EFFICIENT MULTIPLE INDEPENDENT MOTION SEGMENTATION FROM AN ACTIVE PLATFORM BY UTILIZING MODIFIED RANSAC

摘要

In this paper, an efficient approach to segmentation of different independent motion areas from a moving platform is described. This approach is implemented on a stereo vision system, depth information could be computed by matching feature points between stereo images. For two consecutive frames, ego-motion is estimated from the optical flows, including depth information belonging to the background, which has a larger space distribution comparing to those of independent moving objects. In order to distinguish different motion areas, we proposed a modified version of RANSAC mechanism, which could handle the problem of multiple model extraction in a noisy environment.