Articulated Model Registration Using Segmented Detection and Closed-Form Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for three-dimensionally restoring articulated models of bodies from images, such as hands or whole bodies, using deep learning techniques face high computational complexity and power consumption, leading to long processing times and a risk of local minimum issues.
Innovation Solution
An articulated model registration method and apparatus that detects body images, extracts landmarks, generates a three-dimensional body shape model, and estimates positions on a three-dimensional coordinate using a conversion estimating unit and posture estimating unit, reducing computational load and avoiding local minimum problems by employing a reference straight line and closed-form solutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If inverse kinematics technique with Jacobian matrix or conjugate gradient method is used to three-dimensionally restore articulated model, then estimation accuracy is improved, but computation amount increases and processing time is extended
Solution Approach 1:
The patent segments the articulated model into multiple body parts (hand, upper arm, forearm, etc.) and processes each part separately using detection units. This segmentation reduces the overall computational complexity by breaking down the complex inverse kinematics problem into smaller, more manageable sub-problems that can be solved independently and then combined.
Solution Approach 2:
The patent performs preliminary detection of two-dimensional landmarks and body part positions before conducting three-dimensional restoration. By pre-detecting key points and establishing initial configurations, the system reduces the computational burden during the actual 3D reconstruction phase, avoiding the need for complex iterative optimization from scratch.
2Measurement precision
If inverse kinematics technique is used to three-dimensionally restore articulated model, then estimation accuracy is improved, but power consumption of portable device increases
Solution Approach 1:
By dividing the articulated model into separate body parts and processing each independently, the patent reduces the overall computational load and energy consumption. Each detection unit handles a specific body part with fewer calculations compared to processing the entire model as a single complex system.
Solution Approach 2:
The patent employs simplified detection models for each body part that can be quickly computed and discarded, rather than using a single complex iterative optimization model. This approach trades off some theoretical optimality for significantly reduced computational cost and energy consumption, making it suitable for portable devices.
3Measurement precision
If inverse kinematics technique with cost function optimization is used, then articulated model restoration is achieved, but processing time is excessively long
Solution Approach 1:
The patent segments the restoration process into parallel detection units for different body parts, enabling simultaneous processing rather than sequential optimization. This parallelization dramatically reduces processing time while maintaining accuracy, as each body part can be restored independently and concurrently.
Solution Approach 2:
The system performs preliminary detection of 2D landmarks and body part configurations before 3D restoration, establishing initial conditions that eliminate the need for time-consuming iterative optimization during the main restoration phase. This pre-processing step accelerates the overall processing speed significantly.
4Measurement precision
If cost function optimization using Jacobian matrix is used, then articulated model estimation is improved, but risk of local minimum problem increases
Solution Approach 1:
By segmenting the articulated model into separate body parts with independent detection units, the patent reduces the dimensionality of each optimization problem. Smaller, segmented optimization problems have fewer local minima and are less prone to convergence issues compared to a single large-scale optimization problem.
Solution Approach 2:
The patent establishes preliminary 2D landmark detections and initial body part configurations before performing 3D restoration. These preliminary detections provide reliable starting points that guide the optimization process away from local minima and toward the global optimum, improving convergence reliability.
Data Source
AI summary
The present invention relates to an articulated model registration method and apparatus which three-dimensionally restore an articulated model of a body, such as a hand, an upper body, and a whole body, from an image and to acquire a two-dimensional landmark from a two-dimensional image to generate and estimate a three-dimensional articulated model corresponding thereto.