Traffic Sign Detection in Complex Scenarios based on YOLOV7

Abstract

In recent years, thanks to the development of the automobile industry and the advancement of industrial intelligence, driving assistance systems have been developing rapidly under the requirements of national policies and the demands of the consumer market. Traffic sign detection module, as an indispensable part of the driving assistance field, is in urgent need of new target detection algorithms in this field to solve the problems of poor detection accuracy, low computing speed and poor adaptability of the existing algorithms to the vehicle hardware. To solve the application problem, in this paper, we propose a traffic target detection algorithm based on YOLOV7. The model is trained using the CCTSDB dataset and tested in complex scenarios such as different lighting conditions and weather conditions. The mAP of YOLOV7 is stable at over 82.46% and the mAP value of YOLOV7-tiny is stable at over 78.57%, demonstrating the effectiveness of our method that it has the ability to detect and recognize our traffic signs in practical scenarios. In addition, this paper conducts a comparative study on the commonalities and differences of the above two models and draws the corresponding conclusions. In summary, YOLOV7 series can be effectively used in traffic sign detection work.