Light adaptive visual inertial SLAM: a dynamic optimization framework for low light environments

Since SLAM systems are prone to localization failure caused by insufficient feature point extraction in low- light environments, this paper proposes a dynamic optimization framework based on ORB-SLAM3, aiming to address the robustness limitations of SLAM systems under dark lighting conditions. This paper implements two key contributions: one is real-time low-light enhancement frontend: Building upon Contrast Limited Adaptive Histogram Equalization (CLAHE) by improving the clip limit parameter determination through incorporation of image standard deviation, enabling adaptive parameter adjustment. This resolves the ineffectiveness of fixed-parameter CLAHE optimization; the other one is dynamic sliding-window Bundle Adjustment (BA) optimization mechanism: Proposing a window size adaptation algorithm based on real-time feature point count, which dynamically balances optimization efficiency and stability by regulating bundle adjustment scope according to feature density. This addresses BA divergence caused by feature sparsity in low-light conditions. Experimental results demonstrate that the proposed algorithm significantly enhances robustness in low-light environments. Compared with ORB-SLAM3, the improved method increases the average accuracy by 19.89% in different lighting, and reduces the Root Mean Square Error (RMSE) of the Absolute Trajectory Error (APE) by 23.87% under dark and complex conditions.