Research On Multi-Feature Fatigue Detection Based On Machine Vision

As a means of transportation for every household,automobiles bring convenience and also cause frequent traffic accidents.Intelligent driving technology has won an unprecedented upsurge.Driver fatigue detection technology is an important part of it.On the basis of intensive study of relevant methods and principles,considering the generalization of application,and in order to detect driver fatigue more accurately and steadily,this paper proposes a non-contact fatigue detection algorithm,and develops a driver fatigue detection system with multi-feature fusion based on machine vision.Firstly,the driving video captured by camera goes through a series of image preprocessing.And the face is initially located by AdaBoost algorithm based on MBLBP features.And then the fatigue features are extracted and calculated by using Stasm(Stacked Active Shape Model)and camera pose measurement.Finally,the driving status is judged by ELM(Extream Learning Machine)classifier and it is transplanted to development board Jetson TX2,realizing on-board fatigue detection.The main research contents of this paper lies in:1.Aiming at the problem that the haar detection features used in the classical AdaBoost algorithm are greatly affected by illumination,MB-LBP features with gray invariance are used to improve the speed of face detection and reduce the influence of illumination on face detection.2.Study a method of location for feature and feature extraction,and a method of rotation angle measurement based on vision.Key feature points are located by Stasm which is improved on the basis of ASM(Active Shape Model),extracting the facial state parameters accurately and uniformly.On the basis of Stasm,the real-time head angle relative to camera is obtained by combining camera pose measurement,extracting the head state parameters accurately.3.In order to improve the robustness and accuracy of the algorithm,several specific fatigue features are used to judge the fatigue state together.Four specific fatigue features,including PERCLOS(Percentage of Eyelid Closure),blink frequency,yawn frequency and nod frequency,were calculated by combining various basic state parameters with the initialization of the specific driver.4.In order to improve the speed and reliability of state classification,ELM is used for classification.Set up the experimental platform and collect simulated driving videos of volunteers.Take the four specific fatigue features calculated by the above algorithm as the sample set,which train and test the classifier to complete the judgment of driving status.The program is transplanted to the development board Jetson TX2 to realize onboard fatigue detection.Through testing and verification,under different illumination conditions,the overall detection rate of fatigue detection method in this paper reaches 86.19%.It provides information basis for driving fatigue early warning,avoiding driving fatigue to achieve safer driving.