3D Facial Reconstruction Using Blendshape Priors for Narrow Baseline Motion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional structure-from-motion techniques are not suitable for generating accurate three-dimensional models of facial geometry from two-dimensional images captured with small camera or target object movement, as facial areas are often smooth and texture-less, leading to high depth uncertainty and unreliable key point extraction.
Innovation Solution
The method employs facial geometry priors, such as blendshapes, to impose constraints on the model and refine the three-dimensional reconstruction, improving accuracy and robustness for dense facial reconstruction using narrow baseline motion observations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional structure-from-motion techniques are used to reconstruct three-dimensional models from two-dimensional images, then the method can handle large camera movements with large angle variation, but it fails to produce accurate models when camera movement is small (narrow baseline)
Solution Approach 1:
The patent applies preliminary action by pre-defining facial geometry priors and blendshapes before the reconstruction process. These pre-established geometric constraints and facial expression models are prepared in advance to guide the narrow baseline reconstruction, enabling accurate depth estimation even when camera movement is minimal.
Solution Approach 2:
The patent changes parameters by transitioning from traditional structure-from-motion approaches to a method that incorporates facial geometry priors and blendshape coefficients. This parameter transformation allows the system to work effectively with narrow baseline observations by leveraging learned facial geometric constraints rather than relying solely on camera motion parallax.
2Reliability
If traditional key point extraction methods are used on facial images, then the process is simple, but it produces unreliable results on smooth and texture-less facial areas
Solution Approach 1:
The patent introduces facial geometry priors and blendshape models as intermediaries between the two-dimensional images and the three-dimensional reconstruction. These intermediaries provide geometric constraints that bridge the gap caused by insufficient texture information in facial areas, enabling reliable key point extraction and depth estimation without directly relying on image texture.
3Manufacturing precision
If dense reconstruction is performed from narrow baseline observations, then the model captures fine facial geometry, but traditional methods produce high depth uncertainty
Solution Approach 1:
The patent implements feedback by using facial geometry priors and blendshape models to constrain and refine the reconstruction process. The pre-defined facial geometric constraints provide continuous feedback during optimization, guiding the narrow baseline reconstruction toward anatomically plausible solutions and reducing depth uncertainty in the resulting three-dimensional model.
Data Source
AI summary
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.