Kalman Filter-based Algorithms for Estimating Depth from Image Sequences

Abstract

Using known camera motion to estimate depth from image sequences is important in robotics applications such as navigation and manipulation. For many applications, having an on-line, incremental estimate of depth is important. To permit the blending of new measurements with old estimates, it is essential that the representation include not only the current depth estimate, but also an estimate of the current uncertainty. Kalman filtering provides the needed framework to integrate new measurements and reduce the uncertainty over time. Previous applications of Kalman filtering to depth from motion have been limited to the estimation of depth at the location of a sparse set of features. In this paper, we introduce a new pixel-based (iconic) algorithm that estimates depth from an image sequence and incrementally refines its estimate over time. We also present a feature-based version of the algorithm which is used for comparison. We compare the performance of both approaches mathematically, with quantitative experiments using images of a flat scene, and with qualitative experiments using images of a realistic outdoor scene model. The results show that the method is an effective way to extract depth from lateral camera translations. Our approach can be extended to incorporate general motion and to integrate other sources of information such as stereo. The algorithms which we have developed, which combine Kalman filtering with iconic descriptions of depth, can thus serve as a useful and general framework for low-level dynamic vision.