University of Houston

We present a system that exploits existing video streams from an hospital operating room (OR) to infer the OR usage state through Bayesian modeling. We define OR states based on common surgical processes that are relevant for assessing OR efficiency. The human motion pattern within the OR is analyzed to ascertain usage states. The system proposed takes advantage of a discriminatively trained part-based human detector as well as a data association algorithm to reconstruct motion trajectories. Human motion patterns are then extracted using kernel density estimation and a Bayesian classifier is used to assess OR usage states during testing. Our model is tested on a large collection of videos and the results show that human motion patterns provide significant discriminative power in understanding usage of an OR.