A method for constructing 3D human facial images which maintains the characteristics of video input image is proposed. First, a personalized 3D polygonal facial model is constructed from two orthogonal facial images. Video input frames are modified, and the movements of facial feature points are tracked and measured based on the contraction rule of facial muscles. The vertices of the 3D polygons are altered accordingly, and the modified input frame is texture mapped directly onto to the forementioned model. This method was implemented using an ordinary CCD camera and computer, and an experiment using one person's data showed that realistic facial images is reconstructed with high speed and precision.