Abstract: A visual SLAM method based on ORB-SLAM3 is proposed, which integrates brightness perception adaptive low-light enhancement and lightweight descriptor optimization, to address the problems of sharp reduction in feature point count, decreased descriptor discriminability, unstable matching, and tracking loss in complex imaging conditions such as low-light and weak texture. Firstly, to address the issues of excessive enhancement, noise amplification, and detail distortion caused by significant differences in brightness distribution and uneven local lighting in different scenes, a two-level decision adaptive low-light enhancement module is designed. Candidate enhancement results are generated through global brightness determination and exposure compensation. By adaptively switching between global and pixel-wise fusion modes according to the proportion of dark areas and brightness dispersion, the proposed method enhances the observability of dark regions while suppressing overexposure and maintaining geometric consistency. Secondly, to address the issue of insufficient robustness of traditional ORB binary descriptors in low-light conditions, a lightweight descriptor enhancement network is introduced to fuse the original descriptors with the geometric attributes of keypoints. Through self-attention enhancement and Transformer-based cross enhancement, contextual relationships between keypoints are modeled to improve descriptor discriminability and cross-frame consistency. On the ETH3D low-light sequence, the overall performance of the system was evaluated through quantitative error indicators and trajectory estimation results. The experimental results showed that the proposed method can significantly reduce trajectory estimation errors and improve system operation stability. Furthermore, the impact of different module combinations on system performance was analyzed through ablation experiments. Finally, the effectiveness and robustness of the proposed method in practical applications were further validated through mobile robot experiments in real low-light indoor scenes.