融合HBST和光照不变特征的回环检测算法

打开文本图片集

中图分类号：TP391.41 文献标识码：A

Abstract： To enhance the accuracy and robustness of loop closure detection in autonomous driving and robot navigation，this paper proposes a loop closure detection algorithm that integrates the Hierarchical Binary Search Tree（HBST） and ilumination invariant features，namely the IIFHBST（Ilumination Invariant Features-Hierarchical Binary Search Tree） algorithm. It combines the efficient search capability of HBST with the anti-interference ability of an ilumination-invariant bag-of-words model. An improved HBST model is constructed for rapid candidate frame screening，and an illumination-invariant bag-of-words model is designed to ensure the robustness of feature matching. Final loop closure candidates are determined through a weighted fusion strategy. The IIF-HBST algorithm is evaluated on the KITTIo6，KITTI07，and Malaga Urban DatasetO7 datasets，achieving F1 -scores of O.936，O.934，and 0.810，respectively，which significantly outperforms the traditional bag-of-words model （Distributed Bag-of-Words version 2， DBoW2）and the HBST algorithm. The results indicate that the IIF-HBST algorithm effectively balances precision and recall in environments with severe illumination changes and complex dynamic conditions.

Keywords： visual SLAM; loop closure detection； binary feature matching; illumination robustness；HBST

回环检测是同时定位与建图（Simultaneous Localization and Mapping，SLAM）系统的关键组成部分，能够识别机器人或自动驾驶车辆是否回到了之前访问过的位置[1]。（剩余12519字）