Research On Human Action Recognition And Interaction Technology In Virtual Environment

The research of three-dimensional virtual environment is one of the most popular research in the computer field.It is the integration of graphics,images,human motion capture and other technologies and virtual reality technology also depends on virtual environment.Therefore,it has attracted wide attention in academic,medical,industrial and other fields.Because the interaction mode in virtual environment is not intelligent enough and the accuracy of intelligent interaction mode is also inadequate.This paper studies human motion data acquisition,human motion feature extraction,human motion recognition,and interaction technology in virtual environment.The main work is as follows:Aiming at the problem that only the position information of joint points is stored in human motion data,which leads to the deformation of three-dimensional model,this paper proposes a method to express human motion by using the angle deviation of each joint point in the skeleton relative to the parent joint point.In order to verify this method,three human motion databases are collected by using OptiTrack motion capture equipment and related data preprocessing methods,and the preview of joint point offset data is realized.Experiments show that the human motion data saved by this method can effectively prevent the deformation of the three-dimensional virtual model and the error of position data is about 0.6 mm.On the basis of human joint point migration data,this paper presents a human motion feature representation method combining this data with key frames.In this method,human motion is divided into static and dynamic features.The static features are represented by the angle deviation of each joint point relative to the parent joint point,and the dynamic features of human motion are expressed by key frames.When selecting key frames,this paper improves the frame subtraction algorithm by adding the second derivative calculation of reconstruction error,so that the number of key frames can be determined automatically.Experiments show that this feature extraction method can not only reduce the amount of human motion data,but also automatically extract human motion features.Aiming at the problem that the hidden state number is difficult to determine and the recognition accuracy is insufficient when using hidden Markov model(HMM)to model a kind of human motion,this paper proposes a key frame based HMM action recognition method.When using HMM for motion recognition,aiming at the problem of large amount of computation and error in discretization of observation state,this paper uses high-dimensional Gauss function to fit human motion data,and completes GMM-HMM motion recognition method.Experiments show that the number of key frames can represent the number of hidden states of HMM,and the recognition accuracy can be improved by fitting human motion data with high-dimensional Gauss function.Aiming at the monotonous problem of interaction technology in virtual environment,this paper studies the real-time action interaction method in virtual environment.In order to realize motion recognition in complex scenes,this paper combines sliding window with GMM-HMM and sets threshold to realize human motion segmentation.Real-time action interaction module in virtual environment is divided into two parts:off-line training and on-line recognition.Finally,the interactive system of Peking Opera action under the stage environment is completed.