Real-Time Suspicious Activity Detection in Bank and ATM Environments Using Meta-Learning and Edge Computing

This paper presents a cutting-edge surveillance system designed to enhance real-time detection of suspicious activities in sensitive environments such as banks and ATMs. By leveraging edge computing technologies and meta-learning models, the system achieves high accuracy in analyzing live video streams from CCTV cameras over RTSP protocols. A robust architecture integrates tracking-assisted object detection and adaptive behavioral classification, ensuring low latency and reliable performance even under resource-constrained conditions. Edge devices such as NVIDIA Jetson process video frames locally, reducing bandwidth consumption while maintaining data privacy. Unlike traditional approaches, the proposed system utilizes meta-learning frameworks to adapt rapidly to novel and evolving threats with minimal retraining, providing a scalable solution for dynamic security challenges. Through extensive experiments, including latency analysis and classification accuracy benchmarks, the system demonstrates significant advantages over cloud-based methods, achieving faster response times and greater operational efficiency. This work lays the foundation for deploying intelligent, adaptive surveillance systems in high-security environments.