Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Real-Time Privacy-Preserving Behavior Recognition System via Edge-Cloud Collaboration

Huan Song, Shuyu Tian, Junyi Hao, Cheng Yuan, Zhenyu Jia, Jiawei Shao, Xuelong Li

TL;DR

The paper tackles the privacy-utility conflict in high-sensitivity surveillance by decoupling semantic risk perception from identity information. It introduces an edge–cloud privacy-preserving perception system based on the AI Flow framework, using SPA-D at the edge to produce irreversible feature embeddings through non-linear mapping and gradient-guided noise. Cloud inference operates solely on these abstract vectors with multimodal, cross-modal models to detect abnormal risk behaviors without accessing raw imagery. The approach enables real-time risk detection in sensitive spaces while ensuring data irreversibility, moving from traditional video surveillance to de-identified behavior perception with strong privacy protections.

Abstract

As intelligent sensing expands into high-privacy environments such as restrooms and changing rooms, the field faces a critical privacy-security paradox. Traditional RGB surveillance raises significant concerns regarding visual recording and storage, while existing privacy-preserving methods-ranging from physical desensitization to traditional cryptographic or obfuscation techniques-often compromise semantic understanding capabilities or fail to guarantee mathematical irreversibility against reconstruction attacks. To address these challenges, this study presents a novel privacy-preserving perception technology based on the AI Flow theoretical framework and an edge-cloud collaborative architecture. The proposed methodology integrates source desensitization with irreversible feature mapping. Leveraging Information Bottleneck theory, the edge device performs millisecond-level processing to transform raw imagery into abstract feature vectors via non-linear mapping and stochastic noise injection. This process constructs a unidirectional information flow that strips identity-sensitive attributes, rendering the reconstruction of original images impossible. Subsequently, the cloud platform utilizes multimodal family models to perform joint inference solely on these abstract vectors to detect abnormal behaviors. This approach fundamentally severs the path to privacy leakage at the architectural level, achieving a breakthrough from video surveillance to de-identified behavior perception and offering a robust solution for risk management in high-sensitivity public spaces.

A Real-Time Privacy-Preserving Behavior Recognition System via Edge-Cloud Collaboration

TL;DR

The paper tackles the privacy-utility conflict in high-sensitivity surveillance by decoupling semantic risk perception from identity information. It introduces an edge–cloud privacy-preserving perception system based on the AI Flow framework, using SPA-D at the edge to produce irreversible feature embeddings through non-linear mapping and gradient-guided noise. Cloud inference operates solely on these abstract vectors with multimodal, cross-modal models to detect abnormal risk behaviors without accessing raw imagery. The approach enables real-time risk detection in sensitive spaces while ensuring data irreversibility, moving from traditional video surveillance to de-identified behavior perception with strong privacy protections.

Abstract

As intelligent sensing expands into high-privacy environments such as restrooms and changing rooms, the field faces a critical privacy-security paradox. Traditional RGB surveillance raises significant concerns regarding visual recording and storage, while existing privacy-preserving methods-ranging from physical desensitization to traditional cryptographic or obfuscation techniques-often compromise semantic understanding capabilities or fail to guarantee mathematical irreversibility against reconstruction attacks. To address these challenges, this study presents a novel privacy-preserving perception technology based on the AI Flow theoretical framework and an edge-cloud collaborative architecture. The proposed methodology integrates source desensitization with irreversible feature mapping. Leveraging Information Bottleneck theory, the edge device performs millisecond-level processing to transform raw imagery into abstract feature vectors via non-linear mapping and stochastic noise injection. This process constructs a unidirectional information flow that strips identity-sensitive attributes, rendering the reconstruction of original images impossible. Subsequently, the cloud platform utilizes multimodal family models to perform joint inference solely on these abstract vectors to detect abnormal behaviors. This approach fundamentally severs the path to privacy leakage at the architectural level, achieving a breakthrough from video surveillance to de-identified behavior perception and offering a robust solution for risk management in high-sensitivity public spaces.
Paper Structure (9 sections, 4 equations, 3 figures)

This paper contains 9 sections, 4 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Architecture
  3. SPA-D Desensitization Algorithm
  4. Privacy Adversarial Game Model
  5. Dual-Feature Blocking Targeting PSZ
  6. PSZ Attention Redirection
  7. Sensitive Value Suppression
  8. System Implementation and Demonstration
  9. Conclusion

Figures (3)

  • Figure 1: The architecture of the Edge-Cloud Collaborative Privacy Protection System.
  • Figure 2: Visualization of the TeleAI Management Dashboard. The interface visualizes real-time analytics including terminal load, behavior risk statistics (e.g., smoking, intrusion), and safety alerts. Note that the system reports a "Violation Alert" for smoking solely through text and icon indicators, with no visual access to the actual scene, verifying the privacy-first design.
  • Figure 3: Hardware Deployment in High-Sensitivity Scenarios. The TeleAI edge device installed on the ceiling of a public restroom. This node executes the source-side irreversible feature mapping, ensuring that raw visual data never leaves the local environment.