Research on traffic sign recognition based on the YOLOv8 algorithm

Abstract

In the domain of autonomous and assisted driving technologies, the precise detection and identification of road traffic signs are of paramount importance within the perception layer of such systems. To counteract this, our approach involves the selective identification of categories with more than 50 instances, using the associated street scene images as a foundational dataset. Through deliberate data augmentation techniques, we effectively combat the challenges posed by data scarcity and class imbalance. Subsequently, the YOLOv8n neural network algorithm and Slicing Aided Hyper Inference are employed to construct the traffic sign recognition model. The efficacy of the proposed method is tested on images acquired through in-vehicle dashcams and direct on-site photography. The empirical outcomes demonstrate that our YOLOv8n-based traffic sign intelligent recognition algorithm attains an exceptional accuracy rate exceeding 93%. This method significantly augments the accuracy and stability of driving assistance systems, thereby substantially improving vehicular safety.