Scholarlink

Prevention and early detection are more effective than addressing issues after severe symptoms have manifested. Security and safety are essential pillars that provide stability and growth to a community. The installation of CCTV cameras in public places serves to enhance these pillars by offering continuous surveillance and monitoring. This study proposes a sophisticated system for the early detection of crimes using real-time analysis of CCTV footage through advanced machine learning algorithms. The system utilizes an Inflated 3D ConvNet (I3D) model from Google DeepMind, specifically trained to detect kinetic motion, to analyze footage in real time. To train the model, a substantial dataset comprising 35,000 classified clips of violent and non-violent behavior is employed. By leveraging advanced techniques such as 3D Convolutional Neural Networks (3DCNN) and Improved Temporal Segment Networks (ITSM), the system can accurately detect instances of violence in the footage. Once detected, the system promptly alerts the user, providing real-time notifications of potential threats, thereby significantly enhancing public safety and security.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Crime Detection in CCTV Footage Using Machine Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • John Adithya,
  • Payam Mortezaei,
  • Khaled Aljawawde,
  • Mohamed Mohamed,
  • Rita Zgheib,
  • Tamer Mohamed

摘要

Prevention and early detection are more effective than addressing issues after severe symptoms have manifested. Security and safety are essential pillars that provide stability and growth to a community. The installation of CCTV cameras in public places serves to enhance these pillars by offering continuous surveillance and monitoring. This study proposes a sophisticated system for the early detection of crimes using real-time analysis of CCTV footage through advanced machine learning algorithms. The system utilizes an Inflated 3D ConvNet (I3D) model from Google DeepMind, specifically trained to detect kinetic motion, to analyze footage in real time. To train the model, a substantial dataset comprising 35,000 classified clips of violent and non-violent behavior is employed. By leveraging advanced techniques such as 3D Convolutional Neural Networks (3DCNN) and Improved Temporal Segment Networks (ITSM), the system can accurately detect instances of violence in the footage. Once detected, the system promptly alerts the user, providing real-time notifications of potential threats, thereby significantly enhancing public safety and security.