SLAM Error Control via Structure Re-identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing SLAM systems using cameras and inertial measurement units suffer from trajectory drift and poor robustness in constructing three-dimensional maps and determining device pose, especially in complex environments with multiple box worlds, leading to inaccurate Manhattan frame estimation.
Innovation Solution
A method that re-identifies points, lines, and Manhattan frames using a spatiotemporally-sensitive global sub map and pose guide verification, incorporating time and space consistency tests to improve the accuracy and robustness of SLAM results by integrating structure re-identification constraints into the bundle adjustment process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing SLAM systems use tracking-matching on feature points and lines, then device pose can be obtained, but trajectory drift occurs and robustness becomes poor
Solution Approach 1:
The patent segments the feature extraction process into two independent streams: point feature tracking and line structure tracking. Each stream operates separately to extract features and compute constraints, then both streams are integrated in the bundle adjustment process. This segmentation allows the system to leverage the strengths of both point and line features while mitigating their individual weaknesses, thereby improving trajectory robustness without sacrificing pose accuracy.
Solution Approach 2:
The patent creates a composite constraint model by combining point-based tracking constraints and line-based structure constraints. The bundle adjustment process integrates both types of constraints simultaneously, forming a hybrid optimization framework that leverages the complementary nature of point and line features. This composite approach enhances trajectory robustness by providing multiple independent constraints that can compensate for failures in individual feature types.
2Measurement precision
If structure re-identification constraint is integrated into bundle adjustment, then SLAM accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary structure re-identification and extraction of line constraints before the bundle adjustment process. By pre-computing the line structure constraints and re-identifying features in advance, the system prepares optimized input data for bundle adjustment, reducing the computational burden during the optimization phase. This preliminary action enables higher accuracy without proportionally increasing overall computational complexity.
3Manufacturing precision
If binocular camera is used instead of monocular camera, then three-dimensional map with actual physical scale can be constructed, but device complexity increases
Solution Approach 1:
The patent replaces complex mechanical calibration systems with a computational approach that uses line structure constraints to directly enforce metric scale. Instead of relying on complex camera calibration hardware and procedures, the system uses the geometric constraints provided by lines in the scene to recover metric scale information computationally. This substitution maintains three-dimensional mapping accuracy while reducing mechanical complexity.
Data Source
AI summary
Provided is a method of simultaneous localization and mapping (SLAM), the method including obtaining a structure of a current image frame, obtaining a structure re-identification constraint based on the structure of the current image frame, and obtaining a SLAM result based on the structure re-identification constraint. By introducing the structure re-identification constraint in a bundle adjustment process to obtain the SLAM result, error accumulation may be more easily controlled, and accuracy and robustness in obtaining the SLAM result may be improved.