Real-Time Vehicle and Distance Detection Based on Improved Yolo v5 Network
Tianhao Wu, Tongwen Wang, Yaqi Liu
Abstract
Because there are various unsafe factors on the road, the testing of the virtual environment is an important part of the automatic driving technology. This paper presents a CARLA vehicle and its distance detection system in a virtual environment. Based on the existing Yolo v5s neural network structure, this paper proposes a new neural network structure Yolo v5-Ghost. Adjusted the network layer structure of Yolo v5s. The computational complexity is reduced, and the proposed neural network structure is more suitable for embedded devices. After testing the new network structure, the detection accuracy of Yolo v5s is 83.36%mAP(mean Average Precision), the detection speed is 28.57FPS (Frames Per Second), and the detection accuracy of Yolo v5-Ghost is 80.76%mAP, the detection speed is 47.62FPS. The paper also detects the vehicle distance based on the pictures obtained by the monocular camera in the CARLA virtual environment. The detected distance error is about 5% on average.