Drowsy driving detection based on driver's steering operation characteristics

This paper aims to conduct drowsy driving tests on driving simulator for studying the method of drowsy driving detection. Firstly, a sample database of driver's three levels of fatigue (alert, drowsy and very drowsy) is built based on the rating on driver's facial video images by experts. Then, the feature indicators describing the steering operation characteristics of driver at different fatigue levels are quantitatively extracted, and the best combination of feature indicators is screened out by using sequential floating forward selection algorithm. Finally a driver fatigue online detection algorithm with three drowsiness levels is established based on support vector machine technique. The test results show that with driving simulator the accuracy rate of the algorithm in three-level drowsiness identification reaches 87.7% with high robustness and practicability.