A social graph driving-based vehicle networking trajectory prediction method

By using a social graph-driven approach, a social tensor graph is constructed and multi-scale convolution and Gaussian decoder are used to predict vehicle trajectories. This solves the prediction error problem of existing methods in complex traffic scenarios and achieves high-precision, real-time vehicle trajectory prediction.

CN122126306APending Publication Date: 2026-06-02INNER MONGOLIA UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
INNER MONGOLIA UNIV OF SCI & TECH
Filing Date
2026-02-06
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing vehicle trajectory prediction methods struggle to simultaneously model the temporal evolution and spatial interaction of vehicles when dealing with complex nonlinear traffic scenarios. Furthermore, they exhibit significant prediction errors and high computational complexity in high-density traffic flow and complex intersection scenarios, making it difficult to meet the real-time requirements of vehicle-to-everything (V2X) networks.

Method used

A social graph-driven approach is adopted. By acquiring historical trajectory data of the target vehicle, a social tensor graph is constructed and a weighted mask matrix is ​​generated. A multi-scale convolutional aggregation module is used to extract local and global social interaction features. Combined with a Gaussian decoder, autoregressive prediction is performed to output the probability distribution of future trajectories.

Benefits of technology

It maintains high prediction accuracy under traffic density scenarios, reduces prediction errors, improves the model's ability to represent complex traffic scenarios, quantifies prediction uncertainty, and is suitable for real-time trajectory prediction in vehicle-to-everything (V2X) networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122126306A_ABST
    Figure CN122126306A_ABST
Patent Text Reader

Abstract

This invention provides a social graph-driven trajectory prediction method for connected vehicles, belonging to the fields of connected vehicles, intelligent transportation, and autonomous driving. The method involves acquiring and preprocessing historical trajectory data of a target vehicle to extract its historical motion features, resulting in an encoder hidden state vector. Centered on the target vehicle, it filters social neighbor vehicles within the communication range, constructs a social tensor graph, and generates a weighted mask matrix. A multi-scale convolutional aggregation module extracts local and global social interaction features, which are then integrated to obtain a social representation vector. The encoder hidden state vector and the social representation vector are fused and input into a Gaussian decoder. An autoregressive mechanism is used to progressively predict the vehicle's position at multiple future time steps, obtaining the probability distribution of the target vehicle's future trajectory. Negative log-likelihood loss is used as the loss function to train the prediction model, outputting the probability distribution value of the target vehicle's predicted trajectory within each prediction step.
Need to check novelty before this filing date? Find Prior Art