A New Structure of Stereo Algorithm Using Pixel Based Differences and Weighted Median Filter

This paper proposed a new algorithm for stereo vision system to obtain depth map or disparity map. The proposed stereo vision algorithm consists of three stages, matching cost computation, disparity optimization and disparity refinement. The first stage starts with matching cost computation, where pixel based differences methods are used. The matching methods are the combination of Absolute Difference (AD) and Gradient Matching (GM). Next, the second stage; disparity optimization utilizes Winner-Takes-All (WTA) technique to normalize the disparity values of each pixel of the image. Finally, for disparity refinement stage, weighted median (WM) filter is added to reduce and smother the noise on the disparity map