable storage medium on which a computer program is stored, wherein the program is executed by a processor to realize the steps of the method for identifying abnormal behaviors of falling down.

[0050] The present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor implements the steps of the method for identifying abnormal behaviors when the program is executed.

[0051] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in further detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. protected range. Any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention also fall within the protection scope of the present invention.