Multi-target behavior identification method and system for monitoring video

Abstract

The invention provides a multi-target behavior identification method and system for monitoring video. The method comprises the following steps of respectively training a target detection model and a behavior recognition model; predicting position information of pedestrians in the current frame of the video, and taking the position information as a target detection box of the current frame; predicting a target tracking frame of the current frame through the previous frame information according to the target detection frame of the current frame, and calculating a target frame matching degree between the two frames; matching the target detection box of the current frame with the target tracking box of the current frame to obtain matching information; estimating a pedestrian target frame coordinate of the current frame, and predicting a target tracking frame coordinate of the pedestrian target in the next frame; cutting pedestrian pictures and storing pedestrian numbers; according to the pedestrian numbers, matching the pedestrian pictures with the same numbers in multiple continuous frames, combining into a list, and storing the pedestrian numbers; and if the length of the list reaches a specified frame number threshold, inputting the pedestrian pictures stored in the list into the behavior recognition model, and calculating the behavior category probability of the list.

Description

technical field [0001] The invention belongs to the field of video analysis, and relates to a surveillance video-oriented multi-target behavior recognition method and system. Background technique [0002] With the improvement of people's security awareness, many places use high-definition cameras instead of patrolling personnel. However, staff are still required to monitor and analyze the captured content, which is still labor-intensive from another perspective. In recent years, several hot research directions in the field of computer vision, such as object detection, object tracking, and behavior recognition, have made good breakthroughs in the academic community. How to apply the achievements in the field of computer vision to real life has become a new problem for engineers. The main reason is that the scenes where security cameras are placed in real life are diverse and complex, and the recognition rate of behavior recognition in practical applications is limited by de...

AI Technical Summary

Problems solved by technology

Due to the limited efficiency of optical flow extraction, the running speed is very slow, and real-time recognition cannot be done, so it is difficult to use in real life

Images