Research On Fatigue Detection Algorithm Based On Facial Features And Deep Learnin

In recent years,the high incidence of car accidents caused by fatigue driving in China has led to an urgent need for predicting and warning the fatigue state of drivers,which can fundamentally control the occurrence of traffic accidents.Currently,existing fatigue driving detection technologies that are based on physiological characteristics and vehicle driving information have limitations in their large-scale application due to issues such as high contact and low detection accuracy.Therefore,this article chooses to adopt a fatigue driving detection technology based on facial features of drivers,and combine it with the advantages of deep learning techniques in object detection,in order to design an effective fatigue driving detection algorithm that can develop into a comprehensive software and hardware platform.The research content includes but is not limited to the following aspects:(1)Research on facial detection and positioning for drivers.Firstly,the working principle of using deep convolutional neural networks(CNN)for driver facial detection was explored,and several common algorithms in the field of object detection were compared,such as R-CNN,FRCNN,SSD,and Yolo models.The Yolo model with good real-time performance and high accuracy was selected to achieve automatic detection of driver facial features.The studyproposes a solution based on the Mobilenet V3-Yolov5 model to address the drawbacks of its large size and high computational cost.This model uses the lightweight network Mobilenet V3 as the backbone network,replacing the previously complex Darknet53 network.The Mobilenet V3-Yolov5 model has a volume size of 36 MB and a transmission speed of 80 frames/s,with accuracy close to the original Yolov5 model.Through experimental verification,this model is more suitable for application in scenarios with limited computing resources such as mobile devices.(2)Research on driver fatigue testing algorithms.This paper adopts Dlib facial key point detection algorithm and facial localization technology to extract facial features of drivers,and then selects fatigue discrimination parameters.By using the PERCLOS parameters and blinking frequency of the eyes to distinguish eye fatigue,calculating the coordinates of the fatigue feature points of the mouth,and combining with the detection of head posture angle,the driver’s fatigue judgment is carried out.The studyextracts and recognizes facial features of drivers,uses weighted average values for fatigue judgment,and considers the influence of head posture angle and facial features.By using multi-dimensional fusion methods,the accuracy of detection is improved.The facial multi feature fatigue detection algorithm designed in this paper can achieve an accuracy of 92.00% for blink frequency recognition,93.2% for yawning recognition,and95.5% for overall testing.(3)Research on software and hardware design and testing of the system.This paper provides a detailed description of the functional components of a fatigue driving detection system,including video capture,facial feature localization and extraction,fatigue state detection,data storage and management,real-time monitoring and alarm functions,etc.At the same time,this paper designed a simple and easy-to-use interactive interface and developed a backend data observation program,which can achieve real-time interactive communication between the detection end and the web end,thereby better monitoring the real-time status of drivers.The test results indicate that the fatigue detection model has good recognition accuracy,and all detection modules of the system can work accurately in real-time.