Research on real-time detection method of driver using mobile phone based on Faster-YOLO

In response to the impact of drivers' use of mobile phones on traffic safety while driving, a lightweight real-time detection algorithm based on YOLOv5n is proposed. The algorithm introduces the C2ff module and CoorAtt attention mechanism to reduce model parameters. Additionally, the RepConv module is incorporated into the feature extraction network to increase the algorithm's receptive field and enhance its robustness and generalization ability. Furthermore, the Layer-wise Adaptive Model Pruning (LAMP) pruning method is employed to further improve the algorithm's inference speed, reduce model parameters, and memory footprint, making it more suitable for embedded devices. Experimental results demonstrate that the improved network achieves a 96% [email protected], with a detection speed of 145 FPS, and model volume reduced to 0.59MB, meeting the requirements for real-time and reliable detection on embedded devices.