This paper presents an original approach for driver vigilance-decrease detection, based on a progressive detection and learning of the driver's successive ways of driving, allowing the diagnostic process to adapt to the driver and environment specific characteristics. This method performs an on-line analysis of the driver's activity on the man-vehicle interface and does not require any specific sensor. It detects significative changes in this driving activity and interprets them in order to generate an alarm as soon as a vigilance-decrease is suspected. The paper describes the principle of the method, and its implementation on an on-board, real-time specific neural network architecture to confirm the ability of the proposed diagnostic system to detect vigilance-decreases including inattention, distraction, fatigue and drowsiness.
Abstract