Weighted Iterative Fundamental Matrix for Vision Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for identifying true feature matches in vision-based navigation, such as RANSAC and extended Kalman filter, face challenges with high computational load and difficulty in distinguishing false matches from noise, especially in uncalibrated camera scenarios, leading to poor performance and errors in navigation solutions.
Innovation Solution
A method that iteratively calculates a fundamental matrix using weighted estimation, updates weights based on distance errors, and selects true feature matches after a specified number of iterations, reducing the impact of false matches and improving accuracy without requiring camera calibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional methods like RANSAC and extended Kalman filter are used to identify true feature matches, then the navigation system can operate without camera calibration, but the computational load is high and false matches cannot be effectively distinguished from noise
Solution Approach 1:
The patent applies preliminary action by performing iterative outlier rejection before final navigation solution computation. The method pre-identifies and removes false matches through multiple iterations of fundamental matrix estimation and chi-square testing, ensuring that only true feature matches proceed to the navigation calculation stage. This preliminary filtering action resolves the contradiction by establishing reliable match identification without requiring camera calibration.
Solution Approach 2:
The patent implements feedback through iterative refinement where the fundamental matrix estimation and outlier rejection process repeatedly uses previous results to improve accuracy. Each iteration uses the chi-square statistic to provide feedback on match quality, adjusting the set of candidate matches and re-estimating the fundamental matrix until convergence or maximum iterations are reached. This feedback mechanism enables reliable distinction between true and false matches without calibration.
2Adaptability or versatility
If conventional methods are used to identify feature matches, then the system can process uncalibrated camera data, but the computational load increases significantly
Solution Approach 1:
The patent applies partial action by performing iterative outlier rejection for a limited number of iterations (typically 3-10 iterations) rather than exhaustively processing all possible match combinations. This partial iteration approach provides sufficient computational effort to distinguish true from false matches while avoiding excessive computational burden. The method processes only the necessary subset of candidate matches through the iterative refinement process.
Solution Approach 2:
The patent extracts and removes false matches from the candidate set through iterative outlier rejection. By calculating the fundamental matrix and chi-square statistics, the method identifies and extracts false matches (outliers) from the mixed set of true and false matches, leaving only true matches for final navigation computation. This extraction process reduces computational load by eliminating erroneous matches early in the pipeline.
3Measurement precision
If iterative fundamental matrix calculation with weighted estimation is performed, then true feature matches can be accurately identified, but the processing time increases with the number of iterations
Solution Approach 1:
The patent performs preliminary action by conducting iterative outlier rejection before final match selection. The iterative process with weighted fundamental matrix estimation is executed in advance to establish a reliable set of true feature matches, which then serves as the foundation for subsequent navigation calculations. This preliminary accurate identification prevents time loss in later stages by ensuring match quality upfront.
Solution Approach 2:
The patent applies partial action by limiting the iterative process to a reasonable number of iterations (typically 3-10) rather than continuing until perfect convergence. This partial iteration provides sufficient measurement precision for practical navigation applications while avoiding excessive processing time. The method achieves adequate accuracy without the diminishing returns of extended iteration sequences.
Data Source
Figure 1
Figure 2A
Figure 2B~2C
AI summary
An example embodiment includes a method (300) for identifying true feature matches from a plurality of candidate feature matches for vision based navigation. A weight for each of the plurality of candidate feature matches can be set (304). The method also includes iteratively performing for N iterations: calculating (306) a fundamental matrix for the plurality of candidate feature matches using a weighted estimation that accounts for the weight of each of the plurality of candidate feature matches; calculating (308) a distance from the fundamental matrix for each of the plurality of candidate feature matches; and updating (312) the weight for each of the plurality of candidate feature matches as a function of the distance for the respective candidate feature match. After N iterations candidate feature matches having a distance less than a distance threshold can be selected (316) as true feature matches