基于知识蒸馏的叉车疲劳驾驶检测算法

Forklift Fatigue Driving DetectionAlgorithm Based on Knowledge Distillation

为解决叉车安全驾驶检测中模型过大、运行缓慢的问题,提出一种基于知识蒸馏的疲劳驾驶检测方法,旨在提升检测性能并减少参数量。文章介绍了YOLO模型和知识蒸馏技术,并定义了分心驾驶的指标,如频繁眨眼、长时间闭眼、打哈欠、拨打电话和吸烟等。通过对这些行为进行标注和对改进后的YOLOv5-S模型进行训练,系统能够实时准确地检测疲劳驾驶迹象。实验结果表明:YOLOv5-L教师模型在测试集上mAP为94.1%,未经蒸馏的YOLOv5-S模型mAP为89.3%,而蒸馏后的YOLOv5-S KD模型mAP提升至92.5%,精度提高3.2%,仅比教师模型低1.6%。该方法适用于资源受限的嵌入式设备和实时应用,为叉车安全驾驶提供了高效可靠的检测方案。

To address the issues of oversized models and slow operation in forklift safety driving detection,this paper proposes a fatigue driving detection method based on knowledge distillation,aiming to enhance detection performance while reducing the number of parameters.The article introduces the YOLO model and knowledge distillation technology,and defines indicators of distracted driving,such as frequent blinking,prolonged eye closure,yawning,making phone calls,and smoking.By annotating these behaviors and training the improved YOLOv5-S model,the system can accurately detect signs of fatigue driving in real-time.Experimental results show that the YOLOv5-L teacher model achieved a mAP of 94.1% on the test set,while the mAP of the YOLOv5-S model without distillation was 89.3%.After applying knowledge distillation,the mAP of the YOLOv5-S KD model improved to 92.5%,precision increase 3.2%,only 1.6%lower than the teacher model.This method is suitable for resource-constrained embedded devices and real-time applications,providing an efficient and reliable detection solution for forklift safety driving.