Accuracy of 3D camera based virtual keyboard

Tablet producers offer many utilities to attract the users or to facilitate the work with the tablet. The virtual keyboard based on a camera, which will replace the touch screen keyboard, can be such a utility in the near future. We work on a project of such a virtual keyboard. Our concept of a 3D virtual keyboard consists of two steps. In the first step we detect the fingertips in 2D images from both cameras. In a second step the fingertip position in 3D space is determined using the triangulation method. For fingertip localization in 2D image we use three algorithms. The first one is based on finding local maxima of the finger contours, and this maximum has to meet certain criteria. The second proposed method uses a curvature function calculated from a traced hand contour - a curvature function has the local maxima at the fingertips. The third method is based on angle between two vectors created from a traced hand contour - angle has a minimum at the point of fingertip. In this paper we discuss the accuracy of two camera system for virtual keyboard and we find out whether the chosen method for fingertip localization in 2D image has an impact on accuracy.