Design of improved road sign recognition system based on deep learning

Abstract

As the basis of intelligent vehicle environment identification, street sign recognition is a necessary condition to assist safe driving. It is of great significance for realizing automatic driving, improving intelligent transportation system and promoting smart city construction. Traditional traffic sign recognition technology is limited for complex scenes, and the recognition effect is also affected by weather and light. The road sign recognition system based on deep learning includes street sign detection and street sign recognition. YOLOv3 deep learning object detection method is adopted for road sign detection, and CCTSDB standard data set is used as experimental data to perform operations such as region trimming, batch conversion of data format, BM3D noise reduction and image preprocessing and enhancement technology. 1800 images were selected to form the training set and test set, which ensured the high efficiency and timeliness of the road sign detection algorithm. The road sign recognition system adopts the image recognition method of Darknet-53, and the algorithm is optimized. The trained system is detected under the test of CCTSDB dataset, and the detection accuracy is over 97%, and the detection time reaches 0.021s. Therefore, the system meets the requirement of accurate real-time processing of traffic sign detection and recognition system.