Robust and direct estimation of 3-D motion and scene depth from stereo image sequences

Abstract

In this paper, we propose a robust algorithm that estimates 3-D motion parameters and scene depth with calibrated stereo image sequences. In the direct method based on the brightness change constraint equation: we show that the previous linear or least-squares methods are prone to error about the structured noises which are dependent on the contents of image such as intensity edges, depth discontinuities and occluded areas. We also show that the proposed algorithm is robust to those error distributions. With respect to the hierarchical motion estimation framework, an effective method for determining the number of iterative refinements, which affects the accuracy of 3-D motion estimation, is also suggested by using the proposed robust estimation framework. Computer simulation with a synthetic image sequence and experiments with various real-image sequences show that the proposed algorithm has improved the estimation accuracy, compared with existing least-squares methods. (C) 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.