基于改进Yolov3的驾驶员疲劳检测

本文基于多信息融合方法研究了驾驶员疲劳检测技术。通过改进的Yolov3算法与卡尔曼滤波算法的结合进行人脸检测。利用一种基于提升树的算法实现脸部关键点检测。并基于单位时间里眼睛闭合时间所占的百分比(PERCLOS),最长持续闭眼时间和哈欠次数这三个特征进行多特征融合的疲劳检测。在实车录制数据集上进行验证,实验证明该方法平均识别正确率达92.5%, 具有较高的准确率,针对复杂环境有较强的鲁棒性,对于将来的研究有着重大意义。

Driver fatigue detection based on improved Yolov3

In this paper, driver fatigue detection technology was studied based on multi-information fusion method. Face detection was carried out by combining the improved Yolov3 algorithm with Kalman filtering algorithm. An algorithm based on lifting tree was used to detect key points of face. The multi-feature fusion fatigue detection was performed based on the Percentage of Eyelid Closure Over the Pupil Over Time (PERCLOS), the longest continuous eye closure time and the number of yawns. The experimental results show that the method has a high average recognition accuracy of 92.5% and strong robustness against complex environment. It is of great significance for future research.