Skip to Main Content
An algorithm based on stereo vision is proposed to generate 3D point cloud of human face. A two-step matching strategy from sparse to dense is developed. Firstly, an improved seeds growing algorithm is utilized to acquire sparse matching set of high confidence. Secondly, using sparse matching set as guidance, piecewise dynamic programming is carried out and the dense matching is completed. Finally, calibration data and matching relationship are used to recover 3D information of human face. The proposed algorithm combines the advantages of local and global methods of stereo matching which significantly improves the matching accuracy and speed. The experimental results show that the proposed algorithm can produce smooth and dense 3D point cloud model of human face.