A Review of One-Stage Detection Algorithms in Autonomous Driving
Jiaqi Fan, Tianjiao Huo, Xin Li
Abstract
With the deep research on autonomous driving, the target detection algorithms based on 2D images have become a hot topic in recent years. In this paper, we mainly study the six mainstream deep learning detection algorithms, namely YOLO, YOLOv2, YOLOv3, YOLOv4, SSD and RetinaNet, which are used as the representative algorithms of the one-stage object detection methods. This paper is a review of the working principles of these six detection algorithms in deep learning. Firstly, this paper explains and compares the detection network structure, loss function and improvements of these six detection algorithms in detail according to the order in which they were presented. Especially the YOLOv3, YOLOv4, SSD and RetinaNet algorithms, which are the most common used algorithms at presented with a very high detection speed and detection precision for real-time detection of the autonomous driving. Secondly, by comparing the detection accuracy and speed of the YOLOv3, YOLOv4, SSD and RetinaNet algorithms, this paper gives their respective suitable application scenarios.