3D Data Field Conversion for Machine Vision Pose Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine vision techniques face inefficiencies in processing three-dimensional data, particularly in searching for patterns, due to the need to access neighboring points in 3D data, which interrupts parallelization and reduces processing speed.
Innovation Solution
The method involves converting three-dimensional data into a densely-populated field where each cell has associated values, allowing for the determination of representative data and testing model poses by summing dot products of probes with field vectors, thereby avoiding the need to search for neighboring points and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional three-dimensional data processing methods are used to search for patterns, then the system can identify pattern locations and orientations, but the processing speed is reduced due to the need to access neighboring points in 3D data which interrupts parallelization
Solution Approach 1:
The patent creates a copied representation of three-dimensional data by generating a two-dimensional projection or map that contains encoded information about neighboring points. This copy allows probe evaluations to access spatial relationships without requiring random access to the original 3D data structure, thereby enabling continuous parallelization and improving processing speed while maintaining pattern recognition accuracy.
2Measurement precision
If a multi-step approach with different resolution models is used to refine pattern pose identification, then the accuracy of pose identification is improved, but the processing time increases due to iterative refinement steps
Solution Approach 1:
The patent performs preliminary processing by pre-computing and storing a two-dimensional representation of the three-dimensional data that encodes neighboring point information. This preliminary action allows subsequent pose identification steps to operate more efficiently on the pre-processed data structure, reducing the computational burden of iterative refinement while maintaining high accuracy in pose identification.
Solution Approach 2:
The patent transforms three-dimensional data into a two-dimensional representation that preserves essential spatial relationships. This dimensionality change allows the system to maintain measurement precision for pose identification while reducing the computational complexity and processing time associated with operating directly in three-dimensional space during iterative refinement steps.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The techniques described herein relate to methods, apparatus, and computer readable media configured to perform various aspects, including testing a pose of a model to image data, such as testing a pose of a model in three-dimensional data. Three-dimensional data of an object is received, the three-dimensional data comprising a set of data entries. The three-dimensional data is converted to a field comprising a set of cells that each have an associated value, comprising determining, for each cell value, representative data based on one or more data entries from the set of data entries of the three-dimensional data. A pose of the model is tested with the field to determine a score for the pose.