Traffic Sign recognition for smart vehicles based on lightweight CNN implementation on mobile devices

Abstract

The concept of smart vehicles is becoming an essential feature that ensures driver comfort and security. Smart vehicles are equipped with intelligent systems based on advanced technologies that perform a set of tasks for the mentioned purposes. Recognizing Traffic sign is one the most important systems that guarantee a high-security level. However, it is difficult to develop the best traffic sign recognition system due to numerous obstacles. such as weather conditions, geometric deformation, and most important is the material limitation. In this work, we proposed the implementation of a lightweight convolutional neural network (CNN) model on a mobile device to overcome the mentioned challenges. The proposed CNN combines high performances and low computation complexity. Evaluating the proposed model on publicly available datasets proved its efficiency. Besides, the implementation of the CNN model on the pynq platform demonstrates the possibility of using a wide range of mobile devices for the inference of the proposed model.