情報投影提示型インタフェースにおける指差し入力手法の検討

抄録

We propose a robust method for hand and fingertip detection on a projected dialog box. We have developed an interface system that enables the user to input and obtain information in a position-free manner within a specific area of the environment using a projector-camera system. The system detects hand-waving by the user as the system start cue; the system then tracks the hand, and projects a command dialog box at the position specified by the user. The system then observes the button that the user points at with their fingertip and thus receives commands. The conventional fingertip detection method used skin color information and a template matching process with a circular template. While this method was effective, it was not completely robust to changes in dynamic illumination, particularly in the color extraction phase. In this paper, a new fingertip detection method that does not mainly require color information is proposed. Hand candidates are detected as newly appearing objects in the projection area using background subtraction with bipolar radial reach correlation (BP-RRC). The system then judges whether the hand candidate is a real hand or a ghost using a combination of optical flow and edge information. If the candidate is judged to be a ghost, it is then regarded as part of the background. The real hand region is thus detected and the fingertip position is calculated based on the convex hull of the hand. Experiments show that the proposed fingertip detection method's performance is superior to that of the conventional method.