Vehicle Target Detection on Campus Roads Based on YOLOv8

During the epoch characterized by the rapid advancement of artificial intelligence technology, driverless technology is changing day by day, and in this field, accurately identifying vehicles and pedestrians on the road has become one of the key tasks to promote the development of driverless technology. In this paper, the team collected vehicle, pedestrian and non-motorised vehicle datasets on campus roads, used five YOLOv8 network models with different depths for training, and analysed the training results in detail. Through the comparative analysis of diverse network models' performance, this paper concludes that YOLOv8x has the best overall performance, with the highest detection mAP values among all the models, especially the mAP50-95 is at least 1.71% higher than the other models, which highlights the excellent performance of the large model. In the context of campus road vehicle detection, YOLOv8s demonstrates commendable performance, exhibiting a detection accuracy of 91.26% and a mAP (mean Average Precision) value of 95.45%. Although the detection accuracy of YOLOv8s is not the highest among all models, YOLOv8s is more suitable for campus road conditions, considering that YOLOv8s has a faster training speed and lower time cost as a small model. The main objective of this paper is to find the YOLOv8 network model with excellent performance to accurately identify the driving conditions on campus roads so as to improve the safety and reliability of vehicle travelling on campus.