Geoarc Vehicle Localization Using Angular Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for vehicle localization, such as GPS-denied environments, face challenges with accuracy and efficiency, particularly in outdoor settings where active sensing systems are impractical and passive camera-based approaches are computationally complex and error-prone.
Innovation Solution
The use of geoarcs, which are three-dimensional mathematical models associating determined angles of view with feature pairs, to constrain possible camera locations, allowing for robust and efficient localization by accumulating constraints from multiple geoarcs, even with errors in angle measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scene matching methods are used to match camera imagery with geotagged imagery, then localization can be achieved, but the computational complexity increases and error propagation occurs due to the need to match many individual features
Solution Approach 1:
The patent extracts only the essential geometric constraints (angles between feature pairs) from the full image matching problem, discarding the need to match all individual features. By taking out just the angular relationships between corresponding feature pairs in the query and reference images, the system achieves localization without the computational burden of complete scene matching.
Solution Approach 2:
The patent replaces the mechanical feature-matching process with a geometric constraint-based system. Instead of relying on complex image processing and feature correspondence algorithms, the system substitutes a mathematical approach using angular measurements and geometric relationships to directly compute camera position and orientation.
2Adaptability or versatility
If correlation-based approaches are used for image matching, then localization can be performed, but the system becomes sensitive to rotation, scale, perspective, and lighting variations
Solution Approach 1:
The patent changes the parameters used for matching from intensity-based correlations to geometry-based angular measurements. By measuring angles between feature pairs rather than comparing pixel intensities, the system becomes invariant to lighting variations, scale changes, and perspective distortions, while maintaining high localization accuracy through precise angular constraint propagation.
3Measurement precision
If SLAM systems are used to create 3D models and estimate camera position, then localization can be achieved, but the computational complexity increases significantly
Solution Approach 1:
The patent extracts only the necessary angular constraints from image data without performing full 3D reconstruction. By taking out just the angle measurements between feature pairs and using these to directly constrain camera position, the system avoids the computationally intensive processes of creating and processing complete 3D models while still achieving accurate localization.
4Measurement precision
If inertial measurement units are used for localization, then short-term positioning can be achieved, but the location estimate error grows over time
Solution Approach 1:
The patent introduces a geometric constraint system based on angular measurements as an intermediary between image data and camera localization. This mediator provides absolute geometric constraints that do not accumulate error over time, unlike inertial systems, allowing the system to maintain accurate localization indefinitely by periodically constraining the solution space with new angular measurements from feature pairs.
Data Source
Figure 1~11
Figure 2~12
Figure 3~5
AI summary
A method of spatial localization, comprising: (i) acquiring from a sensor supported by a first vehicle of a group of vehicles, an image including a portion of a physical surface of each vehicle of a subset of the group of vehicles, from a position spaced away from the subset of the group of vehicles; (ii) identifying features in the acquired image; (iii) associating identified features in the acquired image with identified features in a three-dimensional reference representation of the subset of the group of vehicles; (iv) selecting a plurality of pairs of features in the acquired image having corresponding identified pairs of features in the reference representation; (v) for each selected feature pair in the acquired image, determining an angle of view between the features of the feature pair, and generating a three-dimensional geoarc associating the determined angle of view and the feature pair in the reference representation corresponding to the selected feature pair; and (vi) identifying locations where the geoarcs for the selected feature pairs overlap.