SLAM Back-End Optimization Using Parallax-Weighted Reprojection Error
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing back-end optimization methods for simultaneous localization and mapping fail to accurately represent the relationship between reprojection error and depth of points, leading to low accuracy in pose estimation due to the lack of spatial information description.
Innovation Solution
The method involves determining a relative reprojection error by providing weights to the reprojection error through parallax, which is the baseline length between projection points on the camera frame, to enhance the robustness of the optimization process and improve spatial localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional reprojection error is used as loss function for back-end optimization, then the optimization process is simple, but the pose estimation accuracy is low due to lack of spatial information
Solution Approach 1:
The patent transforms the traditional reprojection error into a relative reprojection error by changing the parameter representation. Specifically, it uses parallax (baseline length between projection points) as a weighting factor to normalize the reprojection error, converting absolute pixel distance into a relative measure that accounts for spatial depth information. This parameter transformation enables the loss function to reflect both measurement accuracy and spatial relationships without requiring complex additional modules.
Solution Approach 2:
The patent introduces parallax as an intermediary element that mediates between the reprojection error and the optimization process. By calculating the baseline length between projection points of the same 3D point across different views and using it to weight the reprojection error, the parallax acts as a bridge that injects spatial depth information into the otherwise simple reprojection error metric, thereby improving pose estimation accuracy while maintaining optimization simplicity.
2Reliability
If reprojection error is used without spatial weighting, then the calculation is computationally efficient, but the relationship between reprojection error and depth of points is not accurately represented
Solution Approach 1:
The patent changes the parameter representation of reprojection error from absolute pixel distance to relative distance normalized by parallax. By dividing the reprojection error by the baseline length (parallax) between projection points, the patent creates a depth-aware error metric that accurately represents the relationship between reprojection error and point depth, improving reliability without requiring complex computational structures.
Solution Approach 2:
Instead of directly modeling the complex relationship between reprojection error and depth through multiple variables, the patent inverts the approach by using parallax (which is easily computable from camera geometry and point coordinates) as a weighting factor. This inversion simplifies the computational process while achieving accurate spatial relationship representation, as parallax naturally encodes depth information through the baseline length between views.
Data Source
AI summary
The disclosure provides a method, device and storage medium for back-end optimization of simultaneous localization and mapping. In the method, a target three-dimensional space point is determined in three-dimensional space points of a plurality of key frames located by a simultaneous localization and mapping system. In addition, a reprojection error is determined based on the target three-dimensional space point, and a weight is provided for the reprojection error by preprocessing the maximum parallax corresponding to the target three-dimensional space point, to obtain the relative reprojection error.