Abstract: In traditional local matching algorithms, the window-based aggregation methods are usually used to obtain disparity map; also, the segmentation and plane fitting methods are used in disparity refinement. The performance of the algorithms is greatly dependent on the window size, and precision of segmentation and data fitting methods. In this paper, a stereo matching algorithm based bilateral filter is proposed, which is called double weighting aggregation. The fast cost aggregation method and hyperbola are employed to smooth the matching cost. The proposed algorithm is beneficial to improve the stereo matching accuracy since it is independent on the window size. The simulation results show that the proposed algorithm can obtain higher accuracy disparity map in low-texture and depth discontinuity regions. Besides, superior disparity map and 3D reconstruction results are also achieved in real stereo images.

Key words: communication, stereo matching, weighted aggregation, fast cost aggregation, disparity refinement, 3D reconstruction