以單台彩色深度攝影機實現基於視角之遠距直接選取技術

Abstract

目前市面上智慧電視的體感操作以移動游標為主，使用者藉由移動手掌來操作螢幕上的游標、透過手勢執行點選指令。過去有許多研究探討如何徒手直接選取螢幕上的目標，而非透過移動游標的方式，以增加體感操作的效率。然而，前人所提出的遠距直接選取技術需要特別的硬體設置，以準確的偵測使用者動作，因此要將該技術布置於普通環境中是有困難的。在本論文中，我們提出FingerShot系統，一種基於視角之遠距直接選取技術，且只需要使用單台彩色深度攝影機，即可讓使用者遠距直接選取螢幕上之目標。我們提出之即時偵測演算法會追蹤使用者的上半身骨架、眼睛及指尖，計算眼睛到指尖之射線與螢幕的交點，即使用者瞄準的位置。我們的系統也支援雙手操作及多人使用。本系統之使用者測試結果顯示，在距離螢幕至少1.6公尺遠的五種不同位置，當使用者閉起非慣用眼僅用單眼瞄準時，可點選到螢幕上11公分寬的目標；當雙眼皆張開時，可點選到12.3公分寬的目標。

Several research studies have attempted to enable remote direct-pointing to increase the effectiveness of freehand remote controlling compared with controlling through an indirect cursor. Nonetheless, since previous remote direct-pointing solutions usually required specific hardware settings for reliable sensing, to deploy such remote direct-pointing methods in the real environment is difficult. In this paper, we present FingerShot, a perspective-based absolute pointing method that requires only one RGBD camera to enable users to directly point to a remote display. The proposed real-time sensing algorithm defines the remote direct-pointing position on the display by tracking users' upper bodies, eyes and fingertips reliably, as well as supporting bi-manual and multi-user interactions. User study results show that users can effectively select on-screen gridded 11cm-width targets with their non-dominant eye closed, and 12.3cm-width targets with their both eyes open from five different positions at least 1.6m away from the display using our remote direct-pointing technique.