Detection Of Driver Drowsiness Based On Steering Operation And Vehicle State

Decreased vehicle control due to driver drowsiness is one of the majorcauses of serious traffic accidents. Study on real-time drowsy driving detectionmethod, is of great significance to improve road traffic safety. Analysis of agiven driver’s vehicle operating characteristics under different drowsinesslevels has shown the possibility of drowsiness detection. This method, whichhas proved to be one of the most practical technologies, is a non-intrusivemonitoring and is free from the influence of illumination. However, there arestill many challenges posed by unapparent steering operation and vehicle statechanges when a driver becomes drowsy, individual difference of operatingcharacteristics and complicated traffic environment. In order to establish areal-time drowsy driving detection algorithm with high performance, this paperfocuses on the key issues in analysis of drowsiness operating characteristics,drowsiness feature space modeling, optimal feature selection and drowsinessstate inferring. Experiments on driving simulator are performed to testify theaccuracy and robustness of the proposed methods as well.According to a thorough analysis on the fluctuation characteristics of thesteering operation and vehicle state parameters when a driver becomes drowsy,quantitative measures for detecting drowsiness are extracted from timesequence information of these parameters. The differences of each measurewith varying drowsiness levels are tested by the analysis of variance (ANOVA),and measures with statistically significant differences are selected.In order to get the best subset from the universe of the measures, Anoptimized measures selection algorithm is established. This algorithm takes theperformance of support vector machine algorithm (SVM) as evaluation criterionand uses the search strategy of sequential forward floating selectionalgorithm(SFFS) to select optimal measures combination from those withstatistically significant differences, meanwhile the drowsiness detection modelis established based on SVM. The results show that the rate of accuracy of themodel reaches86.1%. The individual difference of operating characteristics is considered. Thispaper uses the first twenty minutes’ steering operation and vehicle state data ofa driver as reference data, takes measures selected from the reference data asreference measures, and uses ratios of measures and reference measures toestablish the drowsiness detection model. This model classifying driverdrowsiness state as three levels reaches an accuracy of87.7%.In order to detect the driver drowsiness state when the car is crossing thelane, the car’s lane crossing is classified as active lane changing and lanedeparture caused by drowsiness. The differences of the steering operation andthe vehicle state parameters between the two classes are analysed. Measuresmeasuring the differences are extracted from the parameters and selected bymeasures optimized selection algorithm. Then, a drowsiness detection modelwhich detects the driver drowsiness state when the car is crossing the lane isestablished, and the rate of accuracy of this model is93.9%.