A Fast and Accurate Stereo Matching Algorithm Based on Epipolar lLne Restriction

A novel algorithm is proposed for solving the overlapped marker matching issues and satisfying speed requirement in a 3D video motion capture system. Firstly, the mean markers' longitudinal coordination difference in two parallel cameras is calculated to reduce 3D matching error. Secondly, the epipolar geometry restriction is used to make the corresponding marker be restricted in a line. Finally, integrating the mean vertical coordination difference and the epipolar geometry restriction, the searching region of the corresponding marker is reduced to the neighborhood of one estimated point. Consequently, the matching speed is improved greatly due to the reduced searching region. The clustering algorithm and the minimum of the weighted distance are used to obtain the accurate matching points. Several experiments results verify the effectiveness of our algorithm.