Stereo Video Depth Estimation Based on Disparity Map Propagation

Abstract

Stereo matching algorithm is used to estimating the depth value correspond with two frames taken from left and right cameras. Due to the movement of the objects, the output of depth map will easily vibrate without confidence. Moreover, surrounding information which cover by other objects may cause significant impacts and problems for the users. Therefore, the stability of depth maps is a crucial point for precisely outcomes. This paper proposes a new notion that considers the motion of objects, the frames in different time, and the relationship between left and right frames to propagate a new depth map. The regions of each objects are estimated using quadtree and contrast context histogram. The experimental results show the proposed method surpass conventional stereo matching methods.