Geometric Feature Matching for Visual Navigation Drift Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Visual navigation systems suffer from pose drift due to residual errors in pose computation, which accumulate over time, making it challenging to maintain accuracy in location determination and trajectory estimation.
Innovation Solution
The system generates and compares 3D maps of local and prior features, using geometric information to determine a transformation that maps recently observed features to their previously observed counterparts, thereby correcting for drift and facilitating loop closure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If loop closure is used to correct pose drift, then position accuracy is improved, but the search space for feature matching increases significantly
Solution Approach 1:
The patent segments the feature matching process into two distinct stages: (1) generating a candidate set of prior features within a threshold distance of current local features, and (2) determining correspondences between the candidate set and local features. This segmentation reduces the search space from all prior features to only those within a spatial threshold, making loop closure computationally feasible while maintaining position accuracy.
Solution Approach 2:
The patent applies local quality by focusing the search and matching process on features within a specific spatial threshold distance of current local features, rather than searching the entire environment. This localized approach prioritizes regions most likely to contain loop closure features, reducing computational complexity while maintaining matching accuracy.
2Ease of operation
If appearance based feature matching is used for loop closure, then feature identification is simplified, but matching accuracy deteriorates due to viewpoint and illumination changes
Solution Approach 1:
The patent introduces geometric information as an intermediary between appearance-based detection and final feature matching. The system first identifies candidate features using appearance-based methods, then uses geometric constraints (threshold distance and spatial arrangement) as an intermediary validation step to confirm matches, thereby maintaining simplicity while improving accuracy.
Solution Approach 2:
The patent combines multiple matching criteria into a composite approach: appearance-based feature detection combined with geometric constraint validation. This composite method leverages the simplicity of appearance-based identification while adding geometric verification to compensate for viewpoint and illumination changes, achieving both ease of operation and high accuracy.
3Reliability
If geometric matching is performed on all prior features, then matching completeness is improved, but computational efficiency deteriorates
Solution Approach 1:
The patent segments the prior features into two groups: (1) features within a threshold distance of current local features (candidate set), and (2) other prior features. The geometric matching is performed only on the candidate set, which significantly reduces the number of comparisons while maintaining completeness for relevant loop closure features.
Solution Approach 2:
The patent applies partial action by performing geometric matching only on the subset of prior features within a threshold distance, rather than all prior features. This partial approach is sufficient for detecting loop closure events while dramatically improving computational efficiency, avoiding the excessive action of checking all features.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A first map comprising local features and 3D locations of the local features is generated, the local features comprising visible features in a current image and a corresponding set of covisible features. A second map comprising prior features and 3D locations of the prior features may be determined, where each prior feature: was first imaged at a time prior to the first imaging of any of the local features, and lies within a threshold distance of at least one local feature. A first subset comprising previously imaged local features in the first map and a corresponding second subset of the prior features in the second map is determined by comparing the first and second maps, where each local feature in the first subset corresponds to a distinct prior feature in the second subset. A transformation mapping a subset of local features to a subset of prior features is determined.