A Method of Human Motion Feature Extraction and Recognition Based on Motion Capture Device

Abstract

In recent years, motion capture devices have been widely used in 3D film special effects, animation generation, digital media and other virtual reality fields. The purpose of this paper is to achieve data acquisition, motion feature extraction and recognition of human motion by using motion capture device. First, we use the OptiTrack motion capture device and related data preprocessing methods to collect motion data, and realize the preview of joint point offset data. Then, a human motion feature representation method which combines the collected data with the key frame is designed. For the selection of key frames, we improve the frame subtraction algorithm by adding the second derivative calculation of reconstruction error to achieve the number of key frames automatic determination. In addition, in order to solve the problem of huge amount calculations and error existing in discretization of observation state, we use high-dimensional gaussian function to fit human motion data, and finally apply the method of Gaussian-Mixture Hidden Markov Model (GMM-HMM) for motion recognition. Experiments show that the method has achieved remarkable performances in human motion extraction and recognition.