Research on Vehicle and Pedestrian Detection and Recognition Algorithm Based on YOLO in Adverse Weather Conditions

To address the challenges of vehicle and pedestrian detection in adverse weather conditions, this paper introduces an enhanced MRLNet (Multi-scale Residual Light Network) algorithm based on the YOLOv8 model. This algorithm incorporates multi-scale feature fusion and image enhancement techniques to bolster the perception of small targets. Additionally, it integrates the Fine-Grained Channel Attention module to focus on critical feature information, and employs the MPDIoU loss function to refine bounding box regression. Experimental results demonstrate that, without pre-trained weights, the algorithm achieves an average accuracy of 70.8% on a self-compiled severe weather dataset, marking a 1.6% improvement over the original YOLOv8 and outperforming other algorithms.