Maximum Disparity Threshold Estimation for Stereo Imaging Systems via Variogram Analysis

摘要

Estimation of the highest disparity value is very important for proper operation of any stereo matching algorithm. Choice of a too big or too small disparity range may cause severe errors in an output disparity map. A guess of this parameter by human is the method which is used very often during experiments, however it is not suitable in real time or autonomous vision systems. In the same way, a usage of the constant disparity range gives often poor results. The paper describes a innovative method of the automatic estimation of a maximum disparity for the real time stereo imaging systems. After short review of the most popular methods of the maximum disparity estimation, we present our method based on the variogram analysis. Finally the experimental results for the real images are also presented.