Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VelocityNet: Real-Time Crowd Anomaly Detection via Person-Specific Velocity Analysis

Fatima AlGhamdi, Omar Alharbi, Abdullah Aldwyish, Raied Aljadaany, Muhammad Kamran J Khan, Huda Alamri

TL;DR

VelocityNet tackles real-time anomaly detection in extremely dense crowds by fusing head-centric detection with dense optical flow to estimate per-person velocities. It then learns normal motion patterns via unsupervised clustering and maps them to four semantic categories (halt, slow, normal, fast) using hierarchical clustering, followed by density-aware, percentile-based anomaly scoring for interpretability and robustness. The approach emphasizes depth-invariant velocity descriptors, per-density regime modeling, and efficient computation to enable real-time deployment in challenging environments. Experimental analysis on a real-world, densely crowded dataset demonstrates effective anomaly discrimination and practical runtime performance, highlighting the framework's potential for practical crowd monitoring and safety applications.

Abstract

Detecting anomalies in crowded scenes is challenging due to severe inter-person occlusions and highly dynamic, context-dependent motion patterns. Existing approaches often struggle to adapt to varying crowd densities and lack interpretable anomaly indicators. To address these limitations, we introduce VelocityNet, a dual-pipeline framework that combines head detection and dense optical flow to extract person-specific velocities. Hierarchical clustering categorizes these velocities into semantic motion classes (halt, slow, normal, and fast), and a percentile-based anomaly scoring system measures deviations from learned normal patterns. Experiments demonstrate the effectiveness of our framework in real-time detection of diverse anomalous motion patterns within densely crowded environments.

VelocityNet: Real-Time Crowd Anomaly Detection via Person-Specific Velocity Analysis

TL;DR

VelocityNet tackles real-time anomaly detection in extremely dense crowds by fusing head-centric detection with dense optical flow to estimate per-person velocities. It then learns normal motion patterns via unsupervised clustering and maps them to four semantic categories (halt, slow, normal, fast) using hierarchical clustering, followed by density-aware, percentile-based anomaly scoring for interpretability and robustness. The approach emphasizes depth-invariant velocity descriptors, per-density regime modeling, and efficient computation to enable real-time deployment in challenging environments. Experimental analysis on a real-world, densely crowded dataset demonstrates effective anomaly discrimination and practical runtime performance, highlighting the framework's potential for practical crowd monitoring and safety applications.

Abstract

Detecting anomalies in crowded scenes is challenging due to severe inter-person occlusions and highly dynamic, context-dependent motion patterns. Existing approaches often struggle to adapt to varying crowd densities and lack interpretable anomaly indicators. To address these limitations, we introduce VelocityNet, a dual-pipeline framework that combines head detection and dense optical flow to extract person-specific velocities. Hierarchical clustering categorizes these velocities into semantic motion classes (halt, slow, normal, and fast), and a percentile-based anomaly scoring system measures deviations from learned normal patterns. Experiments demonstrate the effectiveness of our framework in real-time detection of diverse anomalous motion patterns within densely crowded environments.
Paper Structure (21 sections, 14 equations, 2 figures, 4 tables)

This paper contains 21 sections, 14 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Anomaly Detection in Videos
  4. Velocity Estimation in Crowded Scenes
  5. Motion Representation and Analysis
  6. Methodology
  7. Overview
  8. Motion Estimation Module
  9. Head Detection Module
  10. Velocity Estimation Module
  11. Anomaly Detection Module
  12. Unsupervised Motion Clustering
  13. Semantic Grouping via Hierarchical Clustering
  14. Density-Aware Modeling
  15. Anomaly Scoring
  16. ...and 6 more sections

Figures (2)

  • Figure 1: Architecture overview of VelocityNet: Two parallel streams—head detection and dense optical flow—process the input simultaneously. These outputs merge to produce per-person velocity descriptors, which are clustered during training to establish normal behavior boundaries. During inference, computed anomaly scores are compared against these predefined normal boundaries for real-time anomaly detection.
  • Figure 2: Comparison of linear (yellow) and quadratic (purple) regression models predicting motion magnitude from bounding box area (black dots).