Research On The Analytical Method Of Human Motion Capture Data

Recently, motion capture technology has become more mature, and is widely used in film and television production, game design, sports training, etc. With the rapid accumulation of motion capture data and the development of motion editing and synthesis, it becomes a trend to reuse existing capture data. However, because the capture data may be incomplete, motion capture sequence is too long and the capture data is redundant, the reuse of motion capture data is limited.This thesis focuses on motion data recovery, motion segmentation, motion compression and motion retrieval, and is designed to provide an effective means of data analysis to ensure implementation of the motion reuse technology. The specific contents and research are as follows:(1) Human motion capture data recovery using tensor completion. According to strong temporal and spatial correlation of local human motion and tensor completion theory, the method firstly represents motion feature in the form of 2D matrix. Then the matrix is divided, rearranged, sliced and trajectory-based 3rd-order low rank tensor is constructed. Finally tensor completion algorithm is used to achieve the recovery of missing data. Experimental results show the method outperforms the existing methods.(2) Human motion capture data segmentation based on PGA and PPGA. Due to the inherent nonlinearity and high-dimension of human motion capture data, the methods use nonlinear manifold learning algorithms to reduce nonlinear dimension of the feature data. Then human motion capture data is segmented based on PCA and PPCA segmentation methods. Experimental results show that the methods get better performances.(3) Human motion capture data compression based on Tucker decomposition. Because Tucker decomposition may retain unequal number of principal components on each dimension, the method uses fewer principal components in the high correlation dimension and greatly reduces the space of data storage and transmission bandwidth. Experimental results show that this method can achieve higher compression ratio than other methods.(4) Human motion retrieval based on Tucker decomposition. The method extracts low-dimensional kernel tensor which retains principal components of motion sequence by Tucker decomposition. Then the similarity between motions is measured after matricization process. Experimental results show that the retrieval method has better performance than other methods.