New algorithm for 3D facial model reconstruction and its application in VR

3D human facial model reconstruction is essential to generation of facial animations that is widely used in the field of virtual reality. The main issues of 3D facial model reconstruction based on images are in twofold: one is to track the features of face automatically and the other is to get the realistic-looking human face model. In this paper, we introduce a method on structure from motion. First the accurate corner matches can be acquired by integrating modified image pyramid Lucas-Kanade algorithm and local adjustment algorithm, then a realistic-looking model is generated by camera calibration, motion analysis and texture mapping. Experimental results demonstrate the high accuracy of the corner matching algorithm and the vividness of the facial model.