Improved lightweight YOLOv5s Algorithm for Traffic Sign Recognition

Abstract

A lightweight traffic sign recognition method based on the YOLOv5s algorithm is proposed to address the drawbacks of the current road traffic sign model, such as sluggish detection speed, huge model, and many parameters. To increase the speed of detection, the lightweight GhostNet backbone network is first deployed, which further reduces the parameters and size of the model based on YOLOv5s. Second, the Anchor that is appropriate for the CCTSDB 2021 dataset is recreated using the K-means clustering technique. The NMS algorithm of the original network is then replaced with DIoU-NMS to enhance the recognition of veiled indicators and lower the missed detection rate. To increase the model's detection precision, the CIoU loss function of the original network is swapped out for the EIoU loss function. Research on the CCTSDB 2021 dataset reveals that while the parameters are lowered by 16.5%, the model size is reduced by 16%, and the FPS is increased by 7, the detection accuracy is only dropped by 2.1% when compared to the original YOLOv5s model. The improved algorithm can fulfill the mobile end of many scenarios with a balance of speed and accuracy requirements, as opposed to YOLOv3-tiny and other algorithms.