PrivacySecurityTrustworthinesscs.CV
RL-MoE: An Image-Based Privacy Preserving Approach In Intelligent Transportation System
Abdolazim Rezaei, Mehdi Sookhak, Mahboobeh Haghparast
August 7, 2025
International Conference on Cognitive Machine Intelligence
1 citations
Key Takeaways
-
Converts images to text instead of transmitting raw video: RL-MoE transforms sensitive camera footage into textual descriptions rather than sending images directly, fundamentally reducing privacy risks in traffic monitoring systems where cameras are increasingly deployed.
-
Achieves strong privacy protection against reconstruction attacks: The framework reduced the success rate of replay attacks to 9.4% on the CFP-FP dataset, demonstrating that the approach effectively prevents attackers from reconstructing original images from the data being transmitted.
-
Balances privacy and utility through dual optimization: Unlike existing methods that force a choice between privacy (through blurring/encryption) and data quality, RL-MoE uses reinforcement learning to simultaneously optimize for both semantic accuracy of descriptions and privacy preservation, avoiding the traditional trade-off.
-
Enables trustworthy smart city infrastructure: By providing a practical solution for privacy-preserving visual data in Intelligent Transportation Systems, this approach supports the development of autonomous vehicles and smart city networks that can collect necessary operational data without compromising citizen privacy rights.
Abstract
The proliferation of AI-powered cameras in Intelligent Transportation Systems (ITS) creates a severe conflict between the need for rich visual data and the right to privacy. Existing privacy-preserving methods, such as blurring or encryption, are often insufficient due to creating an undesirable trade-off where either privacy is compromised against advanced reconstruction attacks or data utility is critically degraded. To resolve this challenge, we propose RL-MoE, a novel framework that transforms sensitive visual data into privacy-preserving textual descriptions, eliminating the need for direct image transmission. RL-MoE uniquely combines a Mixture-of-Experts (MoE) architecture for nuanced, multi-aspect scene decomposition with a Reinforcement Learning (RL) agent that optimizes the generated text for a dual objective of semantic accuracy and privacy preservation. Extensive experiments demonstrate that RL-MoE provides superior privacy protection, reducing the success rate of replay attacks to just 9.4\% on the CFP-FP dataset, while simultaneously generating richer textual content than baseline methods. Our work provides a practical and scalable solution for building trustworthy AI systems in privacy-sensitive domains, paving the way for more secure smart city and autonomous vehicle networks.