Swin Detr: Pedestrian and vehicle detection under low light conditions based on Swin Detr

Currently, deep learning-based object detection algorithms can rapidly and accurately locate vehicles and pedestrians under normal conditions. However, in low-light environments, the detection accuracy of these algorithms significantly declines due to image blur and the interference of complex backgrounds. To address the issue of poor detection robustness in low-light conditions, this paper proposes Swin DETR to meet the demands of object detection algorithm research in such environments. Additionally, a comparison will be made with Fast RCNN and YOLO v7. This research holds significant importance and research value for expanding the application scenarios of autonomous vehicles.