Robot Visual Positioning With Point-Line Features and Reduced Pose Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile robots using visual positioning face challenges in achieving long-term reliable operation due to a narrow visual field, poor robustness to dynamic object occlusion, and environmental changes such as lighting and climate variations, leading to high error rates in feature matching and pose estimation.
Innovation Solution
A globally optimal robot visual positioning method based on point-line features, which involves acquiring a priori three-dimensional maps, calculating pitch and roll angles using gravity acceleration, detecting and matching two-dimensional point-line features, and decoupling the pose estimation problem into rotation and translation components using branch-and-bound algorithms and one-dimensional search enumeration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional visual positioning methods based on feature point matching are used, then the positioning system can operate with low cost and high information extraction capability, but the positioning accuracy and robustness deteriorate under environmental changes such as lighting variations, occlusion, and seasonal changes
Solution Approach 1:
The patent combines multiple feature types (point features and line features) into a unified positioning framework. By merging these different feature representations, the system achieves more robust positioning that is less sensitive to environmental changes, occlusion, and lighting variations compared to traditional point-feature-only methods
Solution Approach 2:
The patent creates a composite feature representation by integrating point features and line features. This composite approach类似于using composite materials, where the combination of different feature types provides superior performance under varying environmental conditions compared to single feature types
2Measurement precision
If guaranteed optimal algorithms based on BnB are used to ensure optimal pose estimation, then positioning accuracy is improved, but the computational speed deteriorates due to enumeration and search in 6DoF space
Solution Approach 1:
The patent segments the 6DoF pose estimation problem into two independent sub-problems: rotation estimation (using BnB algorithm) and translation estimation (using voting method). This segmentation allows each sub-problem to be solved more efficiently than the full 6DoF problem, maintaining accuracy while improving speed
Solution Approach 2:
The patent transforms the 6DoF search space into a more manageable form by separating rotation and translation components. This dimensional transformation enables the use of different optimization strategies for each component, reducing the overall computational complexity while preserving estimation accuracy
3Adaptability or versatility
If feature matching is performed in dynamic occlusion and textureless environments, then the system attempts to maintain positioning capability, but the feature extraction quantity decreases and matching error rate increases
Solution Approach 1:
The patent merges point feature matching and line feature matching into a unified framework. This combination compensates for the weaknesses of individual feature types in challenging environments, maintaining positioning capability when one feature type may fail due to occlusion or lack of texture
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method improves positioning accuracy and robustness by utilizing both point and line features, reducing search space dimensionality, and enhancing adaptability to environmental changes, especially in scenarios with limited visual point features.
Implementation Method 1
calculating a pitch angle and a roll angle of a current pose of the robot according to the currently measured gravity acceleration of the robot and the measurement data of gravity acceleration in the priori three-dimensional map
Data Source
AI summary
Provided is a globally optimal robot visual positioning method and device based on point-line features. The method comprises the following steps: acquiring a priori three-dimensional map of a current scene constructed in advance; acquiring a current image of the robot and the inertial measurement data; calculating a pitch angle and a roll angle of the current robot pose according to the current inertial sensor data and the inertial sensor data in the priori map; matching the two-dimensional point-line features detected in the current image with three-dimensional point-line features in a priori map; separating the rotation and translation of the pose to be solved according to the matched feature pairs, solving the rotation and then solving the translation so as to complete the dimensionality reduction of the search space.