Long-Range Dense Mapping With Enhanced Accuracy via a Flying Variable-Baseline Stereo System

For unmanned aerial vehicle (UAV) swarms equipped with cameras operating in large-scale urban environments, long-range mapping is an effective approach to enhancing safe navigation. Conventional stereo vision systems are inherently limited by their compact structure and fixed baselines, restricting their effective sensing range. This article presents flying variable-baseline stereo (VB-stereo)—a collaborative stereo vision system that utilizes two coordinately flying UAVs to form a spatially flexible stereo configuration for long-range dense mapping. We first propose a collaborative variable-baseline stereo mapping (CVBSM) framework that integrates online variable-baseline (VB) estimation, cross-agent feature association, and sparse-to-dense exponential fitting to achieve long-range dense mapping. Building on this framework, we further analyze the optimal stereo baseline that balances geometric parallax and baseline estimation uncertainty to enhance mapping accuracy across different scene depths. Extensive real-world experiments demonstrate that our approach enables dense 3-D reconstruction up to 70 m, achieving relative errors between 2.3% and 9.6%. Notably, the optimal baseline length is shown to increase consistently with scene depth. This provides effective guidance for adaptive baseline selection, thereby enhancing reconstruction accuracy across targeted depth intervals. These results demonstrate the potential of VB collaboration for long-range UAV perception and open new directions for future research in aerial swarm mapping.Video: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https: //youtu.be/AfTm54kpcSo</uri>