A robust area based disparity estimation technique for stereo vision applications

Abstract

A novel and efficient stereo matching algorithm based on robust disparity estimation even in the presence of occlusions is presented in this paper. The algorithm employs the Sum of Absolute Difference (SAD) approach for measure of similarity between two images. A pre-processing stage is applied to smoothen the sharp changes in pixel values at the object boundaries and also help in reducing photometric distortion and noise. Occlusions are removed by using Left-Right consistency constraint which will be explained further in this paper. After the calculation of disparity, image information is combined with the pixel disparities to obtain a cleaner disparity map. The developed algorithm was tested on benchmarked Middlebury data sets as well as acquired sample images. The results obtained are in line with the results of the ground truth images.