YOLOv8-LLE: An Improved Algorithm for Object Detection of Autonomous Driving Road Scenes in Low-Light Environments

In low-light environments, insufficient illumination greatly reduces the vehicle’s vision system’s ability to detect and recognize vehicles and pedestrians. To overcome this problem, this research proposes an enhanced algorithm, YOLOv8-LLE, specifically designed for object detection in low-light environments. The algorithm firstly introduces Retinexformer for low light image enhancement, then replaces the Spatial Pyramid Pooling-Fast (SPPF) with a focal modulation network FocalNets to suppress the noise more efficiently, and finally introduces the neck of the Efficient Multi-Scale Attention to retain more information of features while effectively reducing the number of network parameters. Due to the lack of low-light scene datasets for autonomous driving, this paper also proposes a low-light image dataset construction method. Experimental results show that the constructed low-light image dataset can better restore the image features in real low-light environment. Compared with the original model, the proposed YOLOv8-LLE improves 2.6%, 6%, 3.1%, and 3.7% in Precision, recall, mAP@0.5, and mAP@0.5:0.95, respectively. Real-world vehicle experiments further validate the superior performance of YOLOv8-LLE in detecting vehicles and pedestrians in actual low-light environments.