MULTI-ZONAL VEHICLE SURVEILLANCE SYSTEM ENABLED BY A PRIVATE PERMISSIONED BLOCKCHAIN
AuthorOTTAKATH, NAJMATH SHARFIYA
MetadataShow full item record
Privacy, security, accessibility, and reliability are the most essential characteristics of a public security system. Existing surveillance systems provide monitoring and surveillance-based security. However, their inference depends on manual monitoring and action, which may result in a delay in response. Using computer vision techniques, automated surveillance with monitoring through anomaly detection and tracking has been made possible. Nevertheless, they require a centralized storage system, which may result in a delay or security breach, provide a single point of failure, and render the system unavailable and unreliable. Several states of the art have proposed the use of blockchain, a decentralized ledger utilizing a private, permission-based network, to improve the framework's dependability. However, an analysis of its viability in relation to the security standards of confidentiality, integrity, and dependability, as well as its use in an enterprise-grade application such as public security, which may necessitate a highly scalable network, is required. In this regard, the proposed framework is a multi-surveillance system for vehicles that performs a privacy-protected image analysis to re-identify vehicles through images captured by various cameras and analyzed using computer vision techniques, which will be accessible to other nodes and/or surveillance zones via the blockchain ledger. A private permissioned blockchain network, Hyperledger Fabric, is evaluated for improved reliability and reduced latency using fast and lightweight image analysis tasks, such as combinations of feature extractors and lightweight CNN (Convolutional Neural Network) models. With different approaches in three domains, the accuracy and time required by an edge-based inference tool are measured, resulting in a lightweight tool for surveillance via a permissioned private blockchain network.
- Computer Science & Engineering [87 items ]