Next-Gen Security Monitoring: Advanced Machine Learning for Intelligent Object Detection and Assessment in Surveillance

Abstract

In response to the synergistic evolution of two rapidly advancing technologies, artificial intelligence and edge computing, we present a tailored system named secure Watch: Advanced Surveillance with Intelligent Object Detection and Evaluation. This system employs a scalable edge computing architecture and leverages multitask deep learning to address pertinent computer vision tasks. Recognizing the diverse potential applications of various surveillance devices, we integrate a smart IoT module for normalizing video data from different cameras. This ensures that the secure Watch system adeptly identifies suitable data for specific tasks. Furthermore, deep learning models are deployed at each secure Watch node to conduct computer vision tasks on the normalized data. To bridge the usual gap between training and deployment of deep learning models, especially for related tasks in the same scenario, we propose a collaborative multitask training paradigm on a cloud server. Simulation results based on publicly available datasets demonstrate continuous support for intelligent monitoring tasks, robust scalability, and enhanced performance achieved through multitask learning.