Continuous Driver Intention Recognition with Hidden Markov Models

The most common cause of accidents in individual road traffic is human failure. Accidents often arise from misbehavior of one or several drivers when inducing a driving manoeuvre. Dangers can occur either when the intented manoeuvre is not well adjusted to the current traffic situation, or when the manoeuvre is not properly announced to the environment so that the intention is misinterpreted. When designing advanced driver assistance systems, it is beneficial to gather information about driver behaviors as accurately and early as possible. This work investigates early driver intention inference with hidden Markov models by observing easily accessible vehicle and environment signals such as pedal positions or global vehicle position on a digital map in real traffic.