Effective Street View Traffic Sign Detection and Recognition Using Deep Learning

In a perfect world, traffic sign detection is highly accurate, but when used with autonomous cars, the accuracy of the detection is decreased by the complicated traffic situations. This work is aimed at detecting and recognizing the street view traffic signs (TSD-R) images from dataset using deep learning techniques, which is a complex problem that has been approached with different CNN models. The article discusses the evaluation of different CNN learning models for traffic sign detection systems. The models were trained and evaluated on the same dataset, ensuring consistency in data preparation and handling. All models were the same amount of epochs of training to guarantee fair comparison and give each model an equal opportunity to learn from the data. After evaluating various CNN learning models, including AlexNet, ResNet50, 34, Vgg16_bn, Vgg19_bn, and LeNet. The results showed that ResNet50 and LeNet outperformed the others for traffic sign detection. The result and comparison graph show that the LeNet achieved the highest accuracy 99.98% with 0.02 less error rate and minimum time taken to train the model.