多重觸控桌面顯示系統之手指偵測

Abstract

隨著顯示科技及輸入技術的發展，將電腦引入日常生活的互動桌面系統(tabletop)越來越受到注目，許多研究致力於如何建構大型的互動顯示系統，基於桌面系統的需求，利用紅外線攝影機及紅外線投燈器，透用影像處理的方式來進行手指偵測及追蹤的系統日益增加。然而，基於電腦視覺技術開發的多重觸控手指偵測系統會受到紅外線投燈器照射光分佈不均勻的影響，導致多指觸控偵測效果受到影響，打光佳的地方偵測效果較好；另外，現存多指觸控偵測系統在效能上亦不如預期，使用者在操作上會因為偵測時間較長而有延遲的感覺，進而影響指觸人機互動系統的操作性；再者，對一個基於影像處理的方式開發的手指偵測系統而言，為了達到良好的操作性，有許多相關的參數需要調整，對使用者而言，是相當煩雜且耗時的工作。 因此， 在本研究中我們開發了一個運用圖形處理器加速的高效能多重觸控手指偵測系統，解決了操作時因效能不佳而延遲的問題，另外，我們提出了一個方法來均一化偵測手指時的前景亮度，解決了紅外線投燈器照射光分佈不均勻的問題，還提出一個基於取樣搜尋的方法來協助使用者決定適用於多指觸控系統的參數。

Advances in display and input technologies either based on electronic devices or computer vision have led to booming popularity of direct-touch interactive surfaces. Since there is a trend to use larger display, computer vision approach is adopted for direct-touch interaction due to low cost. However it suffered from non-uniform lighting problem. Besides, the computation efficiency is not good enough in existent multi-touch detection systems. Furthermore, determining parameters for detection system includes try-and-error process which is tedious and time-consuming. Therefore, we modified and designed a finger detection system based on previous work to improve drawbacks mentioned above. In this work, we develop a GPU-accelerated multi-touch detection which makes the computational efficiency fast enough to reduce the delay while manipulating tabletop system. In addition, we propose a method to normalize foreground to solve the non-uniform lighting problem. Furthermore, we propose a sampling-based parameter search method to help user determine an appropriate set of parameters for multi-touch detection system.