AAM Based HCI for an Intelligent Wheelchair

This paper proposes an active appearance model (AAM) based human computer interface (HCI) to control an intelligent wheelchair (IW), namely, RoboChair. Adaboost is applied as the face detection module, while AAM is used to undertake face tracking task. Inverse compositional image alignment is implemented to fit the trained AAM. Subsequently, an straightforward method is carried out to estimate the face direction by comparing the current face shape to the template face shape, so that the intelligent wheelchair could be actuated just according to the head gestures. A well-designed decision making module is deployed to identify whether AAM is correctly tracking the face at the current frame. Results from simulation experiments show the robustness and veracity of Adaboost face detection, AAM face tracking, and face direction estimation presented. An image sequence of our local desktop simulation is provided to demonstrate its feasibility and reliability.