Object Pose Estimation Using Segmented 3D Search Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for estimating the position and attitude of a target object in a real space require searching a 6-axis space, leading to increased computational cost and time due to the extensive search space.
Innovation Solution
An information processing apparatus that acquires depth information and captured images using sensors, generates candidate solutions by referencing two-dimensional data from both depth information and a three-dimensional model, and calculates the position and attitude using matching processes between these data sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full 6-axis space search is performed to estimate position and attitude, then measurement precision is improved, but computational time and computational cost increase significantly
Solution Approach 1:
The patent segments the 6-axis search space into two separate spaces: a 3-axis position space and a 3-axis attitude space. This segmentation allows independent processing of position and attitude estimation, reducing the computational complexity from searching the entire 6-axis space to searching two smaller 3-axis spaces separately, thereby decreasing computational time while maintaining estimation precision.
Solution Approach 2:
The patent transforms the problem by introducing a dimensional decomposition approach, where the original 6-dimensional search space is broken down into two separate 3-dimensional search spaces. This dimensionality change enables more efficient computation by reducing the search space volume from O(N^6) to O(N^3) + O(N^3), significantly reducing computational time while preserving measurement precision.
2Measurement precision
If a full 6-axis space search is performed to estimate position and attitude, then measurement precision is improved, but computational cost increases significantly
Solution Approach 1:
The patent segments the 6-axis search space into two separate spaces: a 3-axis position space and a 3-axis attitude space. This segmentation allows independent processing of position and attitude estimation, reducing the computational complexity from searching the entire 6-axis space to searching two smaller 3-axis spaces separately, thereby decreasing computational cost while maintaining estimation precision.
Solution Approach 2:
The patent transforms the problem by introducing a dimensional decomposition approach, where the original 6-dimensional search space is broken down into two separate 3-dimensional search spaces. This dimensionality change enables more efficient computation by reducing the search space volume from O(N^6) to O(N^3) + O(N^3), significantly reducing computational cost while preserving measurement precision.
Data Source
AI summary
At least one processor of an information processing apparatus performs: a depth information acquiring process of acquiring depth information; a captured image acquiring process of acquiring a captured image; a generating process of generating, with reference to first two-dimensional data and a three-dimensional model regarding the target object, a candidate solution regarding at least one selected from the group consisting of the position and the attitude of the target object in a three-dimensional space, the first two-dimensional data being obtained with reference to the depth information; and a calculating process of calculating, with reference to second two-dimensional data and the three-dimensional model and with use of the candidate solution, at least one selected from the group consisting of the position and the attitude of the target object in the three-dimensional space, the second two-dimensional data being obtained with reference to the captured image.