Camera Pose Estimation Using Equidistant Contour Points

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing camera pose estimation methods in manufacturing, such as those used for robots in aircraft production, face challenges in accurately determining the position and orientation of cameras relative to complex objects, especially when objects are featureless or have limited distinctive features, which can lead to incorrect robot positioning and potential damage.

Innovation Solution

A system and method that utilize a monocular camera to capture a 2D image of an object, identify a unique feature, segment the contour into equidistant points, align these points with a 3D reference model, and perform perspective-n-point calculations and bounded nonlinear optimization to refine the camera pose estimation, minimizing projection errors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional silhouette matching with rendered 3D model is used for camera pose estimation, then the method can work with simple models, but it fails when objects are featureless or have limited distinctive features

Engineering Contradiction:
Improvecamera pose estimation accuracyVSAvoidapplicability to featureless objects
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent segments the object contour into multiple equidistant points rather than treating it as a continuous silhouette. This discretization allows the system to work with specific geometric features (the equidistant points) even on featureless objects, transforming the continuous contour matching problem into a discrete point correspondence problem between 2D image and 3D model.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from 2D silhouette matching to 3D point cloud alignment by synthesizing equidistant points along the contour and matching them with corresponding points on the 3D reference model. This dimensional enhancement allows the system to exploit depth information and 3D geometry, improving pose estimation reliability for objects lacking distinctive 2D features.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple unique features are required on the object for accurate pose estimation, then estimation accuracy improves, but the method becomes inapplicable to featureless objects

Engineering Contradiction:
Improvecamera pose estimation precisionVSAvoidobject feature requirements
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent performs preliminary synthesis of equidistant points along the object contour before pose estimation. By pre-defining these points based on contour geometry rather than requiring pre-existing features, the system eliminates the need for manual feature placement or object modification, making it applicable to featureless objects while maintaining precision through the mathematical regularity of equidistant point distribution.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates a virtual copy of the object contour as a point cloud by synthesizing equidistant points, which is then matched against the 3D reference model. This synthetic point cloud serves as a feature-rich representation even for featureless objects, allowing accurate pose estimation without requiring physical features to be present or manufactured on the actual object.

Inventive Principle:
Principle #26Copying

3Ease of operation

If contour is divided into equidistant segments starting from a unique feature, then point alignment with 3D model becomes feasible, but the method requires at least one unique feature to start the segmentation

Engineering Contradiction:
Improvecontour segmentation processVSAvoidapplicability to completely featureless objects
Core Design Contradiction:
Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent introduces asymmetry by selecting a specific starting point on the contour (such as the topmost point in image coordinates) to begin equidistant segmentation. This asymmetric reference point provides a consistent, deterministic starting position for point synthesis without requiring distinctive object features, enabling the method to work on completely featureless objects while maintaining operational simplicity through a clear, repeatable starting criterion.

Inventive Principle:
Principle #4Asymmetry

Data Source

PatentEP3229208B1Camera pose estimation
Publication Date: 2018.10.31 THE BOEING CO
  • EP3229208B1 patent drawingFigure 1
  • EP3229208B1 patent drawingFigure 2
  • EP3229208B1 patent drawingFigure 3

AI summary

A system (20) for determining a camera pose relative to an object (22) including a unique feature (42) may include a monocular camera (24) configured to produce an image (48) of the object (22), and a processing unit (26) in operative communication with the monocular camera (24). The processing unit (26) may be configured to: identify the unique feature (42) of the object (22) in the image (48) produced by the monocular camera (24), synthesize at least four points (50) along a contour (40) of the obj ect (22) in the image (48) using the identified unique feature (42) as a starting point (52), synthesize a same number of points (60) as synthesized in the image (48) along a contour (40) of the object (22) in a reference model (56), the reference model (56) preprogrammed into a memory (28) of the processing unit (26), correlate the points (60) from the reference model (56) to the image (48), and determine a pose of the monocular camera (24) based on the correlated points (60).