Scholarlink

With the increasing demand for intelligent systems capable of autonomously detecting anomalies in streaming videos for enhanced public security, the field of video surveillance has seen significant developments. To further enhance anomaly detection capabilities, a novel temporal context integration is introduced into the proposed framework. The incorporation of (RNNs) or attention mechanisms provides a more dynamic understanding of anomalous events and improves the framework’s ability to discern complex temporal patterns. While existing method’s widespread deployment faces challenges related to long training times and the lack of interpretability in decision-making processes. To address these issues, an effective framework is proposed, to extract context features, leveraging pre-trained models. Notably, the framework introduces a novel approach to interpretability by combining Shapley additive explanations (SHAP) and autoencoder. This research contributes not only towards the advancement of anomaly detection in video but also offers a practical solution for fast deployment and interpretable decision-making in real-world circumstances.

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 Temporal Fusion Networks for Dynamic Anomaly Detection in Surveillance Video

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Debstuti Biswas,
  • Srujana Maddula,
  • Atul Pandey,
  • Niharika Dinesh,
  • Jatin Chopra,
  • Shitiz Upreti

摘要

With the increasing demand for intelligent systems capable of autonomously detecting anomalies in streaming videos for enhanced public security, the field of video surveillance has seen significant developments. To further enhance anomaly detection capabilities, a novel temporal context integration is introduced into the proposed framework. The incorporation of (RNNs) or attention mechanisms provides a more dynamic understanding of anomalous events and improves the framework’s ability to discern complex temporal patterns. While existing method’s widespread deployment faces challenges related to long training times and the lack of interpretability in decision-making processes. To address these issues, an effective framework is proposed, to extract context features, leveraging pre-trained models. Notably, the framework introduces a novel approach to interpretability by combining Shapley additive explanations (SHAP) and autoencoder. This research contributes not only towards the advancement of anomaly detection in video but also offers a practical solution for fast deployment and interpretable decision-making in real-world circumstances.