Analysis of Patch Shapes in Dense Matching for Reducing Disparity Errors in UAV Stereo Images

초록

Traditional dense stereo matching for 3D reconstruction from UAV images have limitations in complex boundary and low-texture areas, leading to degraded accuracy. To address this, we propose a method that integrates multi-shape patches, each specialized for different directional characteristics, and estimates the optimal disparity using a voting-based support aggregation. Experiment results show that the proposed method reduced the end-point-error by up to 38.7% compared to using a single square patch and enabled successful disparity estimation in complex structures and low-texture areas. This demonstrates that the proposed method was a highly effective strategy for generating high fidelity 3D spatial information. © ACRS 2025.All rights reserved.

키워드