What we see is what we do: a practical Peripheral Vision-Based HMM framework for gaze-enhanced recognition of actions in a medical procedural task