A multi-agent trajectory anomaly detection system, method and application

By combining spatiotemporal feature extraction and trajectory generation modules, along with Mahalanobis distance and uncertainty penalty terms, the problem of insufficient detection accuracy and real-time performance in existing technologies is solved, achieving efficient and accurate trajectory anomaly detection in autonomous driving systems.

CN122412883APending Publication Date: 2026-07-17ANHUI UNIV OF SCI & TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202610545617.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-23
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing trajectory anomaly detection technologies in autonomous driving systems suffer from insufficient detection accuracy, inadequate real-time performance, and insufficient ability to distinguish between longitudinal and lateral errors, making it difficult to meet the requirements of millisecond-level real-time response and safety.

Method used

A spatiotemporal feature extraction module is used to extract vehicle trajectory features through a gated recurrent unit network and a multi-head graph attention network. Combined with a trajectory generation module, the desired trajectory is generated through linear interpolation path and uncertainty is quantified. The anomaly score is calculated using Mahalanobis distance and uncertainty penalty term to achieve directional error differentiation and robustness improvement.

Benefits of technology

It achieves millisecond-level real-time inference, improves detection accuracy and anomaly recall rate in complex traffic scenarios, significantly enhances robustness to visual tracking noise, captures social game behavior between vehicles, and reduces false alarm rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122412883A_ABST
    Figure CN122412883A_ABST
Patent Text Reader

Abstract

本发明提供一种实时多智能体轨迹异常检测系统、方法及应用，系统接收目标车辆及邻居车辆的历史轨迹数据(3秒30帧)，时空特征提取模块通过线性投影、门控循环单元时序编码和多头图注意力空间交互计算，提取并融合时空特征，输出条件上下文向量c。轨迹生成模块在推理阶段以条件上下文向量c为条件，从初始噪声通过欧拉积分生成期望正常轨迹，并通过蒙特卡洛采样计算均值轨迹和不确定性。异常评分模块接收真实轨迹、期望轨迹和不确定性，通过马氏距离和不确定性惩罚计算异常分数A并输出。本发明系统单样本推理时间低，系统吞吐量大，满足自驾毫秒级实时控制需求。
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • A fishing vessel trajectory anomaly detection method based on deep learning

    CN117034440B

  • Self-adaptive fragment division and data evolution automatic driving vehicle track generation method and system based on diffusion model

    CN118886303A

  • Automatic driving lane changing trajectory planning method based on deep learning

    CN120963695A