Lightweight and Efficient Convolutional Neural Network for Traffic Signs Classification

Recently, Intelligent Transportation Systems (ITS) has obtained a large interest in scientific research, due to the intense increase in the number of vehicles in the traffic scene. In fact, ITS is able to solve many problems using computer vision, such as traffic signs recognition. Lately, Convolutional Neural Network (CNN) approaches have been applied in traffic signs classification due to the robust feature extraction with size and rotational invariance. However, the majority of the work realized in this task focuses on accuracy rather than the number of required parameters, which makes applications of traffic signs classification inappropriate for real-time uses. To solve this issue, we propose a lighter and efficient CNN model called Lightweight Traffic Signs Network (LTSNet), which requires fewer parameters while having good accuracy. The experiments are performed on the public benchmark dataset of traffic signs GTSRB to prove the effectiveness of our proposed network in terms of accuracy and parameter requirements.