Performance Comparison of Road Traffic Sign Recognition System Based on CNN and Yolo v5

Exemplary drivers prioritize adherence to traffic rules, and implementing traffic sign detection technology effectively supports such behavior, enhancing road safety for all road users. This paper brings a comparative analysis of road traffic sign recognition systems based on Convolutional Neural Network (CNN), and You Only Look Once version 5 (Yolo v5) on the German Traffic Sign Recognition Benchmark (GTSRB) dataset. The proposed recognition system consists of four stages: 1) dataset annotation; 2) data pre-processing to lower noise; 3) sign detection and classification; 4) performance analysis. The developed models can detect and recognize forty-three classes of traffic signs covering all existing traffic signs under varying light and noise conditions as available in the dataset. Libraries and tools like Keras, TensorFlow, SKlearn, OpenCV2, Google Colab, etc., have been used to pre-process the data in Python. The processed images will serve as the CNN and Yolo v5 training datasets. According to experimental findings, the CNN model outperforms the YOLO v5 model in successfully identifying the majority of traffic signals with a precision of 96.231%. The proposed system would be helpful for the development of autonomous vehicles and provide effective driving assistance messages.