This paper presents a user interface to assist the occupant of a wheelchair. It uses vision to recognize the gestures formed against a back-lit surface. Gesture recognition algorithm uses a combination of gesture contour geometry and a non-dimensional quantity derived using the contour curvatures. The gestures are identified in a position and orientation independent manner. The developed gesture recognition system is implemented on a wheelchair. Two different modes of operations, namely, the manual mode and the autonomous mode are provided for the user. In manual mode the user continuously interacts with the wheelchair. In autonomous mode the user action is limited to choosing a desired destination by pointing to a zone of a known map displayed on the back-lit surface. Results are presented to show the performance of the wheelchair in both modes.