Application Research Of Driving Fatigue Detection Based On Facial Features

Fatigue driving is one of the main causes of traffic accidents.In order to avoid traffic accidents caused by fatigue driving,it is necessary to propose a method to determine the driver’s driving status in real-time and accurately,and to provide timely warning.The aim of this paper is to develop a fatigue driving detection system based on the fusion of driver’s facial features.The system inputs real-time driving video from the driver’s side camera,and based on the face detection model,face key point localisation model and fatigue driving status determination model,realises real-time driving status detection and status alerting to the driver on the display side,thus effectively avoiding traffic accidents.The main tasks are as follows:(1)Designing a lightweight face detection model.In order to deploy a fatigue detection system on a platform with limited computational power and to avoid the face detection model consuming too much computational resources in the detection process,this paper uses two methods,network channel pruning and knowledge distillation,to design the YOLOv5 s model in a lightweight way.Through testing and comparing the experimental models,the network channel pruning method with more outstanding detection effect was chosen as the final lightweight design solution.(2)Design of face key region localisation and head pose estimation models.Compared with the common eye and mouth region detection methods,which are computationally intensive and have weak anti-interference capability,the face key point localisation method has better real-time performance and higher accuracy.In this paper,three face keypoint localisation methods,Dlib,PFLD and Mediapipe face mesh,are experimentally compared and the Mediapipe face mesh algorithm is chosen to perform better in all aspects for the localisation of the driver’s eye,mouth and head regions.At the same time,the traditional head posture estimation algorithm was used to obtain the driver’s head pitch angle value by calculating the Euler angle,so that the driver’s head posture could be estimated.(3)Designing a fatigued driver determination model.When a driver is fatigued,he or she will blink rapidly,tilt the head and yawn and nod frequently with the eyes closed.In order to determine and remind drivers of their driving status,this paper fuses the calculated feature parameters of the eye,mouth and head regions and combines them with fatigue parameter thresholds to make a determination.Based on the requirements of the fatigue detection system,the system software environment was developed using the Streamlit tool.During the development process,the above model was embedded into the fatigue driving detection system,and the performance of the system was analysed and tested through experiments,which showed that the fatigue driving detection system designed in this project can meet the practical application requirements.Finally,the system was ported and deployed to the Jetson Xavier NX embedded platform in order to realise the practical application of the fatigue driving detection system.