Driver Assistance: Drowsy/Fatigued Driver Detection

The Office of Crash Avoidance Research, which is part of the National Highway Traffic Safety Administration NHTSA reported in 1991 that one of the leading causes of single and multiple car accidents is driver drowsiness. The National Highway Transportation Safety Administration estimates 100,000 crashes annually involve driver fatigue, which causes more than 40,000 injuries. The Fatality Analysis Reporting System indicates an annual average of 1,544 fatalities due to driver drowsiness related accidents. However, the problem of driver drowsiness may actually be much worse than these statistics would indicate. Most of the time it is very difficult to attribute drowsiness as a cause to an accident because of the non availability of physical evidence. There may be many incidents in which the initial cause of the loss of vehicle control is drowsiness, but is reported as something else.

CISR is actively developing a system which detects drowsiness by monitoring steering patterns. The basic strategy in our research is as follows:

In the initial stage, direct measurements of driver fatigue, along with driving performance and vehicle related variables, are recorded during extended driving sessions on a driving simulator.
In the second stage, a neural network is trained off line to identify patterns in the data sets that are consistent with the behavior of tired drivers, as indicated by the direct measure of fatigue.
In the final evaluation stage, the neural net (already trained) would be given driver performance data in real time during an extended drive and be expected to predict driver state.

CISR has tested this approach using data for a number of different experiments. These experiments include:

The US DOT’s Federal Highway Administration gathered data in a driving simulator on 12 drivers in an extensive experiment. Drivers were deprived of sleep for up to 60 hours of continuous wakefulness. Their subsequent driving performance was measured by a simulator data acquisition system. This analysis was performed on this data and an artificial neural network with a specially designed input fanning process was developed for complete pattern recognition. The developed ANN is proven to be successful in predicting drivers' sleepiness based on their driving performance with more than 90% accuracy, as tested by using the simulator data. This research is continuing. Work is being done to explore additional patterns in drivers' performance and to develop methods to provide even more accurate results.
CISR performed an experiment in its Driving Simulator Laboratory with more typical hours of continuous wakefulness. In a morning experimental session, subjects drove the simulator for sixty minutes. During this session, subjects were fresh and had no sleep deprivation or fatigue. Unless the driver was too sleepy or tired to continue, the second experimental session consisted of sixty minutes of simulated highway driving and took place between 1:00 am and 3:00 am. In this session, the subjects were sleep-deprived and were susceptible to falling asleep while driving the simulator. Eye closure data was also collected during this experiment. The model again identified drowsy and wakeful steering behavior, calculated over fixed period of time, with good accuracy.
CISR performed an experiment in its Truck Simulator Laboratory to test this validity of this approach for large commercial vehicles. During this experiment truck drivers drove a truck simulator under various levels of fatigue and sleep deprived conditions. Data was recorded for parameters related to driver eye closures and vehicle driving activity. Data recorded in this experiment was used to develop the detection algorithm. Analysis of the results from this experiment shows very similar results to the previous experiments.