3D Enclosing Body Estimation from 2D Polygon Projections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D object recognition methods, particularly those based on machine learning, are resource-intensive and lack interpretability, requiring large amounts of annotated training data and manual effort, which is time-consuming and expensive, and the learned functions are 'black boxes' that do not provide insights into how they arrive at their results.
Innovation Solution
A computer-implemented method for determining a three-dimensional envelope using a mathematical optimization procedure with an objective function comprising fewer than 100,000 terms, which is interpretable and explainable, involving the steps of receiving two-dimensional minimally circumscribing polygon data, applying a projection matrix, and optimizing position, size, and orientation of the envelope using separate objective functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning-based 3D object recognition is used, then recognition accuracy is improved, but resource consumption and training time increase significantly
Solution Approach 1:
The patent segments the 3D object recognition task into multiple 2D recognition tasks from different camera viewpoints. Each 2D task is simpler and requires less training data, while the combination of multiple 2D results achieves accurate 3D object identification without the resource burden of direct 3D machine learning training
Solution Approach 2:
The patent introduces 2D image recognition results as an intermediary step between raw sensor data and final 3D object understanding. Instead of directly training complex 3D recognition models, the system uses 2D recognition from multiple angles as intermediate representations to infer 3D object properties
2Measurement precision
If black-box machine learning models are used for 3D recognition, then recognition performance is improved, but interpretability and accountability are lost
Solution Approach 1:
The patent breaks down the black-box 3D recognition into transparent 2D recognition steps from multiple viewpoints. Each 2D recognition step can be individually interpreted and validated, making the overall 3D recognition process explainable through the combination of these interpretable intermediate results
Solution Approach 2:
The patent replaces the abstract, opaque mathematical transformations of black-box machine learning with a more transparent geometric approach using projection matrices and coordinate transformations. This substitution makes the recognition process more interpretable while maintaining accuracy
3Measurement precision
If annotated training data is used for supervised learning, then learning accuracy is improved, but data preparation time and cost increase
Solution Approach 1:
The patent creates a multi-functional system where 2D image data serves multiple purposes: it provides training data for 2D recognition models, serves as input for generating 3D object hypotheses, and enables viewpoint transformation. This multi-use approach reduces the need for extensive specialized annotated training data
Solution Approach 2:
The patent uses 2D image copies from different camera viewpoints as proxies for expensive 3D annotated data. By transforming and combining these 2D copies through projection geometry, the system achieves 3D recognition capabilities without requiring direct 3D annotated training examples
Data Source
Figure 1
Figure 2~3
Figure 4~5b
AI summary
The invention relates to a computer-implemented method for determining a three-dimensional enclosing body (16), preferably a cuboid, for an object represented on first data (12), wherein the first data (12) comprise at least three spatial dimensions, wherein information data of a two-dimensional minimal surrounding polygon for the object represented on second data are received, a projection matrix which defines a mapping of a three-dimensional data point (14) of the first data (12) onto a two-dimensional data point in the second data is received, and wherein the three-dimensional enclosing body (16) of the object is determined by means of a mathematical optimization method and an objective function (Z1, Z2, Z3), wherein the objective function comprises fewer than 100,000 terms (T1, T2, T3, T1', T2', T1", T2").Furthermore, the invention relates to a data processing device comprising means for carrying out the above method, as well as a computer program product comprising instructions that, when executed by a computer, cause the computer to carry out the above method. The invention also relates to a computer-readable data carrier on which the above computer program product is stored.