The Design of Hand-Pointing Interface for Large Display
| Autor: | Jian-Liang Pan, 潘建良 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 Aiming at developing a natural hand gesture interface for human-machine interactions on large displays, this research develops novel techniques for tracking and recognizing hand gestures in the video sequence based on vision-based technologies. During the tracking and recognition, the proposed method tackles not only changes in environment illumination, but also the affine motion of hand gestures. For hand detection, instead of using commonly-used skin color detection methods, the proposed method exploits edge information and frame differencing method for extracting moving edges for tolerating illumination variations and handling complex backgrounds. We also design a robust feature for attaining the rotation-invariant detection and recognition of the defined hand gestures before initializing the tracking. The hand tracking starts as soon as the initial detection is finished. With the detected fingertip position, the proposed method automatically generates new hand templates for tracking. The Lucas-Kanade tracker is vulnerable to the interference of complex backgrounds. Therefore, the proposed method applies the distance transform on the detected moving edges to compose the template for tracking before using the Lucas-Kanade tracker for accurate hand motion estimation. The fingertip detection would start whenever the tracking fails due to unacceptable tracking errors or gesture changes. With the dynamically changed template, the tracking and recognition get more robust than the results of using only one fixed template. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|