Scholarlink

This paper addresses the critical task of identifying suspicious activities, or anomaly detection, within human behaviour analysis to enhance public safety. Traditional CCTV systems relying on human monitoring are insufficient given the increasing threats from violence and accidents. We propose an automated security system using a CNN-LSTM hybrid model (Long- term Recurrent Convolutional Network) to analyse CCTV footage for real-time detection of anomalies like fighting, robbery, and firing, achieving an accuracy of 91.5%. Implemented with Keras and TensorFlow, the system offers AI-driven alarms, email alerts, and simplified PDF reports, ensuring prompt responses to potential threats. The novelty of this approach lies in combining CNNs for spatial feature extraction and LRCNs for temporal dependencies, addressing limitations of traditional and standalone neural network models. This integrated architecture is specially tailored for anomaly detection in sequential video data.

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

Suspicious Human Activity Detection Using CNN and LRCN

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mitali Dinesh Mahajan,
  • Prathamesh Sanjay Patil,
  • Sanika Sandeep Kurale,
  • Aditi Jagannath Patil

摘要

This paper addresses the critical task of identifying suspicious activities, or anomaly detection, within human behaviour analysis to enhance public safety. Traditional CCTV systems relying on human monitoring are insufficient given the increasing threats from violence and accidents. We propose an automated security system using a CNN-LSTM hybrid model (Long- term Recurrent Convolutional Network) to analyse CCTV footage for real-time detection of anomalies like fighting, robbery, and firing, achieving an accuracy of 91.5%. Implemented with Keras and TensorFlow, the system offers AI-driven alarms, email alerts, and simplified PDF reports, ensuring prompt responses to potential threats. The novelty of this approach lies in combining CNNs for spatial feature extraction and LRCNs for temporal dependencies, addressing limitations of traditional and standalone neural network models. This integrated architecture is specially tailored for anomaly detection in sequential video data.