Behavior identification method based on skeleton video

Abstract

The invention discloses a behavior identification method based on a skeleton video. The behavior identification method comprises the steps: carrying out frame extraction on each segment of video in adata set, obtaining a video training sample, and extracting the skeleton joint point information of each frame; initializing a skeleton joint point connection relation matrix according to the skeletonjoint point physical connection relation so as to preliminarily train the spatial domain graph convolutional neural network; after the parameters of the spatial domain graph convolutional neural network are fixed, training a skeleton joint relation reasoning network through the spatial domain graph convolutional neural network to obtain a new skeleton joint connection relation matrix; updating parameters of the spatial domain graph convolutional neural network through the new skeleton joint point connection relation matrix; and during application, framing a to-be-identified video, extractingskeleton joint point information and then sending the skeleton joint point information to the skeleton joint point relation reasoning network to obtain a skeleton joint point connection relation matrix, and sending the framed video and the skeleton joint point connection relation matrix to the updated spatial domain graph convolutional neural network to obtain the pedestrian behavior category.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a skeleton video-based behavior recognition method. Background technique [0002] The purpose of behavior recognition is to give the behavior category of the target in the video. It is a very important research field in computer vision. With the development of key point detection algorithms and depth cameras, skeleton-based behavior recognition methods have been applied in many fields. Such as monitoring scene early warning, human-computer interaction, virtual reality, etc. Compared with behavior recognition methods based on other modalities such as RGB, the skeleton-based behavior recognition method can more robustly extract human body shape and structure information, and the behavior recognition method based on skeleton information can remove color and texture The interference of features has made it achieve good recognition performance in most behavior categories. ...

AI Technical Summary

Problems solved by technology

The GCN-based method considers the topological structure information of the skeleton joint points, but these topological structure information is artificially defined in advance, and sometimes these topological structure information cannot fully describe the connection between the joint points

Images