A trajectory anomaly detection system, an online trajectory anomaly detection method

The trajectory anomaly detection system based on the causal generative model decomposes into trajectory likelihood estimation and bias-reduced scaling estimation. It utilizes a variational autoencoder (VAE) to model the causal relationship between the start-end point pairs and the trajectory, thus solving the problem of decreased performance of existing methods on unseen start-end point pairs and achieving efficient online detection.

CN117423228BActive Publication Date: 2026-06-05INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INST OF COMPUTING TECH CHINESE ACAD OF SCI
Filing Date
2023-09-22
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing trajectory anomaly detection methods perform well on start-end point pairs seen in the training set, but their effectiveness drops on unseen start-end point pairs. Furthermore, they are computationally inefficient and difficult to apply to large-scale datasets and online detection tasks.

Method used

A trajectory anomaly detection system based on a causal generation model is adopted, which is decomposed into trajectory likelihood estimation and debiasing scaling estimation. The variational autoencoder (VAE) is used to model the causal relationship between the start and end point pairs and the trajectory and the road segment preference, respectively. The anomaly score is calculated through trajectory likelihood estimation and debiasing scaling estimation.

Benefits of technology

It improves the model's out-of-distribution generalization ability, can efficiently handle unseen start-end point pairs, supports online trajectory anomaly detection tasks, and meets the detection requirements with low time complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117423228B_ABST
    Figure CN117423228B_ABST
Patent Text Reader

Abstract

The application provides a trajectory anomaly detection system for detecting whether an occurred trajectory between a start point and an end point in a road network is abnormal, the occurred trajectory being a trajectory matched with road network information, and the system comprises: a trajectory likelihood estimation module configured to generate a model for obtaining a first score by performing likelihood estimation on the occurred trajectory by using variational inference; a debiasing scaling estimation module configured to generate a model for obtaining a second score by performing debiasing estimation on the occurred trajectory by using variational inference; and an anomaly determination module for calculating a trajectory anomaly score according to the first score and the second score of the occurred trajectory according to a preset rule. The application decomposes anomaly detection into trajectory likelihood estimation and debiasing scaling estimation, fully models the causal relationship between the start point and the end point and the influence of road section preference, completes debiasing with very low time overhead, can efficiently process the occurring trajectory, and supports online detection tasks.
Need to check novelty before this filing date? Find Prior Art