Scholarlink

<p>In recent years, the need for efficient violence detection methods has become increasingly critical for various security applications. Existing approaches leveraging deep learning (DL), particularly Vision Transformers (ViTs), often suffer from high complexity and overhead, limiting their real-time applicability. This paper proposes a lightweight spatiotemporal transformer-based violence detection method that integrates ViT with a Convolutional Block Attention Module (CBAM) and Long Short-Term Memory (LSTM) for spatial and temporal feature extraction, respectively. By replacing the multi-head self-attention (MSA) mechanism in ViT with CBAM, we significantly reduce the number of parameters and computational costs while enhancing the model’s ability to focus on localized patterns essential for violence detection. Our approach is validated on four benchmark datasets, achieving accuracies of 99.81%, 99%, 99.50%, and 99.10% on AIRTLab, Industrial Surveillance, RWF2000, and Hockey Fights, respectively. The proposed method demonstrates a good balance between model complexity and computational efficiency, making it suitable for real-time violence detection applications.</p>

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

Lightweight Vision Transformer with CBAM and LSTM for Efficient Violence Detection in Image Sequences

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Aishvarya Garg,
  • Swati Nigam,
  • Rajiv Singh

摘要

In recent years, the need for efficient violence detection methods has become increasingly critical for various security applications. Existing approaches leveraging deep learning (DL), particularly Vision Transformers (ViTs), often suffer from high complexity and overhead, limiting their real-time applicability. This paper proposes a lightweight spatiotemporal transformer-based violence detection method that integrates ViT with a Convolutional Block Attention Module (CBAM) and Long Short-Term Memory (LSTM) for spatial and temporal feature extraction, respectively. By replacing the multi-head self-attention (MSA) mechanism in ViT with CBAM, we significantly reduce the number of parameters and computational costs while enhancing the model’s ability to focus on localized patterns essential for violence detection. Our approach is validated on four benchmark datasets, achieving accuracies of 99.81%, 99%, 99.50%, and 99.10% on AIRTLab, Industrial Surveillance, RWF2000, and Hockey Fights, respectively. The proposed method demonstrates a good balance between model complexity and computational efficiency, making it suitable for real-time violence detection applications.