Most current challenge of hand touch detection in interactive projector system is how to detect finger touching with a single camera. In this paper, we introduce a projector camera system that enables users to interact with computers by touching on arbitrary surfaces with bare-hand. The realization of touch detection is through a novel vision-based algorithm performing in three steps: 1) foreground extraction according to a predicted image generated from geometric and photometric calibration matrix; 2) fingertip detection based on the curvature of hand contour points; 3) touch detection by encoding a local region of self-adaptive structured light into the projected image. In our approach, through the disparity of pattern codes projected on the fingertip, it is able to distinguish immediately whether a touch action really takes place without complex calibration and triangulation. Extensive experiments on foreground extraction, fingertip detection, and touch judgment are presented to show the robust performance of this system.