A Robust and Fast 3D Face Reconstruction Method Using Realsense Camera

In this paper, we propose a fast method to reconstruct a plausible 3D face by deforming a standard face model, constraint to 78 landmarks acquired by Intel realsense camera. First, we calculate the front view coordinates of the 78 landmarks by the Eulerian angle information acquired and scale the front view coordinates to fit the standard model. Second, we calculate a transformation matrix which calculate the displacement of all the points in the face through the displacement of the landmarks. Finally, based on the results of the second, we adjust all the points of the model to best fit the real face. Experimental results demonstrate the effectiveness and efficiency of our method, as well as the robust adaptation to face pose and efficiency of the algorithm.