Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ReeFRAME: Reeb Graph based Trajectory Analysis Framework to Capture Top-Down and Bottom-Up Patterns of Life

Chandrakanth Gudavalli, Bowen Zhang, Connor Levenson, Kin Gwn Lore, B. S. Manjunath

TL;DR

ReeFRAME is a scalable Reeb graph-based framework designed to analyze vast volumes of GPS-enabled human trajectory data generated at 1Hz frequency, and its linear algorithmic complexity relative to the number of time points ensures scalability for anomaly detection.

Abstract

In this paper, we present ReeFRAME, a scalable Reeb graph-based framework designed to analyze vast volumes of GPS-enabled human trajectory data generated at 1Hz frequency. ReeFRAME models Patterns-of-life (PoL) at both the population and individual levels, utilizing Multi-Agent Reeb Graphs (MARGs) for population-level patterns and Temporal Reeb Graphs (TERGs) for individual trajectories. The framework's linear algorithmic complexity relative to the number of time points ensures scalability for anomaly detection. We validate ReeFRAME on six large-scale anomaly detection datasets, simulating real-time patterns with up to 500,000 agents over two months.

ReeFRAME: Reeb Graph based Trajectory Analysis Framework to Capture Top-Down and Bottom-Up Patterns of Life

TL;DR

ReeFRAME is a scalable Reeb graph-based framework designed to analyze vast volumes of GPS-enabled human trajectory data generated at 1Hz frequency, and its linear algorithmic complexity relative to the number of time points ensures scalability for anomaly detection.

Abstract

In this paper, we present ReeFRAME, a scalable Reeb graph-based framework designed to analyze vast volumes of GPS-enabled human trajectory data generated at 1Hz frequency. ReeFRAME models Patterns-of-life (PoL) at both the population and individual levels, utilizing Multi-Agent Reeb Graphs (MARGs) for population-level patterns and Temporal Reeb Graphs (TERGs) for individual trajectories. The framework's linear algorithmic complexity relative to the number of time points ensures scalability for anomaly detection. We validate ReeFRAME on six large-scale anomaly detection datasets, simulating real-time patterns with up to 500,000 agents over two months.

Paper Structure

This paper contains 15 sections, 3 equations, 5 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Temporal Reeb Graph Construction (TERG)
  5. Making Temporal Reebs Describable
  6. Multi-Agent Reeb Graphs (MARGs)
  7. Fusing TERGs with MARGs
  8. Test Trajectory Analysis
  9. Combining Scores
  10. Experiments
  11. Datasets
  12. Feature Extraction, Hyperparameter Tuning, and Results
  13. Scalable Inference on Cloud
  14. Future Work and Discussion
  15. Acknowledgement

Figures (5)

  • Figure 1: Set of seven one-day trajectories of an agent, contributing to appear/disappear/connect/disconnect events. Events will be later used to construct Reeb graph shailja2021computational for the agent.
  • Figure 2: Architectural overview of the proposed anomaly detection framework, ReeFRAME. The pipeline splits into two key modeling processes: Agent Level Modeling, where Temporal Reeb Graphs (TERGs) are generated for individual agents, and Population Level Modeling, where a Multi-Agent Reeb Graph (MARG) is constructed to model broader population patterns. The TERG and MARG features are combined in the Feature Ensembler, which integrates and analyzes the data to produce Detections, identifying agents that deviate from normal behavior.
  • Figure 3: Demonstration of Incremental Reeb graph Construction using a toy example. (a) Sample set of three sub-trajectories; (b) Reeb graph constructed from the sub-trajectories; (c) A new sub-trajectory that will be introduced into the Reeb graph; (d) Updated Reeb graph accommodating the new sub-trajectory.
  • Figure 4: Anomaly Detection Performance of ReeFRAME.
  • Figure 5: AWS Architecture Diagram, demonstrating the cloud readiness of ReeFRAME and the strategies that we used to run the pipeline on cloud.