| Driver fatigue is the major reason of severe traffic accidents, which in recent years has become the focus of traffic safety. The visual detection technology can realize non-contact measurement, and the features it detects are intuitive. So at present the visual detection technology has become the research hotspot and the mainstream in the field of driving fatigue. According to the driver fatigue evaluation research before, one or more fatigue characteristics are applied using Bayesian networks, fuzzy reasoning, artificial neural network, machine vision to estimate driver fatigue state. Its limitation lies in ignoring the drivers’ mental state change which is along with time. Hidden Markov Model reasonably reflects the drivers’ mental state changes with its own characteristic information process. It can describe the driver fatigue state in time of overall nonstationarity and local stability. It is an ideal driver fatigue assessment model. This article selects 20 subjects using simulate driving device to simulate driving on the highway. Using SMI-HED eye-tracker records visual characteristics during the driver is in the driving process, using physiological parameter tester records the driver’s ECG, galvanic skin and respiratory and other physiological signals in the process of driving. The main works as follows:1. In all kinds of driver fatigue assessment model, this paper analyzes the Hidden Markov Model (Hidden Markov Models, referred to as the HMM) driver fatigue assessment model. Select parameters PERCLOS, AECS, PERLVO as the important parameters which can reflect driver fatigue state. Respectively choose a single parameter PERCLOS and multi-parameter PERCLOS, AECS, PERLVO as parameter variables to build fatigue assessment model based on HMM. Driving simulation experiments obtain the experiment data which is used as the model’s training sample data and evidences to verify the accuracy of the model. Use Baum-Welch algorithm (which is also known as before and after algorithm) to get the final HMM driver fatigue assessment model parameters.2. In order to determine the driver’s mental state, this paper calculated posterior probability of observations to get fatigue probability of each observation values, and compared with the fatigue probability of physiological parameter tester for analysis. This paper adopts the HMM algorithm Viterbi algorithm to inference the most likely state during the time that generated observation sequence. Using physiological parameter instrument parameters such as ECG, EEG dates which was collected to during the same time to determine the drivers’ mental state, and then verify the exactness of the model.3. Experiments were carried out to get the data to build HMM model, and verify the accuracy of driver fatigue state assessment model. Through data analysis, regardless of the HMM driver fatigue evaluation model based on the single parameter o multiple parameters, it can response the driver fatigue is a process of change over time. At the same time, through comparative analysis of the two models, The probability curve which is get from the model that using multi-parameter PERCLOS, AECS, PERLVO to build is agree with the driver’s real mental state. |
