ELL-ORB-SLAM: Improved Robust Monocular Visual SLAM Algorithm Based on Image Enhancement and Dynamic FAST Threshold for Low-light Environment

The accuracy of visual localization estimation heavily relies on the quality of input images and feature point extraction, with variations in illumination significantly impacting matching results under challenging conditions. To address this, we propose an enhanced ORB-SLAM algorithm (ELL-ORB-SLAM) based on dynamic adaptive FAST thresholds and image enhancement techniques, including illumination map estimation, improved Truncated Adaptive Gamma Correction, and CLAHE, to improve performance in low-light conditions and refine feature point extraction. Experiments conducted on the EuRoC, KITTI, and ETH3D datasets demonstrate that ELL-ORB-SLAM significantly outperforms the state-of-the-art ORB-SLAM3, particularly in challenging environments. Additionally, real-world tests using a monocular camera showcase the system's robustness and versatility across diverse settings.