Modeling Research On Driver Fatigue

In order to reduce traffic accidents caused by driver fatigue, people get drivers' psychological signals through a lot of testing devices to study driver's states while driving, and has achieved great achievements. However, the detection devices affect drivers'driving process seriously, thus impeding physical and psychological research results from being applied in the vehicle early warning system. With the rapid development of sensor and computer technology, driving signal detection has characterized by precision, stable and efficiency. At the same time, it helps to remove the obstacles of realizing early warning system of driving.This paper briefly discusses the causes of fatigue driving, current research results and trends by scholars at home and abroad. Combining the national and worldwide research status and driving simulation chamber in Beijing University of Technology we designed the experiment, and acquired driving behavior signals. This article firstly completed preprocessing of driving behavior data, secondly carried on vector quantization of driving behavior, and finally identified the state of drivers. Driving behavior data preprocessing includes:wavelet filter to get rid of the interference; Gaussian normalization to analyze driving behavior data; FFT transform and PCA algorithm to complete data fusion on signals deviating from the centerline and signals of steering wheel angle effectively; vector normalization of driving behavior characteristics to eliminate impact of differences in magnitude of each component on distance measure. LBG algorithm and hierarchical clustering algorithm are applied for vector quantization of driving behavior vectors, and the quantitative results of the two algorithms were compared and analyzed. Driving state determination includes:making use of Baum-Welch algorithm for establishing Hidden Markov Model of normal driving, and analyzing driving status with before and after algorithm. In addition, based on ideology of Hidden Markov Model, keeping driving behavior vector sequence, we introduce information entropy to analysis on the driving state.