Abstract
Self-driving vehicles are considered one of the most significant advanced technologies in computer science and artificial intelligence. The primary objective of autonomous driving is the elimination of human supervision from the workflow of autonomous cars. The advantages of this technology are to improve people's lives by minimizing traffic, eliminating parking spaces in cities, boosting collective fuel efficiency, and lowering accidents. As autonomous driving becomes more incorporated into our daily lives, effective solutions are essential to its challenges. One of the most important tasks for autonomous driving is the automatic recognition of traffic signs. Artificial intelligence is broadly used in the object detection fields and especially in traffic sign detection tasks, while Convolutional Neural Networks (CNNs) are utilized in many computer vision tasks. In this paper, we conduct a comparison study of object recognition challenges for self-driving automobiles, in particular for traffic signs and traffic signals detection. Two standard datasets and six different deep CNNs architectures are utilized to carry out experiments. The data augmentation method is employed to solve the unbalanced datasets problem. Experiments show that for the GTSRB dataset, the DenseNet201 model with Adam optimizer achieved the top accuracy (98.66%). For the LISA TL dataset, the InceptionResNetV2 model with Adam optimizer achieved the highest accuracy among the other networks (98.84%).