Violation Detection Method Based on Improved YOLOv5s

In order to reduce the number of traffic accidents caused by violations, violations are detected, and drivers or passengers who violate regulations are warned and punished to encourage them to comply with traffic rules. Traditional methods for violation detecting in complex environments have low detection accuracy and poor adaptability to the environment. In response to these problems, the YOLOv5s network model is improved. The SPP module is replaced by the ghostSPPF module to reduce the number of model parameters, the CBS module in the backbone network is replaced by the MP2 module to upgrade the channel dimensionality, the C3 structure is replaced by the C2f structure, and the BCEwithlogitsloss loss function is replaced by the QFocalLoss loss function. The improved network model has been used to detect the violation of drivers using mobile phones while driving. The mAP of the improved YOLOv5s network model can reach 93.02%, which is higher than the mAP of 91.37% of the original YOLOv5s network model. The detection frame rate of the improved YOLOv5s network model can reach 87.33FPS, which can meet the needs of real-time traffic violation detection.