Object Pose Tracking via 3D Model Contour Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object pose tracking systems face challenges with complexity, such as objects with multiple parts or rapid movement, leading to poor robustness and increased calculation time, especially when using feature points or lines, resulting in difficulty in maintaining real-time tracking.
Innovation Solution
A computer-implemented method for object pose tracking that involves obtaining an initial pose, determining a viewing angle, searching for a corresponding 3D model, calculating a projected contour, and iteratively correcting the pose using the Levenberg-Marquardt algorithm to improve accuracy and efficiency, while reducing the need for frequent initialization calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature points or lines are used for tracking, then tracking is effective for slowly moving objects, but robustness deteriorates when object speed increases or when many similar straight line features exist
Solution Approach 1:
The patent changes the tracking parameter from traditional feature points/lines to contour features. This parameter transformation enables the system to maintain robustness for rapidly moving objects while preserving tracking accuracy, directly resolving the contradiction between reliability and measurement precision under different motion conditions
Solution Approach 2:
The patent implements a dynamic tracking approach by continuously updating the contour model and adapting to object motion. The system dynamically adjusts tracking parameters and performs real-time contour extraction, allowing it to maintain both robustness and precision whether objects are stationary or moving rapidly
2Adaptability or versatility
If too many features are adopted for tracking, then tracking coverage is improved, but calculation time is greatly increased affecting real-time performance
Solution Approach 1:
The patent extracts only the essential contour features from objects, removing unnecessary internal features and details. By focusing solely on the outer contour boundary, the system achieves comprehensive tracking coverage while significantly reducing the number of features that need to be processed, thereby maintaining real-time performance
Solution Approach 2:
The patent segments the tracking problem into two parts: (1) extracting contour features from the image, and (2) matching these contour features with the 3D model. This segmentation allows efficient processing by handling only the essential contour information separately from other object details, improving both coverage and processing speed
3Measurement precision
If initialization calculations are performed frequently to maintain accuracy, then pose tracking precision is improved, but system complexity and calculation burden increase
Solution Approach 1:
The patent performs preliminary action by pre-building a 3D model of the target object from multiple views before tracking begins. This pre-processing step stores the essential geometric information needed for tracking, eliminating the need for frequent initialization calculations during runtime and reducing system complexity while maintaining accuracy
Solution Approach 2:
The patent creates a 3D model copy of the target object that can be repeatedly used for tracking without requiring re-initialization. This digital twin or virtual model serves as a reference that can be matched against incoming images continuously, replacing the need for frequent initialization calculations and reducing computational burden
Data Source
AI summary
The present disclosure relates to robot technology, which provides an object pose tracking method and apparatus. The method includes: obtaining an initial pose of the object, and determining a first viewing angle of an object with respect to a camera based on the initial pose; searching for a first 3D model corresponding to the first viewing angle; calculating a projected contour of an initial pose using the searched first 3D model, and obtaining feature point(s) of the projected contour; calculating a pose change amount based on the feature point(s) of the projected contour and feature point(s) of a real contour; and performing an iterative calculation on the pose change amount to obtain a pose correction amount, and obtaining a corrected pose of the object based on the pose correction amount. In the calculation process, the 3D model can be selected to improve the calculation speed by fewer feature points.