Self-supervised learning and skeleton information behavior identification method

Abstract

The invention discloses a self-supervised learning and skeleton information behavior recognition method, and relates to the technical field of computer vision. The method comprises the following steps: S1, constructing a configurable depth model; S2, in a network pre-training stage, obtaining a pre-training sample according to a preset optical flow prediction task, wherein the pre-training sample comprises a skeleton video and a label of an optical flow prediction task automatically generated by a machine; training the transformation network by using the pre-training sample to obtain an initial parameter theta' of the converter network; S3, in a network fine tuning stage, initializing the converter network according to the initial parameter theta ', and constructing a fine tuning depth model in combination with the initialized converter network and the randomly initialized fine tuning classification network; and S4, inputting a to-be-identified bone video into the trained fine adjustment depth model, and outputting a classification prediction result by the fine adjustment classification network. According to the invention, on the premise of ensuring high precision, human body behavior identification with better effect, robustness and generalization is realized.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a behavior recognition method of self-supervised learning and bone information. Background technique [0002] Human behavior recognition technology is an important and active basic research topic in the field of computer vision. The technology analyzes and classifies images or videos containing human actions to predict what is happening in the scene. With the development of video acquisition sensors and video surveillance, human behavior recognition has gradually become a research content with a wide range of application scenarios in intelligent monitoring, human-computer interaction, intelligent robots, etc., and has attracted more and more researchers' attention. At present, human behavior recognition is mainly aimed at the understanding of video data. [0003] At present, the research on human behavior recognition is mainly divided into two categories, namely recogni...

AI Technical Summary

Problems solved by technology

[0005] However, in the problem of human action recognition based on human skeleton video information, how to effectively extract the timing dependence information between different video frames is the main difficulty

Because compared with RGB images, human skeleton data has less data but more complex information, and human behavior is closely related to the action process in the entire video. If the information implicit in the timing cannot be effectively used, the accuracy of recognition will be affected. Rate

Images