Cross-Detector Visual Localization with Coplanarity Constraints for Indoor Environments.

Area of Science:

• Computer Vision
• Robotics
• Geospatial Analysis

Background:

• Current visual localization (VL) methods require query and map keypoints to use identical detectors.
• This limits map interoperability and the adoption of new, improved keypoint detectors.
• Rebuilding maps is often impractical due to data unavailability or privacy concerns.

Purpose of the Study:

• To formalize and address the challenge of cross-detector visual localization.
• To enable the use of heterogeneous keypoint detectors within a single map representation.
• To ensure the long-term usability and interoperability of feature-based maps.

Main Methods:

• Introduced the problem of cross-detector VL, highlighting spatial discrepancies hindering correspondences.
• Proposed CoplaMatch, a novel approach relaxing descriptor similarity constraints.
• Leveraged 2D homographies and geometric coplanarity constraints by segmenting planar patches.

Main Results:

• CoplaMatch effectively enables accurate visual localization across different keypoint detectors.
• Demonstrated superior performance compared to two state-of-the-art methods in cross-detector scenarios.
• Showcased the feasibility of cross-detector VL without hindering online applicability.

Conclusions:

• CoplaMatch overcomes limitations of traditional VL by enabling detector heterogeneity.
• The proposed method validates the feasibility and opens new avenues for cross-detector visual localization.
• This research enhances the long-term utility and adaptability of feature-based geospatial maps.