Head Gesture based Control of an Intelligent Wheelchair

This paper proposes an integrated approach to realtime detection, tracking and direction recognition of human faces, which is intended to be used as an human-robot interaction interface for the intelligent wheelchair. Adaboost face detection is applied inside the comparatively small window which is slightly bigger than the Camshift tracking window, so that the face precise position, size and frontal, profile left or profile right direction can be obtained rapidly. If the frontal face is detected, canonical template matching is used to tell the nose position. After ascertaining the head gestures, including up, down, left (frontal left or profile left), right (frontal right or profile right), and center, the intelligent wheelchair is expected to speed up, slow down, turn left, turn right or keep speed correspondingly. Experimental results show the robustness in face detection, tracking and direction recognition of the presented method. Finally, a sequence of images of ARIA simulation is provided to demonstrate its feasibility and promptitude.