In view of existing mobile devices which have the limitation of relatively small area of the touch screen, the present invention describes a virtual touch sensing method based on computer vision technology. The method includes the steps of using more than one sensor to detect the coordinates of an indicator in a virtual touching area, and calculating the respective screen coordinates according to the coordinates of the indicator, where the area of the operation surface of the virtual touching area is independent to the area of the screen. The present invention also disclosed a corresponding virtual touch sensing system which provides a predictive control interface, where the area of the control interface is independent to the area of the actual screen.