International Conference on Computer and Knowledge Engineering

Home / 14th International Conference on Computer and Knowledge Engineering

AL-YOLO: Accurate and Lightweight Vehicle and Pedestrian Detector in Foggy Weather

Authors :

Behdad Sadeghian Pour¹ Hamidreza Mohammadi Jozani² Shahriar Baradaran Shokouhi³

1- School of Electrical Engineering, Iran University of Science and Technology 2- School of Electrical Engineering, Iran University of Science and Technology 3- School of Electrical Engineering, Iran University of Science and Technology

Keywords :

Lightweight Object Detection،Adverse Weather

Abstract :

Abstract—One of the most critical tasks in computer vision is object detection. While object detection networks have demon- strated high accuracy in normal weather conditions, they are not reliable in adverse weather. To address this issue, we have enhanced a smaller version of YOLOv5, known as YOLOv5s, to be compatible with challenging weather conditions and devices with limited resources. As a result, we modified the YOLOv5s architecture to extract significant features. We used a lightweight transformer network known as MobileViTv2 with an Inverted Residual Block, which is efficient in terms of computational resources. Furthermore, the YOLOv5 Neck has incorporated the C3RFE module to improve feature extraction efficiency in adverse weather. In order to assess the effectiveness of our suggested method, we conducted a thorough evaluation using a Foggy-Cityscape dataset. The results demonstrate that in comparison to the YOLOV5s, the algorithm has a 33% decrease in the number of model parameters and also increases by about 2% in mAP. Comparative analysis demonstrates the algorithm’s superiority and effectiveness. Our code and pretrained models are available at: https://github.com/BehdadSDP/AL-YOLO.