Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Balancing Accuracy and Training Time in Federated Learning for Violence Detection in Surveillance Videos: A Study of Neural Network Architectures

Pajon Quentin, Serre Swan, Wissocq Hugo, Rabaud Léo, Haidar Siba, Yaacoub Antoun

TL;DR

This study addresses privacy-preserving violence detection in surveillance video by leveraging federated learning and a lightweight Diff-Gated architecture that replaces optical flow with frame differences. It combines transfer learning with One-Cycle training to accelerate convergence and employs spatio-temporal features to boost accuracy while reducing computation. The authors establish a federated-learning ready dataset preparation method and demonstrate that Diff-Gated achieves higher accuracy and lower preprocessing/training times than Flow-Gated, including in a FedAvg-based FL setup. The work highlights the practical viability of privacy-conscious, efficient violence detection in real-world CCTV centers, while also acknowledging memory and non-IID data challenges and proposing directions for future FL improvements.

Abstract

This paper presents an investigation into machine learning techniques for violence detection in videos and their adaptation to a federated learning context. The study includes experiments with spatio-temporal features extracted from benchmark video datasets, comparison of different methods, and proposal of a modified version of the "Flow-Gated" architecture called "Diff-Gated." Additionally, various machine learning techniques, including super-convergence and transfer learning, are explored, and a method for adapting centralized datasets to a federated learning context is developed. The research achieves better accuracy results compared to state-of-the-art models by training the best violence detection model in a federated learning context.

Balancing Accuracy and Training Time in Federated Learning for Violence Detection in Surveillance Videos: A Study of Neural Network Architectures

TL;DR

This study addresses privacy-preserving violence detection in surveillance video by leveraging federated learning and a lightweight Diff-Gated architecture that replaces optical flow with frame differences. It combines transfer learning with One-Cycle training to accelerate convergence and employs spatio-temporal features to boost accuracy while reducing computation. The authors establish a federated-learning ready dataset preparation method and demonstrate that Diff-Gated achieves higher accuracy and lower preprocessing/training times than Flow-Gated, including in a FedAvg-based FL setup. The work highlights the practical viability of privacy-conscious, efficient violence detection in real-world CCTV centers, while also acknowledging memory and non-IID data challenges and proposing directions for future FL improvements.

Abstract

This paper presents an investigation into machine learning techniques for violence detection in videos and their adaptation to a federated learning context. The study includes experiments with spatio-temporal features extracted from benchmark video datasets, comparison of different methods, and proposal of a modified version of the "Flow-Gated" architecture called "Diff-Gated." Additionally, various machine learning techniques, including super-convergence and transfer learning, are explored, and a method for adapting centralized datasets to a federated learning context is developed. The research achieves better accuracy results compared to state-of-the-art models by training the best violence detection model in a federated learning context.
Paper Structure (20 sections, 3 figures, 7 tables)

This paper contains 20 sections, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Violence Detection
  4. Federated Learning
  5. Methodology
  6. Limitation of Classical Classifiers
  7. Transfer Learning and Early Stopping
  8. Transfer Learning and One-Cycle Training
  9. Multi-Channel Input Models using Optical Flow
  10. Multi-Channel Input Models using Frame Differences
  11. Experiment
  12. Dataset Selection
  13. Experimental Setup: Hardware and Software
  14. Data Preparation for Federated Learning: Adapting Traditional Datasets
  15. Federated Learning with Previously Tested Models
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Structure of the feature extraction architecture sernani_deep_2021
  • Figure 2: Structure of the transfer learning architecture sernani_deep_2021
  • Figure 3: Structure of the flow gated architecture cheng_rwf-2000_2020