Relocalization algorithm based on common object planar representation and semantic descriptor matching

By generating a semantic topology graph and using hybrid descriptors for node matching, the robustness problem of the relocalization algorithm under changes in lighting and perspective is solved, and high-precision and stable relocalization effects are achieved, which is suitable for AR and navigation applications.

CN115330861BActive Publication Date: 2025-09-09NORTHEASTERN UNIV CHINA
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210813525.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-11
Publication Date
2025-09-09
Estimated Expiration
2042-07-11

AI Technical Summary

Technical Problem

Existing relocalization algorithms have poor robustness under changes in lighting and perspective, and insufficient single semantic information leads to descriptor matching failure, making it difficult to achieve high-precision and stable relocalization.

Method used

A relocalization algorithm based on common object planar representation and semantic descriptor matching is adopted to generate a semantic topology map and extract hybrid descriptors. Node matching is performed using a graph propagation algorithm combined with semantic neighborhood information, and matching accuracy is improved through attribute and relationship descriptors.

Benefits of technology

The accuracy and robustness of relocalization are improved, and stable relocalization can be achieved in large parallax and dynamic scenes, meeting real-time requirements, and significantly reducing feature extraction and matching time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115330861B_ABST
    Figure CN115330861B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of relocalization technology, and proposes a relocalization algorithm based on common representation of object planes and matching of semantic descriptors. This algorithm extracts semantic objects and planar landmarks from the point cloud map, and then uses the landmarks to construct a semantic topological map. Unlike other brute force graph matching or random walk-based graph matching algorithms, the present invention proposes to use a graph propagation algorithm to combine the node and edge information of the semantic neighborhood to generate a hybrid descriptor. Subsequently, the sKM algorithm is used to match the landmark points of the prior point cloud map and the query point cloud map to improve the accuracy and robustness of the matching, and the matching constraints of the landmark points are used to solve the relocalization pose of the query point cloud map. The algorithm can effectively cope with relocalization work in large parallax or dynamic scenes to serve application scenarios such as AR and navigation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Object-oriented visual SLAM lightweight semantic map creating method

    CN113160401A

  • Mobile robot indoor repositioning method, computer device and product

    CN114526739A