Stereo Vision Based Depth Estimation Algorithm In Uncalibrated Rectification

In stereo vision application, the disparity between the stereo images allows depth estimation within a scene. Detecting conjugate pair in stereo images is a challenging problem known as the correspondence problem. In this paper, an overview of stereo vision is introduced as well as an efficient algorithm and a simple method in depth estimation. We use in this paper the well known matching algorithm, SURF, and we remove the outliers by applying geometric and epipolar constraints. Unlike most of the researchers, we calculate the depth without disparity map. We used the differences between the inliers points resulting from epipolar constraint to get the maximum and the minimum disparity. The depth can be calculated easily from the minimum and the maximum disparity by taking the mean value. To confirm the results we apply on all methods used in estimating the fundamental matrix. Experimental results show that our algorithm in depth estimation works quite robustly. Index Term— Stereo vision, depth estimation, SURF, disparity, correspondence problem, fundamental matrix