A behavior recognition method and system based on improved slowfast

Abstract

The application discloses a behavior recognition method and system based on an improved Slowfast, and the method comprises the following steps: acquiring a video to be recognized; inputting the video to be recognized into a trained behavior recognition model to generate a prediction result of behavior recognition, wherein the behavior recognition model comprises a slow branch, a fast branch and a BFE module, information fusion is performed on each stage of the slow branch and the fast branch through the BFE module, the slow branch samples the input video at a first time sampling step, the fast branch samples the input video at a second frame time sampling step, wherein the second frame sampling step is smaller than the first frame sampling step, the BFE module is used for aggregating spatial information of the slow branch and the fast branch to determine a time attention weight, so as to strengthen the feature information of the fast branch, and the information of the fast branch is fused to the slow branch. The application can improve the recognition accuracy under fast motion.

Images