Hand tracking with sub-pixel precision

Abstract

Human-Computer Interaction has matured from simple keyboard and mouse interface to touch screens, and more recently to 3-D touch screen interfaces. Recently, high precision object tracking algorithms are being deployed for the evaluating the user interaction experience of these interfaces in an unconstrained manner without the need for specially designed data gloves or other equipment restraining equipment. These tracking algorithms are expected to be not only precise, but also robust to instances of occlusions, motion blur and background clutter. Continuing this trend, state-of-the-art object tracking algorithms, capable of sub-pixel precision, were coupled with a carefully chosen de-interlacing technique and used for pinpointing the location of the fingertips between consecutive video frames. An independent study was carried out within the scope of this thesis to determine the most effective de-interlacing filter wherein the bwdif de-interlacing filter stood out and yielded excellent tracking results in conjunction with the ATOM tracker in most of the evaluated normal speed videos, whereas the PrDiMP tracker proved to be a better choice in the domain of the high-speed videos, where de-interlacing is inconsequential.