Performance Evaluation of YOLOv5, YOLOv7, and YOLOv8 Models in Traffic Sign Detection
Fatma Nur Ortataş, Mahır Kaya
Abstract
Autonomous vehicle technology has been developing rapidly in recent years. The main reason for this is the high rate of fatal traffic accidents caused by drivers. The main reason for the emergence of driverless vehicle technology is to reduce traffic accident rates by creating a safe driving environment. With this technology, vehicles are expected to fully recognise the traffic environments they are in and make the correct movement in accordance with the rules. In this context, the detection of traffic signs is a very important area for this technology. In this study, YoloV5, YoloV7 and YoloV8 models were trained for the same hyperparameter values with a dataset containing a total of 4650 photographs of 15 different traffic signs. Prediction, recall, mAP, complexity matrix and F1 score values were analysed. As a result of the study, it was observed that the YoloV8 model gave better results. According to the same dataset and experimental conditions, the YoloV8 architecture provided 14% and 13% mAP50-95 improvement over YoloV5 and YoloV7 architectures,respectively.