Nighttime Road Surface Detection Using Drivable Area Extraction Based on Night Illumination Enhancement

Active monitoring of road surface plays a crucial role in improving driving safety. Currently, most road surface detection methods are based on good illumination conditions during the daytime, with little discussion on nighttime scenarios. Therefore, this paper focuses on road surface detection at night based on visible light images. First, the paper uses an illumination enhancement algorithm to improve images taken under low-light conditions. Then, a two-stage method is employed: first, a novel network with multiple attention modules, MA-ResNet (Multiple Attention Module ResNet), is proposed for horizon detection to extract the region of interest. Next, patches are cropped based on the position of the driving lane and used to fuse classification results using modified EfficientN etV2. Finally, through comparison and ablation experiments, the proposed framework is shown to achieve outstanding detection performance in nighttime scenarios.