Safety Driving Detection Based on Convolution Neural Network

With the rapid development of China’s economy, the popularity of vehicles is increasing year by year, and traffic safety problems emerge one after another. For the sake of automobile safety problems, driver safety detection becomes very necessary. Based on the theory and model of deep learning, a set of safe driving detection model is constructed. We use dlib face recognition database and yolov5 model to design detection algorithms for eyes, mouth and head down, and construct a set of fatigue driving detection algorithms. On this basis, safe driving tests for drinking water, smoking and answering phone calls are added. After designing the safe driving detection algorithm, we build pyside 2 interface visualization, so that we can turn on the camera to detect faces in real time, detect whether there is fatigue and unsafe driving behavior, and give warnings. In order to evaluate the safe driving detection algorithm, we use accuracy rate P, recall rate R and average accuracy of multiple categories mAP as the evaluation indexes of model performance. The experimental results show that the accuracy rate is as high as $93.5 \%$, and the recall rate is as high as $\mathbf{9 3 . 2 \%}$. The accuracy of the detection system is high