基于改进YOLO-MobileNet的近红外图像特征驾驶员人脸检测

Study on Near-infrared Image Feature Driver Face Detection Basedon Improved YOLO-MobileNet

针对暗光环境下驾驶员人脸检测问题,提出了一种基于改进YOLO-MobileNet轻量级深度学习算法的近红外图像特征驾驶员人脸检测方法。以高于人眼敏感范围但在CMOS响应范围内的近红外光对人脸进行补光,使用摄像头采集图像,通过经改进并且轻量化的YOLO-MobileNet网络进行特征提取。试验结果表明:在FADIA数据集上,改进后的算法检测精度为84.93%,YOLOv5s算法为85.37%,改进后检测精度降低了0.44%,但检测速度提高了7.6毫秒。

Aiming at the problem of driver face detection in dark light environment,an improved YOLO-MobileNet lightweight deep learning algorithm based on near-infrared image feature driver face detection method is proposed.The face is filled with near-infrared light that is higher than the sensitive range of the human eye but within the CMOS response range,and images are collected using a camera to extract features by the improved and lightweight YOLO-MobileNet network.The experimental results show that on the FADIA data set,the detection accuracy of the improved algorithm is 84.93%,and the YOLOv5s algorithm is 85.37%,and the improved detection accuracy is reduced by 0.44%,but the detection speed is increased by 7.6 milliseconds.