Edge AI for Autonomous Urban Safety Systems

Urban safety infrastructure is undergoing a significant shift as cities become denser, moredynamic, and increasingly dependent on real-time decision-making. Traditional cloud-basedsurveillance systems, while effective at scale, often struggle with latency and reliability intime-critical scenarios, where even small delays can lead to serious consequences. In thiscontext, Edge AI the deployment of machine learning models directly on local devices hasemerged as a promising alternative. This paper presents a comprehensive review of Edge AI systems for real-time threat detectionin urban environments. It examines how modern deep learning models, includingconvolutional neural networks and transformer-based architectures, are being adaptedthrough model compression techniques such as quantization, pruning, and knowledgedistillation to operate efficiently on resource-constrained hardware platforms like FPGAs,embedded GPUs, and mobile SoCs. Key application areas are explored, including anomalydetection in public spaces, traffic incident response, and acoustic threat detection.However, despite strong progress in controlled settings, real-world deployment remainschallenging. Issues related to hardware limitations, inconsistent model performance undervarying environmental conditions, data privacy concerns, and evolving regulatoryframeworks continue to restrict large-scale adoption. Unlike many existing surveys that focus primarily on model performance, this reviewemphasizes the gap between laboratory results and real-world operational reliability. It furtherhighlights emerging directions such as federated learning, neuromorphic computing, anddeeper integration with smart city ecosystems as critical to the future of urban safety systems.