Research on Multi-Feature Constrained Adaptive Visual SLAM Localization Method Based on Decision Tree
Fei Liu, Y. Ma, Jian Wang, Siqi Wu, Yiwen Zhao, Zhitong Wang · IEEE Access · 2025
This study presents a multi-feature constrained adaptive visual SLAM(Simultaneous Localization And Mapping) localization method based on the decision tree to address the problem of low robustness in visual positioning within complex illumination-interfered scenes, such as glare and shadow. This method builds upon the ORB-SLAM2 algorithm. The system detects illumination interference images by incorporating the C4.5 decision tree method into the front end of ORB-SLAM2. Five indices, including image mean value, standard deviation, average gradient, average brightness, and average saturation, are chosen as the classification criteria for the decision tree. Subsequently, the saturation of the detected illumination interference image is enhanced to yield an image more suitable for ORB feature extraction. This research utilizes the EuRoC and KITTI datasets for validation. Experimental results demonstrate that the localization accuracy of the proposed method surpasses that of the unmodified ORB-SLAM2 algorithm and other state-of-the-art visual SLAM approaches, thereby significantly enhancing the robustness of visual localization in indoor and outdoor complex illumination conditions.