3D Motion Estimation Using Expectation Maximization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the motion of a multiple camera system in 3D space, such as ego-motion estimation in vehicles, are prone to inaccuracies and require significant processing power, often relying on RANSAC algorithms and additional sensors to maintain accuracy, which can lead to errors and high computational demands.
Innovation Solution
The method represents motion parameters in the group of rigid-body motion in 3D and applies an expectation maximization algorithm for local linearization, filtering out outlier motion samples and focusing on reliable sensory information, thereby reducing computation time and improving robustness, using techniques like minimal squared distances and quaternion representations for efficient motion estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RANSAC algorithms and additional sensors are used to maintain accuracy in motion estimation, then measurement precision is improved, but device complexity and processing power requirements increase
Solution Approach 1:
The patent extracts and removes outlier motion samples from the dataset before performing motion estimation. By identifying and eliminating samples with large residuals that deviate from the expected motion pattern, the system achieves accurate motion estimation without requiring complex RANSAC algorithms or additional sensors, thus reducing device complexity while maintaining precision
Solution Approach 2:
The patent performs preliminary filtering of outlier samples before the main motion estimation process. By pre-processing the data to remove erroneous measurements, the subsequent motion estimation can be performed more efficiently with simpler algorithms, reducing both computational complexity and processing requirements while maintaining accuracy
2Measurement precision
If RANSAC algorithms are used to filter outlier landmarks, then measurement precision is improved, but processing power and computation time increase
Solution Approach 1:
The patent extracts outlier landmarks by computing residuals for each landmark correspondence and identifying those that deviate significantly from the expected pattern. By removing these outlier samples before motion estimation, the system achieves precise results without requiring computationally intensive RANSAC algorithms, thus reducing processing power requirements
Solution Approach 2:
The patent applies a simplified filtering approach that processes only the necessary subset of landmarks to identify outliers, rather than performing exhaustive RANSAC iterations on all possible landmark combinations. This partial action approach achieves sufficient precision with significantly reduced computational power
3Measurement precision
If multiple initial motion estimates are generated and evaluated, then measurement precision is improved, but productivity and processing efficiency decrease
Solution Approach 1:
The patent extracts and removes outlier motion samples that would otherwise require multiple iterative estimates to identify and discard. By pre-filtering these outliers based on residual analysis, the system reduces the number of iterations needed for convergence, thereby improving processing efficiency while maintaining the reliability gained from evaluating multiple estimates
Data Source
Figure 1~2
Figure 3
Figure 4a~4b
AI summary
The invention relates to a method of estimating a motion of a multiple camera system in a 3Dspace, wherein the field of view of multiple cameras at least partially coincide. The method comprises the step of calculating a multiple number of initial motion estimates for a motion of the multiple camera system between the time instant associated with a particular image set and a time instant of a subsequent image set, based on 3D position differences of image features in images of the particular set and the subsequent set, captured by aspecific camera. The method further comprises the steps of characterizing the motion by motion parameters represented in the group of rigid-body motion in 3D, and determining an optimal set of motion parameters in the group of rigid-body motion in 3D.