Volumetric Reconstruction for Occluded Object Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pose prediction methods in robotics and automated systems face reduced accuracy in cluttered or densely populated environments, especially when multiple objects are in contact or partially occluded, impairing robot performance.
Innovation Solution
A computer-implemented method that processes image data and depth information to generate volumetric reconstructions, occupancy data, and pointwise feature data for objects, allowing for accurate pose estimation by considering both the target object and its surroundings, using a combination of two-dimensional and three-dimensional feature extractors and regression models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pose prediction methods (template-based, sparse feature-based, dense methods, or hybrid methods) are used, then the methods can process images and depth information, but the accuracy of pose prediction is reduced when multiple objects are in contact or partially occluded
Solution Approach 1:
The patent segments the scene into individual object volumetric reconstructions and separates the target object from surrounding objects through occupancy grid processing. This segmentation allows the system to focus computational resources on the target object while accounting for occlusions from other objects, thereby maintaining high pose prediction accuracy in cluttered environments.
Solution Approach 2:
The patent introduces an occupancy grid as an intermediary data structure that represents the spatial occupancy of all objects in the scene. This intermediary allows the system to reason about occlusions and contacts between objects without directly processing complex object interactions, improving pose prediction accuracy by providing a simplified representation of the cluttered environment.
2Measurement precision
If volumetric reconstructions and occupancy data are generated for all objects, then pose prediction accuracy improves in cluttered environments, but computational complexity increases
Solution Approach 1:
The patent applies local quality by generating high-fidelity volumetric reconstructions only for the target object while using a simplified occupancy grid representation for surrounding objects. This selective approach maintains pose prediction accuracy for the target object while reducing overall computational complexity by avoiding full volumetric reconstruction of all objects in the scene.
Solution Approach 2:
The patent performs partial volumetric reconstruction by focusing computational resources on generating detailed volumetric representations only for objects that are relevant to the pose prediction task (i.e., the target object and immediately surrounding objects causing occlusions). This partial action approach achieves sufficient accuracy without the excessive computational cost of processing all objects in full detail.
Data Source
AI summary
A computer-implemented method of estimating a pose of a target object in a three-dimensional scene includes: obtaining image data and associated depth information representing a view of the three-dimensional scene; processing the image data and the associated depth information to generate a volumetric reconstruction for each of a plurality of objects in the three-dimensional scene, including the target object; determining a volumetric grid containing the target object; generating, using the generated volumetric reconstructions, occupancy data indicating portions of the volumetric grid occupied by free space and portions of the volumetric grid occupied by objects other than the target object; and estimating the pose of the target object using the generated occupancy data and pointwise feature data for a plurality of points on a surface of the target object.