SLAM Map Stitching Using Sparse Feature Relocalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing simultaneous localization and mapping (SLAM) techniques face challenges in stitching together old and new maps when there are changes in camera position and angle, leading to computational intensity and potential failure in tracking the camera's position, especially when the new scene has not been previously mapped.
Innovation Solution
The approach involves initializing a new SLAM map and using sparse feature comparison between keyframes of the old and new maps, employing algorithms like approximate nearest neighbor search and random sample consensus to identify matched features and determine a candidate camera position, allowing for successful stitching even with varying camera angles and positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If direct comparison of two-dimensional keyframe images is used to stitch old and new SLAM maps, then the stitching may work when keyframes are nearly identical, but the technique is computationally intensive and may fail when camera position or angle changes
Solution Approach 1:
The patent segments the keyframe images by extracting sparse features (corners, edges, distinctive locations) rather than comparing entire images. This segmentation reduces the comparison task from full image analysis to feature point matching, significantly decreasing computational intensity while maintaining stitching reliability across varying camera positions and angles.
Solution Approach 2:
The patent extracts only the essential distinctive features from keyframes for comparison purposes. By taking out only the relevant feature points (corners, edges) rather than processing complete images, the system achieves efficient stitching that works reliably even when camera position or angle changes, avoiding the computational burden of full image comparison.
2Productivity
If full sized keyframes are downscaled to reduce computational intensity, then map-stitching becomes faster, but the ability to uniquely identify viewpoints is reduced
Solution Approach 1:
The patent replaces the mechanical approach of downscaling entire images with a computational feature extraction method. Instead of reducing image resolution physically, the system extracts sparse feature points that retain viewpoint identification capability while requiring minimal computational resources for matching, thus achieving fast stitching without sacrificing viewpoint discrimination accuracy.
3Speed
If the camera moves quickly to a new unmapped scene, then exploration is efficient, but camera tracking may be lost and relocalization may fail
Solution Approach 1:
The patent performs preliminary feature extraction and storage during the mapping phase, creating a database of sparse features from keyframes before tracking failures occur. When the camera moves quickly to new scenes, this pre-extracted feature data enables rapid relocalization by comparing current frame features against the stored feature database, maintaining reliability even during high-speed exploration transitions.
Data Source
AI summary
In some examples, a camera tracking failure in a simultaneous localization and mapping (SLAM) process with respect to a first SLAM map may be identified. Responsive to identification of the camera tracking failure, a second SLAM map for the SLAM process may be initialized. A video frame tracked through the second SLAM map may be accessed. Matched features between the video frame and individual keyframes of the first SLAM map may be identified to determine a keyframe subset. The keyframe subset may be analyzed to determine a candidate camera position from among the keyframe subset. The candidate camera position may be tracked with respect to the first SLAM map for subsequent video frames. The first SLAM map may be stitched to the second SLAM map responsive to tracking the candidate camera position with respect to the first SLAM map for the subsequent.