MSFFA-YOLO Network: Multiclass Object Detection for Traffic Investigations in Foggy Weather
Qiang Zhang, Xiaojian Hu
Abstract
Despite significant progress in vision-based detection methods, the task of detecting traffic objects in foggy weather remains challenging. The presence of fog reduces visibility, which in turn affects the information of traffic objects in videos. However, accurate information regarding the localization and classification of traffic objects is crucial for certain traffic investigations. In this paper, we focus on presenting a multi-class object detection method, namely MSFFA-YOLO network, that can be trained and jointly achieve three tasks: visibility enhancement, object classification, and object localization. In the network, we employ the enhanced YOLOv7 as a detection subnet, which is responsible for learning to locate and classify objects. In the restoration subnet, the multi-scale feature fusion attention (MSFFA) structure is presented for visibility enhancement. The experimental results on the synthetic foggy datasets show that the presented MSFFA-YOLO can achieve 64.6 percent accuracy on the FC005 dataset, 67.3 percent accuracy on the FC01 dataset, and 65.7 percent accuracy on the FC02 dataset. When evaluated on the natural foggy datasets, the presented MSFFA-YOLO can achieve 84.7 percent accuracy on the RTTS dataset and 84.1 percent accuracy on the RW dataset, indicating its ability to accurately detect multi-class traffic objects in real and foggy weather. And the experimental results show that the presented MSFFA-YOLO can achieve the efficiency of 37 FPS. Lastly, the experimental results demonstrate the excellent performance of our presented method for object localization and classification in foggy weather. And when detecting concealed traffic objects in foggy weather, our presented method exhibits superior accuracy. These results substantiate the applicability of our presented method for traffic investigations in foggy weather.