REAL-TIME VIOLENCE DETECTION USING DEEP REINFORCEMENT LEARNING

Authors

  • Vaishali Author
  • Amalia Author

Keywords:

Big data, Apache spark framework, Histogram of Oriented Gradients (HOG), Deep Reinforcement Learning (DRL), Double Dueling Deep Q-learning network (????3???? ????????????)

Abstract

Violence is on the rise these days, and in this research, violence that takes place in public spaces is referred to as trembling violence. A real-time violence detection model is needed to monitor people's behavior and take appropriate action when abnormal activities occur in order to prevent these abnormal actions. This work introduces a real-time violence detection system that uses a human intelligence simulation to identify violence in the vast volume of streaming data. The Apache Spark method is used to analyze a vast amount of real-time video feeds from several sources that make up the big data. This approach divides the frames and uses the histogram-of-oriented gradient (HOG) algorithm to get each frame's attributes. Then, based on their characteristics, each frame is categorized as either violent or non-violent. In terms of deep reinforcement learning (DRL), this categorization is carried out using a double dueling deep Q-learning network (???????????? ????????????). When compared to the current deep neural networks, the suggested architecture achieves the best and most accurate results. For real-time violence detection, this model has a classification accuracy rate of 95.08%.

Downloads

Published

2026-06-29