重庆理工大学学报2025，Vol.39Issue(17)：38-44,7.DOI:10.3969/j.issn.1674-8425(z).2025.09.005
交叉注意力驱动的室外双目视觉SLAM稠密建图算法研究
Research on outdoor dense mapping in binocular visual SLAM driven by cross-attention
摘要
Abstract
Traditional visual SLAM algorithms rely on sparse reconstruction,which is insufficient for high-precision environmental perception required for autonomous navigation and obstacle avoidance.This paper proposes a dense mapping model with cross-attention mechanism integrated into the traditional ORB-SLAM3 framework to achieve outdoor dense map construction.The model generates disparity maps,produces color-depth point clouds,and builds high-precision 3D dense maps to meet the requirements of autonomous navigation and obstacle avoidance.Experimental results show the algorithm keeps the errors within 0.5 meters on 90％dense point cloud on KITTI dataset and in real-world outdoor experiments,demonstrating high mapping accuracy and addressing the insufficient environmental information in traditional visual SLAM systems.关键词
双目视觉SLAM/立体匹配/稠密建图/三维重建Key words
binocular visual SLAM/stereo matching/dense mapping/3D reconstruction分类
信息技术与安全科学引用本文复制引用
王立勇,刘毅政,苏清华,宋越,谢智昊..交叉注意力驱动的室外双目视觉SLAM稠密建图算法研究[J].重庆理工大学学报,2025,39(17):38-44,7.基金项目
国家基础加强计划基金项目(2021JCJQJJ0022) （2021JCJQJJ0022）
