Application (pre-grant publication)
DENSE RECONSTRUCTION FOR NARROW BASELINE MOTION OBSERVATIONS
- Number
- 20200105056
- Published
- 2020-04-02
- Filed
- 2018-09-27
- Assignee
- Disney Enterprises, Inc.
- Inventors
- MITCHELL; Kenneth J., DÜMBGEN; Frederike, LIU; Shuang
- CPC
- G06T7/50; G06T17/20; G06T7/251; G06T15/503; G06N3/09; G06N3/0464; G06T17/10; G06N3/08
- Verdict
- Low Notable software
- Source
- Google Patents · FreePatentsOnline
The keeper's note
CV dense 3D reconstruction from narrow-baseline motion.
Abstract
Techniques for constructing a three-dimensional model of facial geometry are disclosed. A first three-dimensional model of an object is generated, based on a plurality of captured images of the object. A projected three-dimensional model of the object is determined, based on a plurality of identified blendshapes relating to the object. A second three-dimensional model of the object is generated, based on the first three-dimensional model of the object and the projected three dimensional model of the object.
Background
BACKGROUNDField of the Invention
Aspects of the present disclosure relate to reconstruction of a three-dimensional model of an object using two dimensional images, and more specifically, though not exclusively, to reconstruction of dense geometry from small camera or target object movement.Description of the Related Art
Estimating three-dimensional structures from two dimensional image sequences, sometimes referred to as structure-from-motion, traditionally requires large camera movements with large angle variation. That is, using traditional techniques, a three-dimensional model for a target object can be estimated using a series of two dimensional images of the object, but creation of an accurate three-dimensional model requires large variations in the location of the target object in the two dimensional images. Many digital cameras can capture a series of two dimensional images in a short window of time, for example capturing a burst of frames surrounding the primary image. But traditional structure-from-motion techniques are not suitable for generating an accurate three-dimensional model using these bursts of two dimensional images, because the movement of the camera and the target object is typically very slight (e.g., from accidental movement by the photographer or target), and so the images do not provide the desired large variations in the location of the target object. This can result in high depth uncertainty.SUMMARY
Embodiments described herein in