Abstract
A number of researches have focused on estimating accurate and edge-preserve disparity map from stereo videos for many years. From left and right images, most previous works apply stereo rectification in the pre-processing step and then estimate the disparity map with stereo matching methods. Many previous works on stereo matching made considerable progress to compute accurate disparity maps in the past few years. In this thesis, we propose a disparity map estimation system for stereo videos. In this work, we develop a novel optical flow based method to estimate accurate, edge-preserving and edge-aligned disparity map. The proposed algorithm improves the TV-L1 optical flow approach by incorporating edge detection, occlusion area detection, textureless region detection and guided filtering to alleviate the problem of inconsistency between the disparity map and color image around object boundary. Furthermore, we propose a temporal filter to improve the temporal consistency of the disparity maps computed from the stereo videos. Experimental results on various real videos are shown to demonstrate that the proposed system effectively improves the accuracy of the disparity maps estimation compared with the TV-L1 optical flow approach.