Human Model Generation from Single Pose via Kinematic Space

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current markerless motion capture methods lack accuracy and efficiency in generating human models with precise joint center locations and seamless free form surface models, especially when using low-quality 3D reconstructions, and require multiple subject poses or dedicated hardware.

Innovation Solution

An automated method that combines a learnt kinematic model space with a shape model using a training dataset to generate subject-specific models from multiple video streams, allowing for the identification of joint centers and seamless mesh registration from a single static pose, incorporating iterative closest point algorithms and Levenberg-Marquardt optimization for accurate joint alignment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If passive methods with three cameras in orthogonal views are used, then the system can be implemented with standard video equipment, but the subject must perform precise movements and the model construction is complex

Engineering Contradiction:

Improveimplementation simplicityVSAvoidsubject movement requirements

Core Design Contradiction:

Ease of manufactureVSEase of operation

Solution Approach 1:

The system uses the subject's natural movements during normal activity to automatically identify joint centers and body landmarks, eliminating the need for the subject to perform specific precise movements or for manual initialization by operators

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces manual initialization and complex parametric model construction with automated functional methods that use motion capture data to compute joint centers and body segment parameters through mathematical optimization

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If functional methods are used to identify joint centers, then some joint centers can be estimated, but the model lacks ankle and wrist joints and biomechanical rigor

Engineering Contradiction:

Improvejoint identification efficiencyVSAvoidmodel accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The system applies a unified functional approach to identify all major joint centers including previously missing ankle and wrist joints, while simultaneously ensuring biomechanical rigor through proper alignment to reference poses and seamless mesh generation

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent uses parameter optimization through Levenberg-Marquardt algorithms to refine joint center locations and body segment parameters, transforming initial estimates into accurate final values that satisfy biomechanical constraints

Inventive Principle:

Principle #35Parameter changes

3Ease of manufacture

If ellipsoidal meta-balls are used for geometric representation, then the method is simple and efficient, but the surface is not seamless and requires equally spaced points for mesh registration

Engineering Contradiction:

Improvemodel generation simplicityVSAvoidsurface continuity

Core Design Contradiction:

Ease of manufactureVSShape

Solution Approach 1:

The patent replaces geometric primitive-based modeling with a functional approach that generates seamless free-form surfaces through mesh registration algorithms that create continuous surfaces with proper point correspondence

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system transforms the surface representation from discrete ellipsoidal meta-balls to continuous free-form surfaces by optimizing mesh correspondence parameters and ensuring seamless transitions between body segments

Inventive Principle:

Principle #35Parameter changes

4Manufacturing precision

If accurate model generation is achieved, then high shape fidelity is obtained, but the algorithm requires manual initialization and laser scan quality data

Engineering Contradiction:

Improvemodel accuracyVSAvoiddata quality requirements

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system automatically initializes all model parameters from motion capture data without manual intervention, making the process fully automated and applicable to various data qualities including standard video-based reconstructions

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent enables accurate model generation from lower-quality visual hull data derived from standard video cameras, replacing the requirement for expensive laser scan quality data while maintaining sufficient accuracy for markerless motion capture applications

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Data Source

PatentUS8180714B2Automatic generation of human models for motion capture, biomechanics and animation

Publication Date: 2012.05.15 POLITECNICO DI MILANO

AI summary

An automated method for the generation of (i) human models comprehensive of shape and joint centers information and/or (ii) subject specific models from multiple video streams is provided. To achieve these objectives, a kinematic model is learnt space from a training data set. The training data set includes kinematic models associated with corresponding morphological models. A shape model is identified as well as one or more poses of the subject. The learnt kinematic model space and the identified shape model are combined to generate a full body model of the subject starting from as few as one-static pose. Further, to generate a full body model of an arbitrary human subject, the learnt kinematic model space and the identified shape model are combined using a parameter set. The invention is applicable for fully automatic markerless motion capture and generation of complete human models.