A simultaneous localization and mapping system based on neural implicit representation

By combining a hybrid scene encoder, a cascaded localization optimizer, and a structured regularizer, the problem of balancing reconstruction accuracy and efficiency in visual synchronous localization and mapping using neural implicit representations is solved, generating high-quality, geometrically reasonable scene models suitable for robot autonomous navigation.

CN121783126BActive Publication Date: 2026-07-21ZHEJIANG UNIV OF TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG UNIV OF TECH
Filing Date
2026-03-05
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing visual simultaneous localization and mapping (SSL) technologies based on neural implicit representations face challenges in balancing reconstruction accuracy and computational efficiency. They also lack effective utilization of the inherent structured geometric priors of the scene, resulting in reconstruction results that are not smooth and reasonable, making it difficult to maintain high accuracy in complex environments.

Method used

A hybrid scene encoder based on an attention mechanism is adopted to fuse mesh structure and feature plane structure. Combined with a cascaded localization optimizer and a structured regularizer, a high-quality implicit scene representation is generated through multi-level spatial encoding, feature fusion, pose optimization and explicit geometric prior constraints.

Benefits of technology

It achieves high-precision and high-efficiency synchronous localization and mapping in complex environments. The generated scene model has both high detail and excellent geometric rationality, which improves the robustness and stability of the system and is suitable for autonomous navigation of robots in real complex scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121783126B_ABST
    Figure CN121783126B_ABST
Patent Text Reader

Abstract

The application discloses a kind of systems for simultaneous localization and mapping based on neural implicit representation, it is related to machine vision and intelligent navigation field, the system includes: scene coding module, for by multilevel space coding structure to input sensor data is analyzed, the implicit geometry and appearance of scene are generated Representation;Positioning optimization module is used to jointly estimate and optimize the pose of robot based on feature matching and the implicit representation of the scene coding module;Structured regularization module is used to extract structured geometric elements from sensor data, and accordingly construct regularization loss function to constrain the map reconstruction process of the scene coding module.The scheme of the present application can enhance the robustness and accuracy of the system in complex, dynamic or texture missing environment, generate dense continuous map with high detail and excellent geometric rationality, realize the effective balance of memory occupation and reconstruction quality.
Need to check novelty before this filing date? Find Prior Art