Traffic sign recognition for self-driving cars with deep learning

The purpose of this research was to create a model for an autonomous car in traffic sign recognition. A high-accuracy model is needed to analyze the signs. Previous studies have mainly been centered on European countries, and the models created in Europe are not applicable to American autonomous cars. The contributions of this paper are twofold. First, this study generated a dataset that was collected and annotated in order to establish a suitable model for the USA. The dataset was custom made and acquired by using camera footage that was converted into individual frames. The dataset was named Cyber Identity and Biometrics Lab Traffic Sign Dataset Version 1 (CIB TS V1). Then, it was annotated into different classes and labels with LabelIMG. With a customized program, we used the annotations to crop out images and categorized them. Second, the data was run through a deep learning algorithm called modified AlexNet. A lighter version of the AlexNet was used for our experiments. Results showed that the model achieved above 99% accuracy on the validation set.