Research On The Detection Method Of Fatigue Driving Based On Driving Behavior

With the continuously growing of vehicle conservation, traffic accidents occurred more and more frequently. Research shows that 57 percent of catastrophic accidents have been estimated to be fatigue related. Therefore, developing a technology that can detect the driver’s fatigue level in real time is very significant to prevent the traffic accidents and improve the road safety. This paper aims at designing a fatigue driving detection algorithm that can fusion multiple driving behavior features, of which purpose is to determine the driver’s driving state by analyzing changes in driving behavior characteristic data.First, the paper studied the present research at home and abroad. On the basis of summarizing the predecessors’ research, the relationship between driving performance and fatigue driving, as well as the causes and influence factors of fatigue driving were introduced. Two experiments(fatigue driving and normal driving) were designed and completed based on a driving simulator. A total of 25 drivers driving behavior data in different driving conditions were collected in the experiments. In the end, a sample database was established by sorting and filtering data.Second, the driving behavior characteristics in fatigue and normal state were analyzed. Statistical methods were used to analyze the fluctuation characteristics of driving behavior data to explain the effect mechanism of fatigue on drivers. In addition, sample entropy was used to analyze the complexity of driving behavior data. According to the thorough analysis, the characteristics of the driving operation and vehicle state parameters when a driver becomes fatigue were clear. Ultimately, we extracted speed, steering wheel angle and lateral position of the vehicle as the parameter for fatigue driving detection.Third, single-parameter detection algorithms were designed based on KNN method on the basis of the extracted parameter for recognizing fatigue driving. In order to optimize the performance of the algorithm based on single parameter, DTW distance was introduced in this paper. The results showed that the performance of most of single-parameter detection algorithms is not well, and the algorithms that were optimized by DTW distance are better.Finally, fusion detection algorithms based on multiple driving behavior parameters were established. First of all, this paper proposed an improved weighted voting method to fuse the recognition result of multiple single-parameter detection algorithms from decision-making level. To compare with the fusion algorithms based on decision-making level, BP and GA_BP neural network method were used to establish a fusion detection algorithm based on multiple driving behavior parameters from character class. The results show that the performance of the fusion detection algorithms based on weighted voting method and GA_BP method are better, but the former is best.