Performance Analysis of Deep Dense Neural Networks on Traffic Signs Recognition

Abstract

Traffic signs are a vital tool of the transport system because they serve to keep pedestrians and drivers readily informed that they can be alerted and notified. Thus, different traffic sign recognition systems were found in the last few years. It is implied that their identification and recognition is a confined issue that signs may be special, distinctive functions, or fragile shapes and solid shapes. Some recent and effective approaches of traffic sign detection and classification showed the success of using deep neural networks in this field. In terms of this domain, the development of an accurate real-time traffic signs recognition system is still a challenging task. This paper discusses the recognition system of traffic signs using four dense CNN-based models, DenseNet121, DenseNet161, DenseNet169, and DenseNet201. However, the present study aims mainly at evaluating the performance of the proposed system using deep dense neural networks on recognizing traffic signs. Results show the feasibility of using DenseNet pre-trained models to perform this task. In terms of testing accuracy, DenseNet201 achieved the best performance with 99.7%.