15. 3d face reconstruction and face assisted video stablization
Name: Jiyang Yu
Grad Year: 2020
We propose an algorithm to reconstruct full 3D model from a single 2D image or uncalibrated video footage. In general, single view reconstruction problem is under-constrained due to dimension loss. However, human faces have many common features which can be used to make the problem tractable. Our algorithm uses a parametric face model to fit a coarse 3D shape to an image. We apply shape from shading constraints on the image to get the accurate shape with a high level of detail. Cast shadows are considered in the shape from shading process. This algorithm can also be applied to selfie video footage to help to estimate camera intrinsics and stabilize the video. Our system is completely automatic with no user input required, which makes it a scalable solution.
Industry Application Area(s)
Internet, Networking, Systems | Software, Analytics