Scholarlink

Detecting violent behavior in surveillance videos is a critical task for enhancing public safety and security. In this work, we propose a deep learning-based approach to automatically identify violent incidents in video footage. Our dataset consists of 2000 real-life surveillance videos, evenly split into two categories: Violence (1000 videos depicting severe violent acts) and NonViolence (1000 videos showing everyday activities such as eating, sports, and singing). We employ a hybrid model combining MobileNet for spatial feature extraction and BI-LSTM for temporal sequence analysis. Our model achieves impressive results, with a precision of 92%, recall of 92%, and an F1-score of 92%, demonstrating its effectiveness in accurately distinguishing violent from non-violent scenes. These findings focus the ability of our methods for real-world applications, like automated surveillance systems, where early detection of violence can considerably improve response times and safety measures.

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

Deep Learning-Based Violence Detection in Surveillance Videos: A Step Toward Safer Cities

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ramya Vaduguru Venkata,
  • Yashodha S. Sindhe,
  • Ashwini N. Shinde,
  • Asaram Janwale

摘要

Detecting violent behavior in surveillance videos is a critical task for enhancing public safety and security. In this work, we propose a deep learning-based approach to automatically identify violent incidents in video footage. Our dataset consists of 2000 real-life surveillance videos, evenly split into two categories: Violence (1000 videos depicting severe violent acts) and NonViolence (1000 videos showing everyday activities such as eating, sports, and singing). We employ a hybrid model combining MobileNet for spatial feature extraction and BI-LSTM for temporal sequence analysis. Our model achieves impressive results, with a precision of 92%, recall of 92%, and an F1-score of 92%, demonstrating its effectiveness in accurately distinguishing violent from non-violent scenes. These findings focus the ability of our methods for real-world applications, like automated surveillance systems, where early detection of violence can considerably improve response times and safety measures.