Enhancing Multicamera-Based 3-D Detection in Low-Light Driving Scenarios With Self-Supervised Illumination Estimation

Vision-based multisensor system for bird’s eye view (BEV) 3-D perception is gaining attention as an alternative to high-cost multi-LiDAR systems and has achieved notable success. However, there is a significant safety concern for future image-based BEV autonomous driving in low-light conditions (such as nighttime) while the limited research on BEV detectors for these scenes. In this article, we attempt to enhance low-light BEV perception with illumination-guided feature fusion. We propose RetBEV, which uses illumination information generated based on the retinex theory to enhance the model’s robustness in varying lighting conditions. Additionally, to address the illumination estimation discontinuity from multiview images that can adversely affect detection, we propose the multiview self-balancing retinex (MVB-retinex), a self-supervised learning mechanism which balances illumination estimation by leveraging overlapping regions between adjacent images. Notably, RetBEV is a plug-and-play module that can be applied to many image-based BEV detector methods and does not require any additional ground truth (GT) supervision. We conduct extensive experiments on the nuScenes dataset, validating our algorithm in nighttime and daytime scenes. Compared to the baseline, our algorithm achieves a 2.34% increase in mean average precision (mAP) on the validation set with minimal computational cost, especially showing a 3.60% improvement in nighttime scene. The experiments demonstrate that our RetBEV effectively improves detection performance in low-light conditions and enhances performance under normal illumination, indicating increased robustness of the BEV detector.