Seminars

View all Seminars  |  Download ICal for this event

Enabling Secure and Scalable Data Analytics on Edge Devices

Series: Ph.D. Colloquium

Speaker: Arun Joseph, Ph.D (Engg.) student, Dept. of CSA, IISc

Date/Time: Nov 03 11:30:00

Location: CSA Auditorium, (Room No. 104, Ground Floor)

Faculty Advisor: Prof. Vinod Ganapathy

Abstract:
Closed-Circuit Television (CCTV) systems play a vital role in public safety, urban monitoring, and criminal investigation by providing continuous visual surveillance. With the increasing urbanization and affordability of cameras, cities worldwide are witnessing an exponential rise in camera density??for instance, Delhi, India, now hosts approximately 1,446 cameras per square mile, one of the highest in the world. These cameras are owned and operated by diverse stakeholders, including law enforcement, government bodies, private organizations, and individual citizens. However, despite this abundance of visual data, collaborative analysis remains a major challenge. City authorities and law enforcement agencies often require access to video footage from privately owned cameras for event reconstruction or investigation, yet private entities are reluctant to share raw footage due to privacy, trust, and data ownership concerns.

To address these challenges, we propose the Public Event Recording and Querying System (PERQS), a distributed and privacy-preserving framework for collaborative CCTV video analytics. PERQS integrates public, private, and individual cameras into a contributory network that enables secure querying of video feeds without requiring raw footage sharing. It employs local video analysis to maintain data sovereignty while ensuring authenticity through hash-based video commitments that make tampering detectable. The system introduces two novel consensus mechanisms??query consensus, which ensures collective agreement across multiple cameras analyzing the same event, and time-travel consensus, which leverages spatiotemporal correlations from nearby cameras to fill blind spots or compensate for missing feeds. To facilitate flexible querying, we design a domain-specific query language, PERQL, inspired by SQL, which allows users to express analytical queries over distributed video data. The systems plug-and-play architecture supports seamless integration with diverse video analytics algorithms and computer vision models, enabling extensibility for future workloads.

To mitigate potential data leakage risks within the blockchain layer that underpins PERQS, we develop Aramid, a privilege separation mechanism for the Hyperledger Fabric permissioned blockchain platform. Aramid isolates peer nodes and enforces strict data access controls, ensuring channel confidentiality and reducing the attack surface within distributed deployments.

Recognizing that PERQS does not support joint video analytics natively, we further evaluate two leading privacy-preserving computation paradigms??Secure Multiparty Computation (MPC) and Trusted Execution Environments (TEE)??to enable collaborative analytics across organizations. We conduct a comprehensive experimental study covering five real-world case studies, including object re-identification, scene similarity detection, vehicle counting, and machine learning??based video fusion tasks. Additionally, we benchmark core image processing operations??such as thresholding, histogram computation, Sobel edge detection, convolution, and thinning??across various MPC configurations to determine the most efficient protocol for video workloads. Our results indicate that TEE-based approaches significantly outperform MPC, achieving up to 70??90? faster execution in compute-intensive tasks, while 3-party replicated secret-sharing (ring)??based MPC offers a reasonable trade-off in settings without trusted hardware. Through detailed evaluation of performance, security, and implementation complexity, we demonstrate that TEE and MPC represent complementary solutions for extending PERQS to support privacy-preserving, scalable, and trustworthy joint video analytics in multi-organizational environments.