Outlier Filtering in 8-Point Camera Motion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current camera motion estimation algorithms, such as the 8-point algorithm combined with RANSAC, face challenges in accurately filtering outliers, leading to unreliable results due to the presence of bad solutions, especially when many outliers are present in the data sets used for estimating vehicle rotation and translation.
Innovation Solution
A method is introduced to filter outliers by selecting and scaling image points, building an 8-by-9 matrix, determining its condition number, and rejecting matrices with condition numbers greater than or equal to a threshold, thereby minimizing errors and improving the robustness of the RANSAC algorithm by reducing the number of outliers sent to it.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the standard 8-point algorithm combined with RANSAC is used for camera motion estimation, then the algorithm can handle outliers in the data, but when many bad solutions are present, the RANSAC algorithm cannot reliably decide which solution is good, leading to unreliable results
Solution Approach 1:
The patent applies preliminary action by computing and evaluating the condition number of the essential matrix before passing data to RANSAC. This pre-evaluation step filters out bad solutions in advance, allowing RANSAC to focus only on promising candidates and reliably identify the correct motion parameters even when many outliers are present in the original data.
2Adaptability or versatility
If RANSAC is used to estimate motion parameters from data containing outliers, then the algorithm can produce results with certain probability, but the presence of many bad solutions prevents reliable decision-making
Solution Approach 1:
The patent introduces the condition number evaluation as an intermediary step between the 8-point algorithm and RANSAC. This intermediary filters and ranks candidate solutions based on their numerical stability, providing RANSAC with pre-filtered, high-quality candidates and enabling reliable solution selection even in the presence of numerous outliers.
3Quantity of substance
If all computed solutions from the 8-point algorithm are sent to RANSAC, then comprehensive coverage of possible motions is achieved, but the computational burden increases and reliability decreases when many bad solutions are present
Solution Approach 1:
The patent extracts and removes bad solutions from the set of candidates before passing them to RANSAC by evaluating the condition number of each candidate's essential matrix. This extraction of poor-quality solutions reduces the computational burden on RANSAC and improves the reliability of the final motion estimation by ensuring only promising candidates are considered.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
A method to filter outliers in an image-aided motion-estimation system is provided. The method includes selecting eight-image-points in a first image received from a moving imaging device at at least one processor; selecting eight-image-points in a second image that correspond to the selected eight-image-points in the first image at the at least one processor, the second image being received from the moving imaging device; scaling the selected image-points at the at least one processor so the components of the selected image-points are between two selected values on the order magnitude of 1; building an 8-by-9 matrix (A) from the scaled selected image-points at the at least one processor; determining a condition number for the 8-by-9 matrix at the at least one processor; and rejecting the 8-by-9 matrix built from the selected image-points when the determined condition number is greater than or equal to a condition-number threshold.