3D Face Model Construction Using Alternating Particle Filters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional bundle adjustment techniques for constructing 3D models from images are hindered by blurry images, leading to reduced accuracy in the resulting models.
Innovation Solution
The use of particle filters, specifically alternating between an M particle filter for constructing a 3D user's face model and an E particle filter for determining extrinsic camera information, iteratively improves hypotheses for the 3D model and camera orientation, leveraging optical flow between images to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bundle adjustment technique is used to construct 3D model from images, then the method is simple and conventional, but the accuracy of the resulting 3D model degrades due to blurry images
Solution Approach 1:
The patent divides the 3D model construction process into two separate particle filters: an E-particle filter for estimating extrinsic camera parameters and an M-particle filter for constructing the 3D face model. This segmentation allows each filter to specialize in specific tasks, improving overall accuracy while managing complexity through modular design.
Solution Approach 2:
The E-particle filter performs preliminary estimation of extrinsic camera information before the M-particle filter constructs the 3D model. This preliminary action provides accurate camera parameter estimates that enhance the subsequent 3D model construction, addressing the accuracy issue caused by blurry images.
2Measurement precision
If particle filters are used to improve 3D model accuracy, then measurement precision improves, but computational complexity increases
Solution Approach 1:
By splitting the computational workload into two separate particle filters (E-particle filter for camera parameters, M-particle filter for 3D model), the patent manages computational complexity through modular architecture while maintaining high measurement precision in each specialized filter.
Solution Approach 2:
The particle filter system uses optical flow information from the images to guide its own particle weighting and resampling processes. This self-service mechanism allows the algorithm to adaptively focus computational resources on the most informative regions, improving accuracy without proportionally increasing overall complexity.
3Loss of information
If optical flow information is exploited between images, then information utilization improves and accuracy enhances, but processing time may increase
Solution Approach 1:
The system performs preliminary optical flow estimation between images before feeding this information into the particle filters. This preliminary action extracts useful motion information that guides the particle weighting process, improving information utilization while keeping processing time manageable through efficient pre-processing.
Data Source
AI summary
Constructing a user's face model using particle filters is disclosed, including: using a first particle filter to generate a new plurality of sets of extrinsic camera information particles corresponding to respective ones of a plurality of images based at least in part on a selected face model particle; selecting a subset of the new plurality of sets of extrinsic camera information particles corresponding to respective ones of the plurality of images; and using a second particle filter to generate a new plurality of face model particles corresponding to the plurality of images based at least in part on the selected subset of the new plurality of sets of extrinsic camera information particles.