Optimization of Wide-Baseline Stereo Matching Based on Multi-View Geometry

Stereo vision plays a crucial role in scientific expeditions and facility maintenance. However, fixed-baseline stereo vision lacks accuracy for long-distance measurements, while wide-baseline stereo vision encounters occlusion issues at short distances. In this study, a wide-baseline stereo matching based on multi-view geometric constraints is proposed. Initially, the camera poses of multiple wide-baseline images are estimated, followed by image rectification and depth estimation. Subsequently, the disparity searching is constrained using geometric relationships among the views, resulting in more precise depth maps. Experiments were conducted to verify the superiority of the proposed method. The results indicate that the proposed method achieves an RMSE as low as 0.15 meters, even with invalid depth points. Compared to traditional stereo vision methods, accuracy is improved by approximately 35%, and robustness is significantly enhanced. In the aircraft runway scenario, overall depth map accuracy is higher, and occluded areas are significantly reduced.