Automobile Driving Behavior RecognitionUsing Boosting Sequential Labeling Method for Adaptive Driver Assistance Systems

摘要

Recent advances in practical adaptive driving assistance systems and sensing technology for automobiles have led to a detailed study of individual human driving behavior. In such a study, we need to deal with a large amount of stored data, which can be managed by splitting the analysis according to the driving states described by driver maneuvers and driving environment. As the first step of our long-term project, the driving behavior learning is formulated as a recognition problem of the driving states. Here, the classifier for recognizing the driving states is modeled via the boosting sequential labeling method (BSLM). We consider the recognition problems formed from driving data of three subject drivers who drove on two roads. In each problem, the classifier trained through BSLM is validated by analyzing the recognition accuracy of each driving state. The results indicate that even though the recognition accuracies of braking and decelerating states are mediocre, accuracies of the following, cruising an stopping states are exceptionally precise.