Stereo Egomotion Voting Strategy via Dimensionality Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating ego-motion from stereo images are limited by their suitability only for low-dimensional problems, high memory and computational costs, and inefficiencies in handling imprecise measurements, especially in complex scenes with moving objects.
Innovation Solution
The method decomposes the six-dimensional ego-motion estimation into two two-dimensional subproblems, allowing for robust computation of camera rotation and translation using a voting strategy, where motion vectors are weighted by 3D distance, and triangulation errors are explicitly modeled to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a voting strategy is used for ego-motion estimation, then measurement robustness is improved, but computational cost and memory requirements increase due to the large number of accumulator bins required
Solution Approach 1:
The patent segments the six-dimensional ego-motion estimation problem into two separate two-dimensional subproblems: one for camera rotation and another for camera translation. This segmentation reduces the accumulator dimensionality from 6D to 2D, dramatically decreasing memory requirements and computational cost while maintaining the robustness of the voting strategy. Each subproblem uses its own reduced-dimensional accumulator, making the system feasible for real-time implementation.
Solution Approach 2:
The patent transforms the high-dimensional ego-motion estimation problem into lower-dimensional subproblems by changing the dimensionality of the search space. Instead of directly estimating six-dimensional motion parameters, the system decomposes the problem into two 2D problems, effectively reducing the computational burden from O(N^6) to O(N^2) complexity while preserving measurement robustness through the voting mechanism.
2Measurement precision
If the accumulator resolution is increased to improve precision, then measurement precision is improved, but memory requirements and computational cost increase significantly
Solution Approach 1:
The patent divides the precision requirements into two separate 2D accumulators instead of one large 6D accumulator. Each 2D accumulator can be resolved at high precision with manageable memory requirements, whereas a single high-resolution 6D accumulator would require prohibitively large memory. The segmentation allows high precision to be achieved practically by distributing the resolution across multiple lower-dimensional structures.
3Adaptability or versatility
If all six ego-motion parameters are estimated simultaneously, then completeness of solution is improved, but device complexity and difficulty of robust estimation increase
Solution Approach 1:
The patent segments the ego-motion estimation into two independent phases: first estimating rotation parameters (three parameters) and then estimating translation parameters (three parameters) given the rotation. This segmentation reduces system complexity by breaking down the coupled 6D estimation problem into two decoupled 3D problems, each further reduced to 2D subproblems, making the overall system more manageable and robust while maintaining completeness.
Solution Approach 2:
The patent performs preliminary estimation of camera rotation before estimating camera translation. By first solving the rotation subproblem and using its results to inform the translation subproblem, the system simplifies the overall estimation process. This preliminary action reduces complexity by fixing certain parameters before solving for others, transforming a complex simultaneous 6D estimation into a sequence of simpler staged estimations.
Data Source
AI summary
Methods and systems for egomotion estimation (e.g. of a vehicle) from visual inputs of a stereo pair of video cameras are described. 3D egomotion estimation is a six degrees of freedom problem in general. In embodiments of the present invention, this is simplified to four dimensions and further decomposed to two two-dimensional sub-solutions. The decomposition allows use of a voting strategy that identifies the most probable solution. An input is a set of image correspondences between two temporally consecutive stereo pairs, i.e. feature points do not need to be tracked over time. The experiments show that even if a trajectory is put together as a simple concatenation of frame-to-frame increments, the results are reliable and precise.