Developing an assistive interface for individuals with spasticity disorders cd.

3. PREVALENT HEAD-AIM DEVICES

A number of head-aim devices have been identified and tested to develop a better understanding of what performance limitations are associated with each device, according to predefined criteria. A specs matrix has also been developed and the top 4 scorers were chosen as the interface software to be augmented. The criteria being used are execution speed, robustness, ease of use, cost, resolution, and spastic features. Table 1 summarizes the findings, with the ranking associated with every device.




3.1. NOUSE

Nouse [4] was created by The Computational Video Group at the National Research Council in Canada. The face-tracking algorithm is based on detecting convex-shape nose features [5] [6]. Tracking such features is accomplished by means of template matching within a window of interest. A facial expression event triggers mouse clicking such as a double eye blink. This technique is inconvenient for people with CP since they may lack adequate muscle control.

3.2. VISUALMOUSE

MouseVision Inc. created VisualMouse [7]. A cross-correlation function between the targeted region of the current and previous frames is used to detect facial features. Double-clicking and right-clicking are possible through jerks, after sufficient dwell time has expired. Lighting conditions are a major constraint of this software. The highest light intensity is necessary to be incident on the tracked feature; otherwise, a dim light in the background is needed. Any gesture in the background could also mislead the algorithm and could cause a failure in cursor location.

3.3. CAMERAMOUSE

The Image and Video Computing Group designed CameraMouse at Boston University [8]. A normalized correlation function between templates of two adjacent frames is used as the detection algorithm [9], [10]. Double and right-clicking are made possible by first using dwell time to select an icon, and then dwelling again over that icon for clicking. The algorithm speed requires that the feature motion be slow. The software thus requires a low-resolution camera. The mouse clicking technique is inconvenient for people lacking adequate muscle control, such as this work's targeted demographic of people with CP.

3.4. HEADMOUSE

The Neural Information Processing Group designed HeadMouse v1.5 alpha at Eötvös Loránd University [11]. Hough transform and wavelets are utilized to track the face. Dwell time is used for clicking; however, double-clicking is very difficult. It also requires low camera resolution, making tracking of spastic motion extremely challenging.