Abstract
This paper presents an original approach for driver vigilance decrease detection. It is based on an evolution detection of the driving activity with respect to the driving behaviour observed at the beginning of the trip. These initial driving conditions are memorized by a neural network, allowing the detection process to adapt to the driver specificity as well as changing driving conditions. Evolutions are analysed in order to select those related to a decrease in the drivers attention level. The proposed method was implemented on an experimental vehicle for validation. Experimental results obtained with ten drivers on motorways in real driving conditions are presented. They show that the method is able to perform a reliable detection of any form of vigilance decrease, including distraction, inattention, fatigue and early signs of drowsiness.