A trajectory completion method, device, apparatus and storage medium

By using a deep learning network model and an inner-outer loop training method, and employing LSTM and heading angle deviation algorithms to complete vehicle trajectories, the problem of discontinuous trajectory data is solved, achieving efficient and accurate trajectory completion, and supporting traffic analysis and autonomous driving.

CN121743694BActive Publication Date: 2026-06-05CHINA XIONGAN GRP DIGITAL CITY TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA XIONGAN GRP DIGITAL CITY TECH CO LTD
Filing Date
2026-02-27
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

The existing trajectory data is discontinuous due to equipment failure, signal obstruction and other reasons during collection, which leads to the destruction of the spatiotemporal integrity of the trajectory and affects the accuracy of traffic analysis and autonomous driving planning.

Method used

A deep learning network model is adopted to extract time, space and motion features by acquiring historical trajectory data in multiple scenarios. The trajectory completion model is trained by alternating internal and external training methods. Long Short-Term Memory (LSTM) network is used for trajectory completion, and missing points are handled by combining heading angle deviation and Bézier curve interpolation algorithms.

Benefits of technology

It achieves efficient and high-precision completion of discontinuous vehicle trajectory data, providing a continuous and reliable trajectory data foundation to support traffic analysis and autonomous driving planning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121743694B_ABST
    Figure CN121743694B_ABST
Patent Text Reader

Abstract

The application provides a trajectory completion method and device, equipment and a storage medium, and relates to the technical field of data processing. The method comprises the following steps: acquiring historical trajectory data in multiple scenes, wherein the historical trajectory data comprises multiple trajectory sequences, and each trajectory sequence comprises multiple trajectory points arranged in time sequence; extracting time, space and motion features of each trajectory point of each trajectory sequence to obtain multi-dimensional feature data of each trajectory sequence, wherein the multi-dimensional feature data comprises time features, space features and motion features of all trajectory points in the corresponding trajectory sequence; training a trajectory completion model by adopting an inner-outer loop alternating training mode based on the multi-dimensional feature data of each trajectory sequence; acquiring a trajectory sequence to be completed, and completing the trajectory sequence to be completed based on the trajectory completion model to obtain a complete trajectory sequence. The application realizes efficient and high-precision completion of discontinuous trajectories.
Need to check novelty before this filing date? Find Prior Art