CornerVINS: Accurate Localization and Layout Mapping for Structural Environments Leveraging Hierarchical Geometric Representations
and Yihong Wu
AI summary
Problem
Existing visual-inertial navigation systems struggle with pose drift and ambiguous data association in structured indoor environments due to the limited discriminability of traditional point and infinite plane features.
Approach
CornerVINS fuses depth-enhanced points, bounded plane patches, and novel 6-D box corners within an extended Kalman filter framework to provide robust, long-term geometric constraints for pose estimation and mapping.
Key results
- First parameterization and integration of 6-D box corners into a visual-inertial system
- Hierarchical mechanism for robust extraction and association of planes and corners
- Depth-enhanced point triangulation that improves initial pose estimation and tracking
- Demonstrated superior accuracy and computational efficiency over state-of-the-art systems
Why it matters
Enables reliable, high-level scene understanding and robust navigation for indoor service robots and autonomous systems in man-made environments.
Abstract
A compact and consistent map of surroundings is crit- ical for intelligent robots to understand their situations and realize robust navigation. Most existing techniques rely on infinite planes, which are sensitive to pose drift and may lead to confusing maps. Toward high-level perception in indoor environments, we propose CornerVINS, an innovative RGB-D inertial localization and layout mapping method leveraging hierarchical geometric features, i.e., points, planes, and box corners. Specifically, points are enhanced by fusing depth information, and planes are modeled as bounded patches using convex hulls to increase their discriminability. More importantly, box corners, lying at the intersection of three orthog- onal planes, are parameterized with a 6-D vector and integrated into the extended Kalman filter for the first time. We introduce a hierarchical mechanism to effectively extract and associate planes and corners, which are considered as layout components of scenes and serve as long-term landmarks to correct camera poses. Exten- sive experiments prove that the proposed box corners bring signif- icant improvements, enabling accurate localization and consistent layout mapping at low computational cost. Overall, the proposed CornerVINSoutperformsstate-of-the-artsystemsinbothaccuracy and efficiency.