Purpose With the rapid development of autonomous driving, vehicles can achieve real-time perception of the surrounding environment by installing cameras. Traffic signs provide rich environmental information, and their accurate recognition can enable autonomous vehicles to timely obtain road rules information. By combining image processing, computer vision technology and deep learning with the transportation field, intelligent and contactless detection of traffic signs can be achieved. Design/methodology/approach Firstly, to address the problems of poor quality and low visibility of some traffic sign images, a BF-IE-MSR enhancement algorithm is proposed. The contrast and information entropy of the enhanced image are higher than those of traditional algorithms and the image quality is significantly improved; Secondly, in response to the problems of large parameters and low accuracy in current models, a lightweight detection model GCS-YOLO11 is proposed. The GhostConv module is introduced to reduce the parameters and the CBAM module is introduced to optimize the detection accuracy of traffic sign targets. Findings The experiment showed that the mAP on the GTSRB dataset is 96.8%, with an overall improvement of 3.7%; the inference time is 10.5ms, with an overall reduction of 0.7ms. Various metrics such as mAP, Params and Inference time are superior to other common deep learning models. In addition, the TT100K dataset is also tested, with mAP increased by 1.6%, Params reduced by 0.46 M and Inference time reduced by 6.5ms, the accuracy and real-time performance improved. It can achieve intelligent and contactless detection of traffic signs, providing technical support for traffic signal control optimization and traffic flow adjustment. Originality/value At present, most research is based on open datasets for detection. However, in the GTSRB dataset, some images have blurry targets and poor visibility. When inputing into the detection model, it can affect the detection performance of the model. Therefore, before model detection, this paper first preprocesses the low-quality images in the dataset to enhance the features of the target to be detected in the images and then inputs them into the improved YOLO model for traffic sign recognition to improve the accuracy and efficiency of model detection.
Yang et al. (2025) studied this question.