Multi-Camera Pose Estimation for Cluttered Bin Picking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine vision systems face challenges in accurately determining the positions and orientations of objects in cluttered environments, particularly when objects are randomly piled in bins, due to expensive 3D laser scanning technologies, high computational requirements, and limitations in precision and reliability, especially for texture-poor objects.
Innovation Solution
A method involving a machine vision system with multiple cameras that process images to identify features at specific depth planes by eliminating background clutter and using homography transformations to isolate features consistently located across camera views, enabling accurate 6D pose estimation of objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D laser scanning technologies are used for pose estimation, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent uses multiple 2D camera images as copies of the object from different viewpoints, replacing the need for expensive 3D laser scanners. By processing these 2D image copies through homography transformations and feature matching, the system achieves pose estimation without requiring complex 3D scanning hardware.
Solution Approach 2:
The patent replaces mechanical 3D laser scanning systems with an optical-based computer vision system using standard 2D cameras. The mechanical complexity of laser scanners is substituted by computational methods including homography calculations, feature detection, and pose estimation algorithms processed by a computer.
2Measurement precision
If 3D laser scanning or structured light projectors are used, then measurement precision is improved, but use of energy and computational resources increase
Solution Approach 1:
The patent extracts only the essential 2D image data from the environment using simple cameras, removing the need for energy-intensive 3D scanning hardware and complex structured light projection. The system processes only the necessary visual information from multiple 2D views to achieve pose estimation.
Solution Approach 2:
The patent uses inexpensive 2D digital cameras instead of expensive, energy-consuming 3D laser scanners. These standard cameras consume minimal energy and can be easily replaced or repositioned, providing a cost-effective and energy-efficient alternative to complex 3D scanning systems.
3Loss of information
If stereo vision approaches with multiple cameras are used, then depth information is obtained, but reliability deteriorates due to erroneous correspondences
Solution Approach 1:
The patent implements feedback through iterative feature matching and homography verification. Detected features from multiple camera views are matched and verified against each other, with inconsistent correspondences being identified and corrected. The system refines pose estimates by continuously checking feature consistency across all camera views.
Solution Approach 2:
The patent performs preliminary feature detection and homography calculation for each camera view before final pose estimation. By pre-processing the images to identify stable features and calculate preliminary homographies, the system establishes a reliable foundation for accurate pose determination, reducing errors from incorrect correspondences.
4Ease of operation
If template-based approaches are used for pose estimation, then ease of operation is improved, but adaptability deteriorates for texture-poor objects
Solution Approach 1:
The patent enhances features on texture-poor objects by analyzing intensity variations, edges, and gradients in the 2D images. By detecting subtle intensity changes and edge features rather than relying on rich textures or colors, the system can identify and match features on objects with minimal surface detail, improving adaptability to diverse object types.
Data Source
AI summary
Methods and apparatus for determining poses of objects acquire plural images of the objects from different points of view. The images may be obtained by plural cameras arranged in a planar array. Each image may be processed to identify features such as contours of objects. The images may be projected onto different depth planes to yield depth plane images. The depth plane images for each depth plane may be compared to identify features lying in the depth plane. A pattern matching algorithm may be performed on the features lying in the depth plane to determine the poses of one or more of the objects. The described apparatus and methods may be applied in bin-picking and other applications.