Direct Estimation of Structure and Motion from Multiple Frames.

摘要

This paper presents a method for the estimation of scene structure and camera motion from a sequence of images. This approach achieves fundamental improvements over existing methods in the following respects: (1) No computation of optical flow or feature correspondence is required. Structure and motion are obtained directly from gradients of image brightness; (2) The method processes image sequences of arbitrary length and exploits the inherent redundancy for a significant reduction in error over time; and (3) No restrictions or assumptions are made about the camera motion or the the surface structure. Both quantities are fully recovered using this method. Our method combines the direct motion vision approach of Horn, Weldon and Negahdaripour with the theory of recursive estimation. Most importantly: it really works and we show results on a variety of real image sequences. Keywords: Structure estimation; Kalman filters; Artificial intelligence. (kr)