基于深度学习的人脸疲劳检测系统

Face fatigue detection system based on deep learning

针对某些场景人脸部分遮挡引起的疲劳检测模块失准、人脸检测模型在算力较低的机器上帧数低等问题,本文提出了一种改进的人脸疲劳检测系统。首先用Mediapipe的Blaze Face和Controursmodel代替Open CV和Dlib对人脸进行识别和检测。利用可分离卷积,使骨干识别网络具有更少的参数量,其次增大了可分离卷积depthwise层的卷积核尺寸,实现更快的下采样。本文通过实验证明了改进的人脸疲劳检测模型具有较高的鲁棒性。对于实时视频输入,所提出的系统可以优化输出帧数,提高实时性。

In order to solve the problems of misalignment of fatigue detection module caused by partial occlusion of face in some scenes and low frame number of face detection model on a machine with low computational power,an improved face fatigue detection system is proposed in this paper.Firstly,BlazeFace and Controursmodel of Mediapipe are used to replace OpenCV and Dlib for face recognition and detection.By using separable convolution,the backbone identification network has fewer parameters.Secondly,the size of the convolution kernel of the depthwise layer of separable convolution is increased to realize faster downsampling.The experimental results show that the improved face fatigue detection model has high robustness.For real-time video input,the proposed system can optimize the number of frames and improve the real-time performance.