Research On Fatigue Detection Algorithm Based On Visual Features

With the increasing number of vehicles and drivers,the world is facing more and more severe traffic condition.The data shows that over 10 percent of major traffic accidents result from fatigue driving.Thus it is of great importance to detect the driver’s fatigue level to decrease the rate of accidents.How to extract fatigue features and how to detect driver’s fatigue level with proper methods have become the hot topic among the research area all over the world.This paper analyses existing fatigue detection methods and proposes a fatigue detection algorithm based on multi-visual features.First,in order to increase the accuracy and steady of fatigue detuection,based on the face and landmark detection result from Adaboost and ASM algorithm,this paper gives 8 visual fatigue features fused for fatigue detection,including blink frequency,PERCLOS,yawning frequency,total yawning time,nodding frequency,total nodding period,turning frequency and total turning time.Then the date was put into different machine learning models for training.By using fork validation method,logistic regression(LR)model shows better performance than other machine learning models,and was selected as the fatigue detection algorithm for detection system.At last,a real-time fatigue detection system was implemented and the size of the detection sliding window is 300 frames.When the system detects that the driver is in fatigue status,it will give an alarm to help reduce the traffic accidents rate.The paper compares the result of the proposed LR algorithm with those of existing P80 and fuzzy system detection model.Result shows that the proposed fatigue detection algorithm based on multi-visual features by using LR model performs the best in recall,precision and accuracy.