TSC18 Convolutional Neural Network for Traffic Sign Classification

The classification of traffic signs holds significant importance in the realm of autonomous vehicles. The primary objective of our research is to effectively perform this classification to mitigate the occurrence of accidents and enhance the overall reliability of autonomous vehicles. This research proposed a TSC18 CNN model, using 18 layers only for the Traffic Sign Classification and performing the evaluation using the pre-trained Convolutional Neural Network (CNN) models. The TSC18 CNN model contains four convolutional layers, three max-pooling layers, one flattening layer, four fully connected layers, and batch normalization and dropout layers. The TSC18 CNN model is trained and validated using the Road Cracks dataset, which contains a traffic image collection of 43 classes. The training set includes 34,799 images that have been labeled, while the validation set contains 4,410 labeled images. The performance of the proposed CNN model is compared with the pre-trained CNN models, including EffecientNet, InceptionNet, and VGG-19, using the same dataset. The training and validation accuracy and losses are presented in graphical form for comparison. Also, the wellknown pre-trained CNN models, including EfficientNet, InceptionNet, and VGG- 19, were tested using the same dataset for the performance evaluation with the proposed model. The TSC18 model exhibited a test data accuracy of 99.21%. The performance of the proposed model is compared with the existing models using the GTSRB dataset. The proposed model performs well as compared to the pre-trained CNN models. The TSC18 CNN model can be used for road crack detection using road images.