Traffic Sign Recognition Using Deep Learning Neural Network and Spatial Transformer

• Published in: 2023 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) pp. 1 - 8
• Main Authors: Baruah, Arindam, Kumar, Rakesh, Gupta, Meenu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.05.2023
• Subjects: Convolution(Conv.); Fully-Connected(Fc); Max-Pooling(Mx-Pool); Model-A; Model-B; Rectified Linear Activation Unit(RELU); Stochastic Gradient Descent (SGD); Traffic Sign Detection (TSD); Traffic Sign Recognition (TSR)
• DOI: 10.1109/ACCAI58221.2023.10199560

Recently, numerous applications of Intelligent Transportation Systems (ITS) have gained increasingly more focus. One of the most crucial functions of ITS is Traffic Signs Detection and Recognition (TSDR) by notifying drivers of the status of road signs and providing helpful information regarding safety procedures which helps to improve safety. A deep learning system is designed that is a combination of Convolution Neural Network(CNN) and Spatial Transformer Network(STN) to obtain maximum accuracy. AlexNet (Model-A) and BaselineNet (Model-B) are the two CNN- architecture-based models and the two sets of datasets used in the study are Belgian Traffic Sign Classification(BTSC) and German Traffic Sign Recognition Benchmark(GTSRB). To test the models in rainy, foggy, extreme sunlight, and low-light conditions augmenters like gaussian-blur, fog, snow, spatter, and contrast have been applied to the images. Further, a comparison is done between the model performance parameters after training both datasets.