Hands-free Wheelchair Navigation Based on Egocentric Computer Vision: A Usability Study

People who suffer from impaired upper-limb mobility find a conventional joystick-controlled powered wheelchair difficult or impossible to drive. Thus, several alternative, hands-free methods for controlling the wheelchair via the chin, the tongue, voice commands, and eye tracking have been developed. A recent method relies on an egocentric camera as the primary sensor and on computer vision technology to track the user’s head motion and translate it to control signals for the wheelchair. In this paper, we evaluate this vision-based wheelchair control approach by conducting a two-round comparative usability study with 21 subjects. In each round, the subjects were required to navigate a powered wheelchair using the visionbased approach through an indoor test area. The subjects also navigated the same route using two baseline approaches: chinbased control, which is a commonly-used hands-free alternative, and manual joystick control, which is the most widely used method for driving a powered wheelchair, but requires hand functionality. We propose to use joystick control as a reference that is available to all researchers in this area. By comparing their methods to manual control, the results of studies such as ours would be immediately comparable to each other. While we do not expect hands-free control methods to enable faster navigation times compared to joystick control, the loss of efficiency can be use to quantify the effectiveness of alternative methods. Our quantitative and qualitative results show that the vision-based control approach is viable for hands-free indoor use. Moreover, the improvement in performance in the second round using our method provides evidence that users can close the gap to joystick control with practice.