A Comprehensive Review of Stereo Matching Algorithms for Depth Map Application

Accurate depth map estimation from stereo images plays a central role in many computer vision applications, including autonomous navigation, robotic perception, and 3D surface reconstruction. Despite extensive research, traditional stereo matching methods continue to face challenges in weakly textured areas, at depth discontinuities, and within occluded regions, which reduces their reliability when applied in complex and unstructured environments. This study provides a structured review of stereo matching techniques, outlining the transition from purely local and global approaches to more advanced hybrid and segment-based frameworks. A particular emphasis is placed on the role of multi-cost matching functions, which integrate complementary descriptors to improve robustness against texture variations. In addition, edge-preserving cost aggregation strategies, such as segment-based and side-window filtering, are highlighted for their effectiveness in maintaining object boundaries while suppressing noise. To further improve performance, disparity optimization methods such as semi-global matching and adaptive refinement are examined for their ability to strike a balance between computational efficiency and accuracy. Experimental analysis using benchmark datasets demonstrates that combining multi-cost descriptors with adaptive filtering enhances both the consistency and quality of reconstructed depth maps. Overall, the findings suggest that segment-aware and texture-robust stereo matching approaches offer strong potential for enabling scalable and real-time stereo vision systems suited to practical deployment.