Region-weighted gradient and motion-aware camera exposure control for robust visual odometry

Abstract

Visual odometry (VO) often suffers from performance degradation in high dynamic range (HDR) scenarios due to uneven illumination conditions, adversely affecting localization accuracy. To address this challenge, we propose a camera exposure control algorithm designed to enhance the robustness of visual odometry by adaptively adjusting the camera’s shutter speed and gain parameters to obtain high-quality images. Specifically, the algorithm employs a gradient-weighted sum of local image regions as a quality evaluation metric, and iteratively updates exposure parameters by computing the first-order derivative of this metric with respect to the exposure settings. Moreover, scene irradiance information is incorporated to determine an appropriate update step size. To further mitigate motion blur, we constrain the camera’s maximum shutter time based on feedback from the visual odometry system. the efficacy of the proposed approach is validated through experiments conducted in both HDR and fast-motion scenarios. Additionally, we integrate the exposure control strategy into two representative VO frameworks, ORB SLAM3 and VINS MONO, and evaluate their performance in both indoor and outdoor environments. The experimental results demonstrate that our exposure control method significantly improves the overall accuracy and robustness of visual odometry systems.