Fast and accurate 3D face reconstruction based on facial geometry constraints and fringe projection without phase unwrapping

Abstract

Phase unwrapping is a critical process and a challenging step in fringe projection profilometry. Methods employing additional auxiliary patterns or embedding signals in the projected fringe pattern can successfully address this issue, but most of them suffer from increased number of structured patterns or depressed signal-to-noise ratio (SNR) of fringe images, which are not expected in real-time high-accuracy application such as 3D face reconstruction. In this paper, a 3D face reconstruction method that completely sidesteps the phase unwrapping process is proposed. Phase order lines (POLs) are first extracted from the left and right wrapped phase map of the stereo vision system respectively, and then the left and right POLs are matched by automatically drawing polylines on the face with facial geometry constraints as prior knowledge. Finally, based on the matched POL pairs, fine matching is performed directly employing the wrapped phase to generate a dense disparity map. Compared with the conventional, the processing time of the proposed method is further improved without sacrificing accuracy. Experiments demonstrate the effectiveness and robustness of the method.