3D Object Pose Estimation via Pre-computed Comparison Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional position/posture recognition apparatuses require extensive processing time to estimate the position or posture of an object due to the complexity of generating and comparing multiple position/posture estimation values, even when the illumination conditions vary.
Innovation Solution
An estimation system that includes image input means, 3D shape data storage, comparison image generation, image positional relationship detection, correction amount calculation, and state correction to reduce the number of comparison image generations and complexity in calculating the object's position or posture, by using 3D shape data and image displacement distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple position/posture estimation values are generated by varying parameters by predetermined variation, then the measurement precision of object position/posture is improved, but the processing time and device complexity increase significantly
Solution Approach 1:
The system performs preliminary action by generating comparison images for multiple position/posture estimation values in advance before actual object recognition. The posture candidate group determination means pre-calculates and stores multiple comparison images corresponding to different position/posture parameters, so that during actual recognition, these pre-generated comparison images can be directly used for matching without re-generating them, thereby significantly reducing processing time while maintaining estimation accuracy
Solution Approach 2:
The system segments the position/posture parameter space into discrete candidate values. Instead of continuously varying parameters, the system divides the parameter space into discrete steps and generates comparison images for specific discrete position/posture combinations. This segmentation allows the system to manage complexity by processing only a finite set of discrete candidates rather than continuous variations, reducing computational burden while maintaining sufficient precision for accurate object recognition
2Measurement precision
If multiple comparison images are generated for different position/posture values, then the measurement precision is improved, but the device complexity and calculation complexity increase
Solution Approach 1:
The comparison image generation means is designed with multi-functionality to handle different illumination conditions universally. The same comparison image generation mechanism can accommodate various lighting scenarios by adjusting illumination parameters, eliminating the need for separate processing chains for different lighting conditions. This universal approach reduces overall system complexity while maintaining the ability to generate accurate comparison images for position/posture estimation under diverse environmental conditions
Solution Approach 2:
The system manages complexity by parameterizing the generation of comparison images. Instead of generating images for all possible position/posture combinations exhaustively, the system uses parameter variation within defined ranges to create a manageable set of representative comparison images. The posture candidate group determination means adjusts parameters (position, posture, illumination) systematically to generate a sufficient subset of comparison images that cover the essential recognition scenarios, reducing calculation complexity while maintaining estimation accuracy
3Ease of operation
If the posture candidate group determination means generates position/posture estimation values by simple parameter variation, then the ease of operation is improved, but the measurement precision deteriorates due to lack of accurate position/posture information
Solution Approach 1:
The system implements feedback through the end determination means that continuously monitors the similarity between comparison images and input images. The feedback loop compares the estimated position/posture against the actual image data and adjusts the estimation accordingly. When the similarity metric indicates sufficient accuracy, the feedback mechanism terminates the search, providing a simple operation interface while ensuring measurement precision through data-driven validation rather than relying solely on predetermined parameter variations
Data Source
AI summary
3D model storage stores the 3D shape data of a target object and illumination base data in advance. A comparison image generator generates, as a comparison image, a reproduced image with the target object being arranged in the position/posture of the current estimation value under the same illumination condition as that for the input image on the basis of the 3D shape data and illumination base data. An image displacement distribution detector segments the comparison image into sub regions and detects the image displacement distribution between the comparison image and the input image for each sub region. A posture difference calculator calculates a position/posture difference value on the basis of the image displacement distribution and 3D shape data. An end determinator outputs the current position/posture estimation value as an optimum position/posture estimation value when determining that the position/posture difference value is smaller than a predetermined threshold value.