Operating Room Surveillance Video Analysis for Group Activity Recognition

Human error during surgery can cause unexpected accidents. Elucidating the causes of human by monitoring behavior surgical team improve safety. Conventional methods include a questionnaire that is presented to team, but this subjective, and its accuracy depends on memory. However, pose estimation an emergent technology able capture person's skeletal information. Objective group behaviors using may assist in identifying error. In study, we utilized surveillance video images entire verify effectiveness proposed method extracting quantifying order identify intraoperative situation. Specifically, attempted extract focusing actions handing over instruments multiple members pay attention simultaneously. method, applied camera model individual based individual's position, facial orientation, body arm bending, wrist position. The ID are then combined input into behavior. was six actual videos estimate behaviors. Accordingly, were accurately involving cooperation addition receiving delivering, as well warranting members. Further, successfully identified important points occurrence characteristic from redundant videos.