Runway Vision Pose Estimation With High-Confidence Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pose estimation methods for vision-based navigation systems in avionics face challenges in providing high-confidence error bounds due to the lack of reliable correspondence maps and the exponential growth of candidate pose solutions, leading to integrity issues and misleading information, especially in the absence of accurate optical pose estimation.
Innovation Solution
A vision-based navigation system that aligns 2D image features with a 3D constellation database using orthocorrection and reprojection techniques to determine a candidate correspondence map with high-confidence ambiguity identification, reducing the complexity of matching features by focusing on geometric relationships and error bounding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large set of candidate CMAPs is considered to ensure completeness, then the scope of candidate pose solutions is broadened, but precise high-confidence error bounding becomes impossible and system integrity deteriorates
Solution Approach 1:
The system performs preliminary orthocorrection of the image plane to align it with the constellation plane before feature matching. This preliminary geometric transformation reduces the search space for candidate CMAPs by pre-establishing the correct geometric relationship, allowing the system to focus computational effort on a reduced set of high-confidence candidates rather than exhaustively searching all possible correspondences
Solution Approach 2:
The system changes the geometric parameters of the image plane through orthocorrection transformation. By applying this parameter transformation, the image coordinates are converted to a coordinate system that matches the constellation database, enabling direct comparison and reducing the number of candidate correspondences that need to be evaluated
2Productivity
If residual monitoring thresholds are loosened to accept more candidate CMAPs, then availability improves, but erroneous or infeasible CMAPs are validated leading to loss of system integrity
Solution Approach 1:
Orthocorrection is applied as a preliminary step before residual monitoring. This pre-alignment of the image plane with the constellation plane using known geometric relationships establishes a strong prior that constrains the search space, allowing the system to maintain tight residual thresholds while still achieving high availability through the reduced candidate set
3Measurement precision
If the number of features to be matched increases to improve pose estimation accuracy, then the precision of pose estimation improves, but the computational complexity grows exponentially
Solution Approach 1:
The system performs orthocorrection as a preliminary transformation that aligns the image plane with the constellation plane before feature matching. This pre-alignment dramatically reduces the computational complexity by eliminating the need to evaluate exponential numbers of geometric transformations, allowing the system to match a large number of features efficiently
Solution Approach 2:
By changing the coordinate system parameters through orthocorrection, the system transforms the feature matching problem from a computationally intensive 3D-2D correspondence search into a simpler aligned-plane comparison, reducing complexity from exponential to polynomial growth with the number of features
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A vision-based navigation system (e.g., for aircraft on approach to a runway) captures via camera (102) 2D images of the runway environment in an image plane. The vision-based navigation system stores a constellation database (206) of runway features and their nominal 3D position information in a constellation plane. Image processors detect within the captured images 2D features potentially corresponding to the constellation features. The vision-based navigation system estimates optical pose of the camera (102) in the constellation plane by aligning the image plane and constellation plane into a common domain, e.g., via orthocorrection of detected image features into the constellation plane or reprojection of constellation features into the image plane. Based on the common-domain plane, the vision-based navigational system generates candidate correspondence maps (CMAP) of constellation features mapped to the image features with high-confidence error bounding, from which optical pose of the camera (102) or aircraft can be estimated.