Vision-based augmentation of a sentient computing world model

This paper presents a work which integrates computer vision information obtained from calibrated cameras with location events from an office-based ultrasonic location system. Bayesian networks are used to model dependencies and reliabilities of the multi-modal variables and perform fusion. Context is represented using a world model which incorporates aspects of both the static and dynamic environment. Information from the sentient computing system is used to guide and constrain the computer vision components, which in turn enhance the accuracy and capabilities of the world model.