Using Sparse Visual Data to Model Human Activities in Meetings
Abstract:
We have recently engaged on the challenging development of an agent to assist users in everyday office-related tasks. In particular, the agent needs to keep track of the state of their users so it can anticipate the user’s needs and proactively address them. The state of the user may be easily available when the user directly interacts with their agent through a PC or PDA interface. However, when the user attends a meeting and interacts with other people, PC and PDA interfaces are not sufficient to give the agents a general view of the environment in which their users are interacting. In this paper, we introduce the CAMEO, the Camera Assisted Meeting Event Observer, which is a physical awareness system designed for use by an agent-based electronic assistant. We then present a particular aspect of CAMEO and main contribution of the paper, namely how CAMEO addresses the problem of extracting and reasoning about high-level features from real-time and continuous observation of a meeting environment. Contextual information about meetings and the interactions that take place with them is used to define Dynamic Bayesian Network classifiers to effectively infer the state of the users as well as a higher-level state of the meeting. We present and show results of the state inference algorithm. 1.
