Precise Object Detection in Challenging Foggy Driving Conditions With Deep Learning

Abstract Recent advancements in deep learning have led to significant improvements in driving perception. However, perceiving the environment accurately during risky situations like fog remains a major challenge. Existing methods for detecting foggy atmospheres struggle with both recognition speed and accuracy. In this study, we propose an object discovery network specifically designed for driving in foggy environments, utilizing a modified version of YOLOv5. Our approach involves building a mist detection network based on a modified ResNet-50 model. To address the lack of features in foggy scenes, we introduce a new feature enhancement module (FEM) and employ the skip attention mechanism. These additions allow the recognition network to focus on the most relevant features in foggy scenes, improving overall performance. Experimental results demonstrate that our suggested foggy multi-target recognition network outperforms the original YOLOv5 in terms of both accuracy and speed. When evaluated on the MS COCO vehicle dataset, our network achieves an impressive accuracy of 81.8%, showcasing its superiority over the original YOLOv5. This advancement in foggy environment perception holds great promise for enhancing driving safety and efficiency. Keywords: accuracy, deep learning, driving perception, foggy environment, mist detection, object discovery, recognition speed, ResNet-50, skip attention, YOLOv5.