Developing AI-Based Surveillance Software

Abstract

In this work, we developed an AI-based surveillance system to be used by businesses to improve public safety, security, and law enforcement efforts. The system has numerous potential applications, including lost item retrieval, efficient emergency response, facial biometric ver ification, and more. Our software utilizes advanced algorithms and state-of-the-art technologies to analyze videos from CCTV cameras and accurately identify individuals and objects in real-time. Despite recent advances in face recognition, small face recognition at scale remains a significant challenge. To address this, we utilized Insightface models for face recognition and achieved an accuracy of 75% with the system processing 15 frames per second. To optimize the system, we implemented several infrastructure changes, including shifting from Elasticsearch to Opensearch, which reduced the loading time of our UI interface from 2 minutes to 10-15 seconds. The entire system is highly scalable and fault-tolerant, capable of processing 60 million images per day due to its implementation in Kubernetes.