A fusion positioning system of heterogeneous dual-domain fixed-capacity asynchronous playback

By using a heterogeneous dual-domain fixed-capacity asynchronous playback fusion positioning system, the problems of long-term error accumulation and low resource utilization in visual-inertial fusion positioning systems are solved, achieving autonomous, high-precision, and long-term stable positioning, which is applicable to fields such as UAVs, unmanned vehicles, robots, and aerospace embedded terminals.

CN122360488APending Publication Date: 2026-07-10UNIV OF ELECTRONICS SCI & TECH OF CHINA
-1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
UNIV OF ELECTRONICS SCI & TECH OF CHINA
Filing Date
2026-04-28
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing deep learning-based visual-inertial fusion positioning systems suffer from long-term positioning error accumulation, poor generalization ability to invisible environments, and weak perception of changing environments. Existing correction methods rely on external signals and cannot adapt to complex electronic warfare environments. Existing online learning methods suffer from catastrophic forgetting, high computational resource consumption, poor real-time performance, and cannot be adapted to resource-constrained devices.

Method used

A heterogeneous dual-domain fixed-capacity asynchronous playback fusion positioning system is adopted. The system is divided into a strong real-time processing OS domain and a weak real-time AI OS domain through a heterogeneous partitioning processing module. This enables real-time data acquisition, asynchronous playback training of parallel deep fusion models, and real-time visual fusion inertial navigation inference. By combining diverse fixed-capacity playback buffers and asynchronous playback training methods, the system is collaboratively designed to achieve autonomous and high-precision positioning.

Benefits of technology

It achieves fully autonomous, high-precision, and long-term stable visual-inertial navigation fusion positioning, adapts to the hardware resource constraints of intelligent agents, improves positioning real-time performance, system robustness and environmental adaptability, eliminates dependence on external signals, and is suitable for complex electronic warfare and dynamically changing environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122360488A_ABST
    Figure CN122360488A_ABST
Patent Text Reader

Abstract

This invention relates to the fields of deep learning and positioning technology, and discloses a heterogeneous dual-domain fixed-capacity asynchronous playback fusion positioning system, comprising: a strong real-time processing OS domain, used to collect environmental image data and agent motion posture data, perform data preprocessing, and then transmit the data to a data storage module; a weak real-time AI OS domain, used to retrieve historical and real-time environmental images and agent motion posture data stored in the data storage module, perform parallel asynchronous playback training of the deep fusion model and real-time visual fusion inertial navigation inference, so as to realize real-time positioning analysis of environmental images and agent motion posture data; and a data storage module, used to receive several data transmitted from the strong real-time processing OS domain and the weak real-time AI OS domain, and perform storage, classification management, and transmission. This invention not only achieves fully autonomous, high-precision, and long-term stable visual-inertial navigation fusion positioning, but also improves positioning real-time performance, system robustness, and environmental adaptability.
Need to check novelty before this filing date? Find Prior Art