Depth map-based disparity estimation technique using multiview and depth camera

Abstract

We present a depth map-based disparity estimation algorithm using multi-view and depth camera system. When many objects are arranged in the 3D space with a long depth range, the disparity search range should be large enough in order to find all correspondences. In this case, traditional disparity estimation algorithms that use the fixed disparity search range often produce mismatches if there are pixels that have similar color distribution and similar textures along the epipolar line. In order to reduce the probability of mismatch and save computation time for the disparity estimation, we propose a novel depth map-based disparity estimation algorithm that uses a depth map captured by the depth camera for setting the disparity search range adaptively as well as for setting the mid-point of the disparity search range. The proposed algorithm first converts the depth map into disparities for the stereo image pair to be matched using calibrated camera parameters. Next, we set the disparity search range for each pixel based the converted disparity. Finally, we estimate a disparity for each pixel between stereo images. Simulation results with various test data sets demonstrated that the proposed algorithm has better performance in terms of the smoothness, global quality and computation time compared to the other algorithms.