一种基于深度学习的疲劳驾驶检测方法研究

Research on Fatigue Driving Detection Method Based on Deep Learning

疲劳驾驶检测对避免车辆事故的发生有着非常重要的意义,对检测方法的实时性和准确率均有较高的要求。为此,提出一种基于深度学习的疲劳驾驶检测方法。首先,使用改进后的目标检测网络YOLOX对驾驶员的面部区域进行定位;然后使用PFLD深度学习模型检测面部关键点,从而计算出眨眼频率、打哈欠频率和点头频率等疲劳特征参数值;最后,通过多特征融合疲劳判定算法判断驾驶员的疲劳状态,从而进行有效的疲劳驾驶预警。大量的实验表明,该疲劳驾驶检测方法在实时性、准确率等方面都取得明显的性能提升。

Fatigue driving detection is very important to avoid the occurrence of vehicle accidents,and has high requirements for the real-time and accuracy of detection methods.To this end,a method of fatigue driving detection based on deep learning is pro-posed.First,the improved target detection network YOLOX is used to locate the driver's facial area,and then the PFLD deep learn-ing model is used to detect key points of the face,thereby calculating feature parameters such as blinking frequency,yawning fre-quency,and nodding frequency.Finally,a multi-feature fusion fatigue determination algorithm is used to determine the driver's fa-tigue state,so as to provide effective early warning of fatigue driving.A large number of experiments show that this method has achieved significant performance improvements in the real-time and accuracy of fatigue driving detection.