GeoFair: Privacy-Aware Representation Learning for Cross-Camera Identity Matching in Urban Surveillance

Rong Fu, Anqi Zhao, weizhi Tang, YIBO MENG, Simon James Fong

Keywords: Representation Learning, Cross-Camera Identity Matching, Geometry-Aware Attention, Privacy-Preserving AI, Fairness in Surveillance, Transformer Architectures, Metric Learning, Urban Video Analytics

TL;DR: We propose a geometry-aware transformer framework for privacy-preserving and fair identity matching across camera networks.

Abstract: This paper presents a geometry-aware transformer framework for cross-camera identity association in urban surveillance systems. The proposed architecture integrates spatial topology constraints into attention mechanisms and employs adaptive metric learning to enhance representation robustness. By embedding camera geometry into feature learning and dynamically adjusting decision boundaries, the system achieves consistent identity matching across disjoint views. Privacy-preserving indexing and fairness-aware optimization are incorporated to ensure ethical deployment. Extensive evaluations demonstrate improved retrieval accuracy, resilience to occlusions and adversarial perturbations, and compliance with privacy standards. This work contributes a scalable and trustworthy solution for intelligent video analytics in distributed environments.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Code Of Ethics: true

Submission Guidelines: true

Anonymous Url: true

No Acknowledgement Section: true

Submission Number: 7357