基于级联神经网络疲劳驾驶检测系统设计

算法通过调整输入图片的大小、扩大最小人脸尺寸以及减小检测窗口层间放缩比例等方式，在保证准确率的同时极大的提升了人脸检测速率，其检测效率为原始MTCNN（multitask casaded convolutional networks）的18倍；构建新的卷积神经网络结构模型检测眼、嘴，网络检测的准确率可以达到95.6%。将设计的检测网络与经过改进后的MTCNN进行级联，在已经检测出来的脸部区域继续对眼、嘴进行分类及定位；设置综合性的疲劳检测函数，加强了由于单一特征而出现的虚检及误检等错误检测，准确率可以达到95.7%。

Design of Fatigue Driving Detection System Based on Cascaded Neural Network

An algorithm is proposed to greatly improves the face detection rate and ensures the accuracy by adjusting the size of input images, expanding the minimum face size, and reducing the scaling ratio between layers of the detection window. The detection efficiency of this algorithm is 18 times higher than that of the original MTCNN. By building a new CNN structure model for the detection of eyes and mouths, we can achieve network detection accuracy of 95.6%. The proposed network is cascaded with the original MTCNN to continue classifying and locating the eyes and mouth in the formerly detected face area. The false detection due to a single feature are improved through the setting of a comprehensive fatigue detection function, and the detection accuracy can reach 95.7%.