Strong maneuvering target tracking method and device based on transformer structure

By employing a high-maneuverability target tracking method based on a transformer structure and utilizing self-attention and cross-attention to extract trajectory features, the problem of large state estimation error in high-maneuverability target tracking is solved, achieving higher tracking accuracy and real-time performance.

CN116563346BActive Publication Date: 2026-05-29TSINGHUA UNIVERSITY
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TSINGHUA UNIVERSITY
Filing Date
2023-06-14
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing technologies suffer from large target state estimation errors in tracking highly maneuverable targets, making it difficult to effectively capture the transition patterns between rapidly changing states, resulting in insufficient tracking accuracy and real-time performance.

Method used

A high-mobility target tracking method based on transformer structure is adopted. By encoding and embedding the measured trajectory, the transformer encoder is used to extract self-attention information and the transformer decoder is used to perform cross-attention fusion. The encoder features with different state information are obtained and residual trajectory prediction is performed to estimate the target trajectory.

Benefits of technology

It improves the accuracy and real-time performance of state estimation for highly maneuverable targets, reduces the difficulty of state estimation, enhances the ability to capture rapidly changing states, and improves tracking accuracy and performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116563346B_ABST
    Figure CN116563346B_ABST
Patent Text Reader

Abstract

A strong maneuvering target tracking method and device based on a transformer structure are provided in the application. The method comprises: encoding and position embedding of a measurement track to obtain track features with state differences; inputting the track features with state differences into a transformer encoder for single-layer state feature self-attention information extraction to obtain encoder features with different state information; inputting the encoder features with different state information and the track features with state differences into a transformer decoder for cross-attention extraction and fusion to obtain decoder features with fused different correlation information; and performing prediction processing based on the decoder features, adding a residual track obtained to a measurement track to obtain a target track. The method can learn the transition law of fast changing states better, improve the accuracy of state estimation of strong maneuvering targets, and ensure the performance of the target tracking task.
Need to check novelty before this filing date? Find Prior Art