Enhanced Local Disparity Map Algorithm Segment-Side Window-based Cost Aggregation and Refinement

Abstract

Accurate disparity map estimation is crucial for applications such as 3D reconstruction, autonomous navigation, and object detection. Local window-based cost aggregation often suffers from edge fattening and texture inconsistency. This paper introduces a Segment-Side Window-based (SSW) stereo matching algorithm that combines Truncated Absolute Difference (TAD), Gradient Magnitude (GM), and Census Transform (CT) to build a robust cost volume. In the proposed approach, SLIC superpixels guide adaptive aggregation, while Side Window Filtering (SWF) preserves edges and enhances texture consistency. Winner-Takes-All optimization and SWF refinement further improve depth accuracy. On the Middlebury dataset, the proposed method achieves 13.3% (Nonocc) and 21.8% (All) bad pixel errors, outperforming BF, GF, iGF, and MF in both edge preservation and texture robustness.