Hierarchical human motion compression with constraints on frames

Abstract

This paper presents a compression method for motion data with special characteristics which can be indicated by motion behavior or specified by user with constraint. Motion capture systems have been widely used to model motion behavior in feature film production, action games and virtual environments on network. In general, the motion data consist of three coordinates or some rotation angles in each frame. For storing motion data and real-time transmission, an uncompressed raw data is often unacceptable. Our method combines wavelet transform and kinematics to implement an efficient motion data compression. Controlling the distortion yielded by the quantization, we enable user to specify constrains of motion. The max shift ROI method is adopted to sign the constraint frames. We further apply an adaptive quantization to establish the optimal position of the joint. Our method achieves 5-20% compression without any visual artifacts. The experiment result shows the validity of our method.