Two-stage process action detection method based on YOLO-TSM

The invention relates to the technical field of computer vision time sequence action detection, and discloses a two-stage process action detection method based on YOLO-TSM, and the method comprises the following steps: S1, obtaining an original process action video stream, and carrying out the preprocessing; s2, constructing a YOLO network to perform reasoning on a process action video stream, and obtaining spatial distribution information and action feature information of a human body; s3, constructing a TSM network, inputting human body space distribution information obtained from the YOLO network, and reasoning by obtaining time features of human body actions to obtain action feature information; and S4, reasoning specific actions by combining the threshold value, the action feature information obtained through the YOLO network and the action feature information obtained through the TSM network. According to the method, a plurality of algorithm models are fused, so that the real-time reasoning rate and the