In foggy driving environments, the precision of object detection by autonomous driving systems is significantly diminished compared to clear weather conditions. To date, there is an abundance of work on object detection, yet research focusing on object detection in foggy scenarios for autonomous driving remains scarce. Firstly, due to the scarcity of publicly available datasets for vehicles in foggy conditions, this experiment constructs a foggy environment vehicle dataset. Subsequently, I integrated an enhancement module based on the NLD (Non-Local Image Dehazing) algorithm into the YOLOv10 model, yielding a new model capable of vehicle detection in foggy environments. Furthermore, I performed the same operation on various other YOLO models and conducted model evaluations. The results demonstrate that the new model, incorporating NLD, demonstrates a marked enhancement in object detection accuracy compared to the original models, enhancing the precision and robustness of vehicle detection in foggy conditions, thus offering greater application value in foggy autonomous driving scenarios. The foggy dataset for the experiment is available on GitHub: https://github.com/yiming676/Hazed-VehicleDataset.
YOLOv10-Based Object Detection Method with Optimized Defogging Algorithm for Autonomous Driving in Foggy Environments
29.11.2024
1772371 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
MFR-YOLOv10:Object detection in UAV-taken images based on multilayer feature reconstruction network
Elsevier | 2025
|