Spatiotemporal 3D motion vector filtering method for robust visual odometry

Abstract

Most of the previous visual odometry methods cannot deal with a large independently moving object that takes up over 50% of the image area. To overcome this problem, the spatiotemporal filter is incorporated into the RANSAC method to filter out false match that occurrs by a large independently moving object. This spatiotemporal filter uses the current and previous motion vector's length and direction. Experimental results demonstrate that the proposed method effectively rejects the motion vectors generated from large independently moving objects and improves the visual odometry accuracy.