A Human Action Recognition Method Based on Gaussian Process Hidden Variable Model

Abstract

The present invention claims to protect a discriminative human action recognition method based on Gaussian process hidden variable model and hidden conditional random field, which mainly includes the following three parts: In terms of acquiring motion data, the skeleton of the human body is obtained through motion capture technology or Kinect somatosensory technology Structure and motion information; In terms of extracting motion features, the Gaussian process hidden variable model with dynamic process and sparse approximation is used to obtain the popular structure of high-dimensional motion information in low-dimensional hidden space to represent motion features; in human action recognition , using discriminative implicit conditional random fields to model the features of temporal motion data and classify actions. The present invention can not only realize the visualization of human body motion features, but also can effectively use the information between the motion sequence data to perform high-precision recognition of human body motion, and is suitable for the field of real-time recognition of human body movements.

Description

technical field [0001] The invention belongs to the field of motion recognition, in particular to a discriminant recognition method based on a human body motion model of a Gaussian process hidden variable model and a hidden conditional random field. Background technique [0002] In the past two decades, human action recognition has become a hot research topic in the fields of computer vision, artificial intelligence, and pattern recognition. People hope that computers can think and understand some signals like the human brain, such as understanding our daily activities, so that computers can interact more naturally with humans. Its goal is to analyze the movements performed by the human body from an unknown image sequence. Recently, human action recognition has been widely used in video monitoring, human-computer interaction, medical care, intelligent security, virtual reality and other fields, and its research has important practical value and significance. [0003] Human...

AI Technical Summary

Problems solved by technology

In terms of feature extraction, it is easily affected by factors such as occlusion, viewing angle, lighting, dynamic background, and camera movement; in terms of classifier design, the algorithm research for time series data processing is still not deep enough to overcome the data independence and universality. and other stringent requirements

At present, a large number of researches focus on human action recognition based on video, and feature extraction is mostly an image processing process, while ignoring the important role of the chain bone structure of the human body in the recognition process.

Images