Modeling bed exit likelihood in a camera-based automated video monitoring application

Hospital inpatients often fall during an exit from the bed or in the ensuing seconds and minutes. Existing fall prevention technologies fail to provide adequate lead time to a patient bed exit or exhibit high rates of false alarms. To address these limitations and reduce risk of falls for patients and hospitals, we have developed a 3D camera-based system, named Ocuvera, for monitoring patients at risk of falling, without requiring a human monitor. The developed automated system looks for cues that predict a likely bed exit. If the system determines that the risk to patient safety is high, the system alerts nursing staff, often with enough lead time to prevent the exit. In this paper we discuss the algorithmic pipeline of the developed system, starting with the raw camera feed and ending with alarms. Emphasis will be placed on computer vision models of behaviors and objects, as well as a machine-learned bed exit risk model.