A driver assistance system that recognizes traffic signs with deep learning

Today, with the rapid development of society and economy, automobiles have become one of the appropriate modes of transportation for almost every household. This is making the road traffic environment more and more complex, and people expect to have smart Vision powered apps that provide drivers with traffic sign information, regulate driver operations or assist vehicle control to ensure road safety. Traffic sign detection and recognition [1], one of the more important functions, has become the hot research direction of researchers at home and abroad. It is essentially the use of on-board cameras to capture real-time images of the road and then detect and identify traffic signs encountered on the road, thus providing accurate information to the driving system. However, road conditions in the real scene are very complex. After many years of hard work, researchers have yet to make the recognition system practical, and further research and refinement is still needed. Traditionally, traffic signs are detected and classified using standard computer vision methods, but manually processing important features of the image is also quite time consuming. With the development and advancement of science and technology, more and more scientists are using deep learning technology to solve this problem. Deep learning algorithms have shown cutting-edge performances in classification tasks. As a result, deep learning-based object detection algorithms have become popular in computer vision tasks. They can be divided into two main categories: Two-stage detection algorithms and one-stage detection algorithms. Two-stage detection algorithms have better performance in terms of positioning and recognition accuracy compared to single-stage detection algorithms. However, single-stage detection algorithms are designed to be faster, making them suitable for real-time applications where detection time is crucial. In this project, a traffic sign detection algorithm, mainly from state-of-the-art single-stage detection algorithms, is presented. Based on the application of road traffic sign detection and recognition, it focuses on the accuracy and high efficiency of detection and recognition. We have built two models of the database GTSDB (German Traffic Sign Detection Benchmark) using the open-source framework YOLOv3 and YOLOv7, and the resulting results are YOLOv7 97.3% and YOLOv3 99.46%. A deep convolutional neural network algorithm is proposed to obtain a model that can classify traffic signs and train traffic sign trainers to learn and identify the most critical of these traffic signs.