On integrating depth from motion and stereo

This paper presents a method to accurately and efficiently extract depth from a sequence of images. The method integrates two techniques of depth extraction, namely, spatial and temporal gradient analysis, and stereo processing to enhance the reliability of the depth extraction. Spatial and temporal gradient analysis has been shown to provide depth in a highly efficient manner. However, due to the small image displacement used in this algorithm, depth with only limited precision can be computed. To enhance the precision, a stereo based matching is performed on the first frame and the last frame of the monocular sequence used in spatial and temporal gradient analysis. The stereo matching is limited to a small region, based on the results of the spatial and temporal gradient analysis. The performance of the approach is demonstrated through experiments with real scenes