Multi-camera recognition of people operating home medical devices

We perform action recognition with a robust approach to recognize action information based on explicitly encoding motion information. This algorithm detects interest points and encodes not only their local appearance but also explicitly models local motion. Our goal is to recognize individual human actions in the operations of a home medical device to see if the patient has correctly performed the required actions. Using a specific infusion pump as a test case, requiring 22 operation steps from 6 action classes, our resulting classifier fused information from 4 cameras, to obtain an average class recognition exceeding 50%.