随着各类车载系统的普及，驾驶员出现分心驾驶行为愈加频繁，检测这种分心驾驶行为并予以提醒将很大程度上保障行车安全。基于视觉的分心驾驶的检测存在分类困难、特征提取繁琐、缺乏定量化评级方法等问题。

针对以上问题，本论文采用了一种基于视觉特征的分心检测算法。利用人脸表情识别技术，通过驾驶员的行为和表情对驾驶员分类。提取驾驶员的人脸信息输入到MobileNetV1神经网络模型进行训练，得到较高精确度的模型。再利用眨眼检测和PERCLOS算法辅助监测疲劳驾驶状态，与模型相结合进行分心驾驶检测。通过实验验证，算法的检出率达到82.8%，正确率达到81.4%，形成了较高精度的分心驾驶检测算法和模型。采用视频流关键帧截取的技术，监测单位时间内的驾驶员状态，当分心驾驶状态出现次数超过阈值时提醒驾驶员，使模型更具稳定性，有较好的实际应用价值。

With the universal apply of high-tech car-mounted system, drivers are more likely to be distracted by these factors. Detect the distracted driving behavior in order to warn the drivers could guarantee the drivers’ safety significantly. There are a plenty of weakness in distracted driving assessment methods based on visual feature: the inconvenience of classify the samples, the intricacy of feature extraction, and the lack of model rating method.

Aiming at the drawbacks above, we proposed our model of distracted driving assessment method based on facial expression recognition. According to the behaviors and expressions of the drivers, we classify the samples into distracted and focus. We extract the faces of the drivers and input them into MobileNetv1, a neutral network for training. A high-accuracy model is created to recognize the distracted driving behavior. Another model based on blink-detect algorithm and PERCLOS algorithm is proposed for driver fatigue detection in order to support the distracted driving assessment. Adequate groups of experiments have been conducted. The verification of the method concluded its effectiveness and superiority with the accuracy of 81.4% and detection rate of 82.8%.