Stereo Matching Based on Improved Matching Cost Calculation and Weighted Guided Filtering

Abstract

Aiming at the problem that the existing local stereo matching algorithm has low matching accuracy in weak texture, disparity discontinuity and occlusion regions, an improved algorithm based on matching cost calculation and weighted guided filtering is proposed. The algorithm first improves the traditional gradient cost (GRAD) and Census transform, normalizes and fuses these two matching costs to form a new matching cost, then proposes a weighted guided filter based on the Kirsch operator and aggregates the matching cost, finally, the method of the winner-takes-all (WTA) is used to complete the disparity calculation, and we use the method of left and right disparity consistency and the quadratic curve interpolation to complete the disparity optimization and obtain the final disparity map. A large number of experiments prove that the proposed stereo matching algorithm has an average mismatch rate of about 5.45% relative to the standard disparity map on the test platform of Middlebury. Compared with most algorithms, proposed algorithm achieves a good matching effect.