Posture Estimation via Difficulty-Weighted Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing posture estimation methods face challenges in accurately estimating the posture of objects formed of multiple parts, particularly when parts are concealed, leading to increased possibilities of erroneous estimations due to high degrees of freedom in candidate postures.
Innovation Solution
A posture estimation apparatus and method that utilize an image input section, posture information database, difficulty level information table, and fitting section to calculate and weight correlation based on estimation difficulty levels, specifically using parallel line components to determine the accuracy of part positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the degree of freedom for concealed parts is increased to expand candidate postures, then the possibility of including correct posture increases, but the estimation accuracy decreases due to unnecessary expansion of candidate postures
Solution Approach 1:
The patent applies local quality by differentiating the treatment of parts based on their concealment status. Concealed parts are assigned higher degree of freedom to expand candidate postures, while non-concealed parts maintain lower degree of freedom to restrict unnecessary candidates. This localized differentiation resolves the contradiction by applying appropriate flexibility only where needed (concealed parts) while maintaining precision where possible (non-concealed parts).
Solution Approach 2:
The patent changes the parameter of degree of freedom dynamically based on concealment detection. When a part is detected as concealed, its degree of freedom parameter is increased to expand the search space for correct posture. When not concealed, the parameter remains low to maintain estimation precision. This conditional parameter change resolves the contradiction between reliability and measurement precision.
2Reliability
If the number of candidate postures is increased to account for concealed parts, then the robustness against estimation errors improves, but the computational complexity increases
Solution Approach 1:
The patent segments the posture estimation problem by treating concealed and non-concealed parts differently. By dividing parts into these two categories based on detectability, the system can expand candidate postures only for concealed parts while maintaining strict constraints for non-concealed parts. This segmentation reduces the overall computational complexity compared to expanding all candidate postures uniformly, while still achieving robustness against estimation errors.
3Ease of operation
If area threshold is used to identify concealed parts, then the detection process is simplified, but the accuracy of concealed part detection decreases
Solution Approach 1:
The patent introduces an intermediary concept of 'degree of freedom' that bridges the gap between simple area-based detection and accurate concealed part identification. Rather than directly using area threshold as the sole criterion, the system uses it as an initial indicator and then adjusts the degree of freedom parameter to refine the detection accuracy, accounting for cases where area alone may be misleading.
Data Source
AI summary
A posture estimation device that is capable of highly precisely estimating the posture of an object comprising multiple parts. Said device (100) comprises: a posture information database (110) that for each of multiple postures, holds posture information that defines the placement of multiple parts; a fitting unit (160) that computes, for each of the parts in an image, a correlation level between the placement of the parts and the posture information; a difficulty level information table (130) that holds an estimation difficulty level that is a degree of difficulty of estimating each part position and computed on the basis of each parallel line components of each of the parts contained in the posture information; and a posture estimation unit (170) that to the correlation level, applies a weighting based on the estimation difficulty level, and on the basis of the weighted correlation level, estimates the posture of the object.