一种鲁棒的基于互信息的实时立体匹配算法

立体匹配算法依然是目前计算机视觉领域中的一个重要的研究热点.现有的绝大多数匹配算法都假定已满足相似性约束条件.然而在野外环境下,光照不均、场景为非朗伯表面、像机间差异都会使该约束条件无法满足,从而导致匹配失败.本文针对这种情况,并从实时应用的角度出发,提出一种鲁棒的基于互信息的实时立体匹配算法.该方法在相关匹配算法中引入互信息概念,并将其转换为相关算法可用的求和形式,通过迭代方法修正互信息值.然后,建立了三种亮度变化模型来验证该算法的有效性.实验结果表明,该算法能很好地抑制立体图像对间的亮度差异,具有很好的实时性与鲁棒性.

A robust algorithm for real time stereo match based on mutual information

Stereo match algorithm is still an important and hot research topic in the computer vision domain. Most of the existing match algorithms are subject to similarity constraint. However, the non-uniform illumination, non-lambertian reflection in the off-road environments, and different camera gains or biases, all can violate the similarity constraint, and lead to match failure. Aiming at this case, a robust and real-time stereo match algorithm based on mutual information is proposed in view of practical applications. Mutual information is incorporated in the correlation based stereo match algorithm, and is transformed to summation form that can be used by the match algorithm. And, the mutual information is amended iteratively. At last, three luminance variance models are built to verify the validity of the proposed algorithm. Experiment results show this real-time algorithm is very robust, and can suppress the luminance difference of stereopsis successfully.