Application of YOLOv8 in Target Detection of Autonomous Vehicles

At present, autonomous driving technology is gradually replacing traditional car driving technology. Object detection is a critical part of autonomous systems. It is of great significance for driving safety, autonomy, improving traffic efficiency, emergency response, etc. However, radar technology for environmental perception and cameras and vehicle sensor networks for road perception have disadvantages such as high cost, susceptibility to weather light, and low resolution. In this paper, an improved autonomous target detection network based on YOLOv8 is proposed, which realizes the high efficiency and high precision detection of multi-scale, small targets and remote objects by introducing structural reparameterization technology, bidirectional pyramid structure network model and new detection pipeline structure model into the main part of YOLOv8 structure. Experiments show that the improved model can detect larger and smaller objects with an accuracy of 65% due to YOLOv8. It is expected to be further utilized in real-life applications, as well as to perform well in autonomous competitions for single-target and small targets detection, e.g., the Formula Student Autonomous China (FSAC).