A Hybrid Deep Learning Model for Efficient Anomaly Detection in Video Surveillance Systems

• Abba M. BALA

  Department of Computer Science, Faculty of Computing, Northwest University, Kano, Kano State, Nigeria

  Author

• Abdurrauf G. SHARIFAI

  Department of Computer Science, Faculty of Computing, Northwest University, Kano, Kano State, Nigeria

  Author

• Umar S. HARUNA

  Department of Cybersecurity, Faculty of Computing, Northwest University, Kano, Kano State, Nigeria

  Author

Traditional CNNs often face challenges in capturing long-range dependencies and contextual relationships within data, which limits their effectiveness in complex tasks like anomaly detection. To overcome these limitations, we propose an innovative enhanced attention-mechanism hybrid model that combines the strengths of CNNs with Transformer architecture. This hybrid model leverages the powerful feature extraction capabilities of four distinct CNN architectures, VGG16, DenseNet121, ResNet50, and MobileNetV2. To process the training data comprehensively, the extracted features are fused and passed through Swin Transformer which integrates attention mechanisms to capture long-range dependencies within the data effectively, and focuses on the most relevant regions of the input data. The effectiveness of this approach is evaluated on the UCF-Crime benchmark dataset using performance metrics such as ROC-AUC, achieving an outstanding accuracy of 99.2% surpassing existing state-of-the-art methods. Moreover, the model’s ability to handle complex video data and extract semantically rich features highlights its potential for real-time surveillance applications where timely and accurate anomaly detection is critical.

• FJET_21_4_4

Copyright (c) 2026 FUDMA Journal of Engineering and Technology

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.