Robot Vision Position Estimation Using Common Structure Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the position/orientation of target objects in robot assembly, such as model fitting, face challenges with local minima issues and high processing costs due to the need for multiple initial value candidates and alignment processes, especially when similar shapes are observed.
Innovation Solution
The technique involves generating a common structure model with geometric features common among verification position/orientation candidates, using this model to align with a reference model, and estimating the position/orientation by verifying the alignment against an image, thereby reducing the need for individual alignment processes for each candidate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple position/orientation candidates are generated and alignment is carried out for each candidate, then the accuracy of position/orientation estimation is improved, but the processing cost increases significantly
Solution Approach 1:
The patent segments the alignment process by identifying and extracting only the common geometric features that appear in multiple position/orientation candidates. Instead of performing full alignment for each candidate, the system divides the work into: (1) extracting common features once, (2) using these features for initial position/orientation estimation, and (3) performing refined alignment only if necessary. This segmentation reduces redundant computations while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary action by extracting common geometric features from the target object before generating multiple position/orientation candidates. These pre-extracted features are stored and reused during the alignment process for all candidates, eliminating the need to re-extract features for each candidate. This preliminary feature extraction significantly reduces processing cost while preserving estimation accuracy.
2Productivity
If model fitting is performed with a single initial value, then the processing cost is reduced, but the accuracy deteriorates due to local minima problems
Solution Approach 1:
The patent addresses the symmetry problem by detecting asymmetric geometric features of the target object and using them to generate position/orientation candidates that account for similar shapes. When symmetric features are detected, the system intentionally creates multiple candidates with different orientations (including flipped versions) to break the symmetry assumption. This allows the system to explore multiple potential solutions and select the correct one, avoiding local minima while maintaining reasonable processing cost.
Solution Approach 2:
The patent introduces common geometric features as an intermediary between the image data and the multiple position/orientation candidates. These extracted features serve as a mediator that provides reliable initial estimates for all candidates without requiring full alignment processing. The intermediary feature set bridges the gap between single-initial-value efficiency and multi-candidate accuracy, enabling the system to start multiple fitting processes from informed initial values.
3Reliability
If full alignment is performed for each verification position/orientation candidate, then the reliability of position/orientation estimation is improved, but the processing time increases
Solution Approach 1:
The patent applies partial action by performing complete alignment processing only for the most promising position/orientation candidates rather than all candidates. The system first performs a quick verification using extracted common features to evaluate all candidates, then performs full alignment only for those that pass a threshold criterion. This partial processing approach maintains reliability for the final result while significantly reducing overall processing time by avoiding redundant full alignment operations.
Data Source
AI summary
A plurality of verification position/orientation candidates for a target object is set. A common structure model including a geometric feature of a part, among geometric features of a reference model representing a three-dimensional shape of the target object, that is common among the candidates is generated. An image including the target object is obtained. A position/orientation of the target object is estimated by verifying the common structure model and the reference model arranged at the plurality of verification position/orientation candidates, against the image.