Target abnormal behavior identification method based on multichannel video stream analysis

The invention relates to the technical field of abnormal behavior recognition, and discloses a target abnormal behavior recognition method for multichannel video stream analysis, and the method comprises the steps: segmenting video stream data into channel frames and optical stream frames with target detection marks, and carrying out the self-attention weight calculation; respectively carrying out weighted coding and feature extraction on the channel frame and the optical flow frame; performing time sequence feature representation on the extracted multi-channel frame feature vector sequence and the extracted flow feature vector sequence to obtain a time sequence feature vector; and constructing an abnormal behavior recognition model to perform abnormal behavior recognition. According to the method, the accuracy of target abnormal behavior recognition is improved based on the feature information of multiple channels and the optical flow information, the time sequence difference of adjacent video frame images i