3D Pose Estimation for Picking Small Objects With Dense Correspondence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pose estimation systems in robotics face challenges in accurately determining the location and orientation of small objects and deformable objects due to limitations in image resolution, capture speed, and field of view, leading to high error margins that can result in missed grasps or improper object manipulation.
Innovation Solution
The system employs a method that computes dense image-to-object correspondences using optical flow and disparity maps to refine the pose estimation of objects by aligning a 3-D model with the observed object, utilizing a Perspective-n-Point algorithm to minimize error and account for deformations, enabling accurate detection of small and deformable objects' configurations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pose estimation methods are used, then the system is simple and fast, but the measurement precision is insufficient for small objects
Solution Approach 1:
The patent segments the pose estimation process into multiple stages: initial pose estimation using traditional methods, followed by refinement using dense image-to-object correspondences. This segmentation allows the system to achieve high precision for small objects by applying complex methods only where needed, rather than using complex methods for all objects.
Solution Approach 2:
The patent applies dense correspondence computation selectively - not to all objects in the scene, but specifically to refine poses of small objects where high precision is critical. This partial application of the complex method reduces overall computational burden while achieving the required precision for challenging cases.
2Measurement precision
If image resolution is increased to improve detection accuracy, then the measurement precision improves, but the capture speed decreases
Solution Approach 1:
The patent processes images at different resolution levels for different purposes: lower resolution for initial object detection and localization, then higher resolution only for refining pose estimates of identified small objects. This segmentation of processing resolutions maintains capture speed while achieving necessary precision.
Solution Approach 2:
The patent performs preliminary pose estimation using lower-resolution images and traditional methods before applying computationally intensive dense correspondence refinement. This preliminary action at lower resolution reduces the data volume that requires high-resolution processing, maintaining capture speed while enabling subsequent precision improvement.
3Productivity
If the field of view is expanded to capture more objects, then the productivity increases, but the measurement precision for individual small objects decreases
Solution Approach 1:
The patent segments the scene into multiple regions and processes objects at different stages: initial detection across the full field of view, then refined pose estimation for small objects in specific regions of interest. This allows comprehensive scene coverage while maintaining high precision for critical small objects.
Solution Approach 2:
The patent applies high-precision dense correspondence methods locally to small objects where it matters most, rather than uniformly across all objects in the expanded field of view. This local application of high-quality processing maintains productivity by avoiding unnecessary computation on larger or less critical objects.
4Measurement precision
If dense image-to-object correspondences are computed, then the pose estimation accuracy improves, but the computational time increases
Solution Approach 1:
The patent computes dense image-to-object correspondences partially - only for objects identified as small or requiring high precision, rather than for all detected objects. This selective computation significantly reduces total processing time while maintaining high accuracy for critical cases.
Solution Approach 2:
The patent performs preliminary object classification and size assessment before committing to computationally intensive dense correspondence computation. This preliminary sorting allows the system to apply high-precision methods only when necessary, reducing overall computation time while preserving accuracy for small objects.
Data Source
AI summary
A method for estimating a pose of an object includes: receiving, by a processor, an observed image depicting the object from a viewpoint; computing, by the processor, an instance segmentation map identifying a class of the object depicted in the observed image; loading, by the processor, a 3-D model corresponding to the class of the object; computing, by the processor, a rendered image of the 3-D model in accordance with an initial pose estimate of the object and the viewpoint of the observed image; computing, by the processor, a plurality of dense image-to-object correspondences between the observed image of the object and the 3-D model based on the observed image and the rendered image; and computing, by the processor, the pose of the object based on the dense image-to-object correspondences.