A Nighttime Vehicle Detection Method Based on YOLO v3
Yan Miao, Fu Liu, Tao Hou, Lu Liu, Yun Liu
Abstract
For the past several decades, the number of vehicles has been increased for many times, leading to a lot of traffic issues. The nighttime accident rate is much higher than that in daytime because of the weak light and the uneven distribution of nighttime brightness. Thus, in this paper an effective nighttime vehicle detection approach is designed. First, the original nighttime images were enhanced by an optimal MSR algorithm. Then, a pretrained YOLO v3 network was selected and fine-tuned by the enhanced images. Finally, the detection network was used to detect vehicles from the nighttime images and outperformed two widely used object detection methods, namely the Faster R-CNN and SSD, on the precision and detection efficiency. The average precision of the proposed method reaches 93.66%, which is 6.14% and 3.21% higher than that of the Faster R-CNN and SSD, respectively.