Context-Dependent Action Interpretation in Interactive Storytelling Games
In this paper, a framework of context-dependent behavior interpretation in interactive storytelling system is proposed. A user can act as one of the role characters in a story to interact with other virtual characters in the system. The authors implemented two levels of action interpretation: activity and behavior. A Microsoft Kinect sensor is used to acquire and recognize user's activities in terms of the information of its body joints that will be trained by a pre-learned model. Then, with multiple-context modeling and the recognized activities, a dynamic Bayesian network is adopted to disambiguate user's behaviors in terms of his intentional and subgoal structure.