Real-Time Smart City Surveillance Using Big Data and Deep Learning-Based Video Analytics

Abstract

With the explosion of smart city infrastructure, vast amounts of surveillance video data are being created, that need scalable, intelligent, and low latency analytics. In this paper, we propose a framework of distributed big data processing and state-of-the-art deep learning models in real time surveillance, named BD-DLVS (Big Data Deep Learning Video Surveillance Framework). It consists of 6 hierarchical levels and includes the Real-Time Intelligent Video Analytics Algorithm (RT-IVAA), which includes video ingestion, anomaly classification and automated alert generation. On the UCF-Crime, VIRAT and ShanghaiTech datasets, experimental evaluation yields a detection accuracy of 96.7%, precision of 95.4%, recall of 94.9% and an average latency of 47ms, outperforming existing approaches in terms of detection accuracy and latency for large-scale deployment in smart cities.