Video Analytics Privacy via Data-Oblivious TEEs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The proliferation of IoT devices and video cameras has led to challenges in securing video analytics in the cloud, as existing technologies are vulnerable to attacks that compromise confidentiality and integrity, particularly due to side-channel attacks on memory access patterns and data-dependent processing.

Innovation Solution

A system utilizing trusted execution environments (TEEs) across central processing units (CPUs) and graphics processing units (GPUs) with data-oblivious algorithms to obscure processing operations and communication, preventing attackers from inferring data-dependent access patterns and protecting against side-channel attacks, while ensuring confidentiality and integrity of video analytics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If video analytics processing is performed in the cloud using conventional methods, then processing capability and productivity are improved, but security and confidentiality are compromised due to side-channel attacks

Engineering Contradiction:
Improvevideo analytics processing capabilityVSAvoidsecurity and confidentiality
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent introduces trusted execution environments (TEEs) as intermediary secure processing units between the cloud infrastructure and video analytics workloads. These TEEs act as mediators that isolate and protect sensitive video data and processing operations from potential attacks on the underlying cloud platform, thereby maintaining both high processing capability and security confidentiality simultaneously

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent implements nested trusted execution environments where secure processing units are embedded within the cloud infrastructure hierarchy. Multiple layers of TEEs are nested within each other, creating defense-in-depth security architecture that protects video analytics processing while maintaining cloud-based productivity benefits

Inventive Principle:
Principle #7Nested doll (Nesting)

2Reliability

If data-oblivious algorithms are used to obscure processing operations, then security against side-channel attacks is improved, but computational complexity and processing time increase

Engineering Contradiction:
Improvesecurity against side-channel attacksVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies data-oblivious algorithms selectively to specific critical processing stages within the TEEs rather than uniformly across all operations. By identifying and protecting only the most vulnerable local processing points (such as memory access patterns and intermediate data transmissions), the system achieves adequate security protection while minimizing overall computational complexity increases

Inventive Principle:
Principle #3Local quality

3Reliability

If multiple trusted execution environments are used across CPU and GPU, then security and confidentiality are improved, but system complexity and communication overhead increase

Engineering Contradiction:
Improveconfidentiality and integrityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the security management and coordination functions of multiple TEEs (across CPU and GPU) into a unified TEE management layer. This consolidation reduces the complexity of managing multiple independent secure environments by providing centralized control over security policies, key management, and inter-TEE communication protocols

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3999987B1Private data analytics
Publication Date: 2024.08.07 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3999987B1 patent drawingFigure 1
  • EP3999987B1 patent drawingFigure 2
  • EP3999987B1 patent drawingFigure 3

AI summary

This document relates to performing video analytics on a cloud device that preserves privacy. One example uses data-oblivious algorithms to process input video data, where the data-oblivious algorithms can assist with preventing attackers from exploiting side-channels induced by data-dependent access patterns.