Abstract
Recent advanced driving systems are used as luxury tools to handle a difficult or repetitive task. One of the most important tasks is traffic signs detection that provides the driver with a global view of traffic signs on the road. A traffic signs detection application should be able to detect and understand each traffic sign. To develop a robust traffic sign detection application, we propose to use the deep learning technique to process visual data. The proposed application is used for an embedded implementation. To solve this task, we propose to use the deep learning technique based on convolutional neural networks. As known, a convolutional neural network needs a big amount of data to be trained. To solve the problem, we build a dataset for traffic signs detection. The dataset contains 10,500 images from 73 traffic signs classes. The images are captured from the Chinese roads under real environmental conditions. The proposed application achieves high performance on the proposed dataset with a mean average precision of 84.22%. Also, the proposed application can be easily used for embedded implementation because of its lightweight model size and its fast inference speed.