Intention recognition for vehicle driving by time-series pattern learning

Real-time intention recognition by sensing a user and its environment is an important function for man-machine interaction. However, when an action can not be divided into a clear sequence of action fractions because of the duration uncertainty, conventional methods of intention recognition axe not applicable to the recognition of partial and short term intentions relating to the action fractions, i.e. micro-level intentions. Our approach to this micro-level intention recognition is to apply a pattern learning function which discovers and utilizes synchronous and temporal relations among the multi-dimensional time-series data of both user and environment. In this paper, we deal with a vehicle driving task as a typical application of the proposed intention recognition method, and examine the realizability of the pattern learning function by Support Vector Machine (SVM). Appropriate pattern learning methods to this problem axe discussed as well as our future plan.