Human skeleton action recognition method and system and medium

Abstract

The invention provides a human skeleton action recognition method and system and a medium. The method comprises the following steps: a data acquisition step: acquiring human skeleton data; an action structure chart inference step: according to the acquired human skeleton movement data, inferring a connection relationship among different joints, and constructing an action structure chart taking thejoints as points and taking the connection relationship as edges; an action structure feature extraction step: introducing the obtained action structure diagram into a depth model formed by spatial diagram convolution and time convolution, and extracting action semantic information of the joint; and an action analysis and recognition step: realizing action recognition according to the extracted joint action semantic information. The relationship of the nodes in the action is modeled, so that a motion mode is better captured, the method is not limited to the skeleton structure of the human body, the relationship of the joints in the action in the motion process and the relationship between the joints and the long-distance nodes in the biological structure are excavated, and the graph structure is generalized.

Description

technical field [0001] The present invention relates to the fields of video analysis and pattern recognition, and in particular, to a human skeleton action recognition method, system and medium. Background technique [0002] Human action recognition has a wide range of applications in the fields of video surveillance, human-computer interaction and virtual reality. Among them, skeletal motion is an expression of human motion, which can be collected by a kinematic camera and represented as a series of joints and bones. Skeleton data can effectively express actions, and has the advantages of strong anti-noise ability and low data dimension. Action recognition using skeletal data has a wide range of applications. [0003] The current method of human skeleton action recognition is usually to stitch together the joint coordinates at each moment, and use the feature extraction method to analyze the entire action sequence. With the development of deep learning technology, convol...

AI Technical Summary

Problems solved by technology

However, this modeling method only considers the relationship between adjacent joints in the biological structure of the human body. For different actions, joints with farther biological structures may have stronger semantic relationships, such as in the process of walking. In , the hands and feet are in coordinated motion, that is, the action relationship, which is stronger than the relationship between the wrist joint and the elbow joint; There is a relationship, such as the shoulder joint and the upper arm on the same side, that is, the structural relationship

Images