The research of vehicle monocular ranging based on YOlOv5

Target detection and distance estimation for autonomous driving have been making remarkable strides in recent years. In autonomous driving, target detection mainly refers to the identification of vehicles, pedestrians, traffic signs, etc. Nowadays, target detection technology has gradually matured, especially with the emergence of Faster R-CNN and YOLO, which has greatly improved the accuracy and efficiency of target detection. Not only to detect objects, we want to know more about spatial and distance information between the vehicle and the objects, however, the complexity of road conditions challenges the ranging methods. When the road is crowded and there are many vehicles, which will block each other, reduces the ranging performance, especially for laser ranging and ultrasonic ranging. In this paper, a monocular visual ranging method based on YOLOv5 is proposed, which is divided into two steps. Stage one detects vehicles and pedestrians, predicts a bounding box for each detected target. The second stage determines the position of the object with the help of the bounding boxes and accurately extracts the feature points, then computes the specific distance base on the geometric relationship. In this paper, we considered various road conditions to improve performance in real-world road scenes and the distance measurement method in this paper is proved to be robust under different road conditions. Since only one camera is required, the equipment cost is greatly reduced. What's more, our model can be combined with lane detection to realize target recognition, object ranging and lane detection at the same time.