Anomaly Detection Using Trajectory Meta-Features

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current anomaly detection methods for connected vehicles face challenges in accurately identifying new intrusions and zero-day attacks, particularly due to high false positive rates and the need for labeled data, which limits their effectiveness in real-world applications with rare anomalous events and privacy concerns.

Innovation Solution

A computer-implemented method using meta-features such as distance, time, and velocity to represent vehicle trajectories in a meta-feature space, applying a clustering algorithm like k-means to detect anomalous trajectories, allowing for early detection of cyber-attacks without requiring labeled data and reducing the need for extensive human expertise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If supervised anomaly detection methods are used, then detection accuracy for known attacks is improved, but the requirement for labeled data and human expertise increases

Engineering Contradiction:
Improvedetection accuracyVSAvoiddata labeling complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs self-training by automatically generating labels from unlabeled trajectory data through clustering algorithms. The anomaly detection model trains itself without requiring external labeled data or human annotation, enabling supervised learning capabilities while eliminating the complexity of manual data labeling.

Inventive Principle:
Principle #25Self-service

2Reliability

If traditional anomaly detection methods are used, then detection capability is improved, but false positive rate increases

Engineering Contradiction:
Improvedetection capabilityVSAvoidfalse positive rate
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The system transforms trajectory data from spatial coordinates into temporal meta-features (distance, time, velocity) to create a new feature space. This dimensional transformation enables the clustering algorithm to better distinguish between normal and anomalous patterns, improving detection capability while reducing false positives through more meaningful feature representation.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Measurement precision

If comprehensive trajectory analysis is performed, then detection precision is improved, but computational time increases

Engineering Contradiction:
Improvedetection precisionVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system extracts only the essential meta-features (distance, time, velocity) from complete trajectory data that are most relevant for anomaly detection. By selecting and analyzing only these critical features rather than processing all trajectory information, the system maintains high detection precision while significantly reducing computational time and resources required.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4141715A1Anomaly detection
Publication Date: 2023.03.01 FUJITSU LTD
  • EP4141715A1 patent drawingFigure 1
  • EP4141715A1 patent drawingFigure 2
  • EP4141715A1 patent drawingFigure 3

AI summary

A computer-implemented method comprising: providing data of at least one of distance, time, and velocity as at least one trajectory feature, the data defining a plurality of vehicle trajectories; computing for each vehicle trajectory a plurality of meta-features by computing, based on the data, at least one of a maximum value, a mean value, and a standard deviation of at least one of the at least one trajectory feature of each vehicle trajectory, so that each vehicle trajectory is represented by a point in a meta-feature space, the point defined by the plurality of meta-features; implementing a clustering algorithm on the points and selecting, based on a result of the clustering algorithm, at least one point as an anomaly; and selecting the at least one vehicle trajectory corresponding to the at least one point as at least one anomalous vehicle trajectory.