抄録
A novel algorithm for automatic human body model initialization and motion tracking based on voxel data is proposed. Model initialization is to calculate the dimension of model and estimate the posture parameters using one single frame. In this paper, the process to initialize skeletal model is based on the novel idea of slicing voxel data along z axis and using some basic algorithms of image processing to extract feature points. In our algorithm, tracking is considered as a global optimization problem, and we use a novel evolutionary algorithm named Probability Evolutionary Algorithm (PEA) to solve it. The whole processes are totally automatic and experiments show this new algorithm is visually accurate and very robust to noise.
