以視訊為基礎之多點式觸控人機介面

Abstract

在本論文中，我們提出一建立於電腦視覺，可適用於任意物體的新式的多點式觸控人機介面，且系統的外觀厚度遠小於現行的電腦視覺觸控系統。我們的系統架構於兩組水平及俯視的攝影機和固定的邊框。當物體接觸螢幕時，水平攝影機得到接觸物體的影像。將兩水平攝影機所得到的物體位置投影回3D座標系，即可得到物體存在的可能位置。透過俯視攝影機分析各個可能位子是否存在物體的可能性，並尋找最有可能的物體位置組合。最後將結果傳送至應用程式端。為了加速在尋找可能位置的時間，我們使用Viterbi演算法和變異係數分析排序。

This thesis proposes a new computer vision based multi-touch human computer interface with any type of objects. The thickness of this system is much smaller than that of the most popular existing technologies, FTIR and DI. To detect the touch position, our system consists of a rim upon the edge of screen, one pair of horizontal cameras and one pair of tilted cameras. First, the possible positions of objects would be enumerated by the foreground in the horizontal views. Then, each position would be projected to the tilted views, and analyzed with the information from tilted cameras. Finally, the positions of existing objects can be separated by the proposed arrangement algorithm, and can be sent to applications by TUIO protocol. To speed up the arrangement process, Viterbi algorithm and variance-dependent arrangement are also applied in this work.