Evaluation of Tracking Methods for Human-Computer Interaction
This paper improved the existing “Camera Mouse” system.
The Camera Mouse system is an interface that processes video input and turns that into input controlling the mouse.
The improvements are made by applying different tracking and filters.
Tracking works by distinguishing facial features and measuring the distance they travel between consecutive frames.
In this study, two trackers are used.
The correlation tracking method is achieved by calculating the differences for each pixel in a 50 X 50 area.
The Lucas-Kanade tracker (LK) uses the principle of optical flow. Constant brightness of a feature is assumed and the feature is then able to be tracked.
Once tracking data has been gathered, it is then passed into a Kalman Filter for further processing. This filtering allows for finer grain input to the tracker thus reducing its search space.
Subjects were asked to use several applications that required different kinds of mouse input.
After studying the trackers, it was determined that the correlation tracker took more time to location the desired features, but produced a lower error rate than the LK tracker.
The Kalman Filter improved tracking accuracy for “non-erratic” mouse movements. The Kalman filter was also more accurate without alternation.
It was also noted that horizontal movement was more accurate than vertical movement. The average range of horizontal movement was 247 pixels while the vertical was 88.