Abstract:

The motion estimation based on image is a basic task in many applications of computer vision. The main goal of it is to estimate the motion of scenes and objects as accurate as possible. In the optical flow estimation, outliers may lead to undesirable results and greatly reduce the motion estimation precision. In order to solve these problems, an improved robust multi-scale algorithm of hierarchical motion estimation is proposed. In this algorithm, Tukey＇s bi-weight function is used to automatically adjust the weights of data with different residual errors and to remove the data with excessive residual errors, and a multi-scale pyramid is employed to accurately estimate the motion vector by an iteration gradually from coarseness to fine. Experimental results show that the proposed algorithm with strong robustness effectively solves the problem of outliers caused by occlusive background and discontinuous motion and greatly improves the precision of motion estimation.

Key words: motion estimation, optical flow, multiple scale, robustness