This paper introduces a collision warning system in which the driver's braking action is predicted by Hidden Markov Models in order to prevent annoying warnings. If the warning of collision sounds before driver's braking action that is intended by the driver, the driver may feel annoying. The Hidden Markov Model attempts to predict the driver's braking action in real-time. To construct the Hidden Markov Models for the prediction of braking action, real experiments were carried out with 18 subjects performing about 1- hr drive. After training offline, we conducted the real experiments using the collision warning system in order to examine online the effect of prediction of braking action in which 19 subjects were participated. The result showed that the number of annoying warning could be reduced, and allowed us to confirm the possibility of the prediction of braking action, using the Hidden Markov Models applied in real traffic situations.
Abstract