By Evany Anne Moses,  M. Brindha,  N. Sivaukumaran

In recent years, the demand for intelligent surveillance systems has grown significantly due to the increasing need for enhanced security in public and private spaces. This journal presents a comprehensive approach that integrates advanced deep learning techniques for anomaly detection, violence recognition, and person re-identification in surveillance videos. The proposed system leverages a hybrid model combining ResNet50 and 3D Convolutional Neural Networks (CNNs) to accurately detect violent activities in real time. Additionally, the system includes anomaly detection to identify unusual patterns in the video feed that may signal potential security threats. The system further incorporates YOLO for high-precision object detection, DeepSort for robust tracking, and OpenPose for pose estimation, enabling real-time monitoring and accurate identification of individuals across multiple frames. Experimental results demonstrate that the integrated system outperforms traditional methods in terms of accuracy, efficiency, and scalability, making it a powerful tool for modern surveillance applications. This work highlights the potential of combining multiple deep-learning approaches to create a more effective and reliable surveillance system capable of addressing diverse security challenges.

Unpublished paper, 2024. 34p.