SlowFast Behavior Recognition Algorithm Based on Residual Frame

In behavior recognition, the accuracy of recognition results is deeply affected by input video processing, which is crucial. Therefore, in order to improve the efficiency and accuracy of SlowFast behavior recognition model, we optimize the fast channel of the model, mainly in the fast channel, first apply residual frames to the video sequence to extract key frame changes, and then conduct behavior recognition based on these processed video data. This improvement aims to reduce computational redundancy, focus on more prominent motion features in the video, and improve the overall recognition efficiency and performance. Finally, the experimental results show that the proposed method improves the recognition accuracy compared with existing methods.