TSDCRF: Balancing Privacy and Multi-Object Tracking via Time-Series CRF and Normalized Control Penalty

2026-03-13

AI Summary

This work addresses the challenge of preserving identity privacy in multi-object tracking, where appearance or positional cues often leak sensitive information, and naive injection of differential privacy noise typically causes ID switches or trajectory fragmentation. To mitigate this, the authors propose the TSDCRF framework, which integrates (ε,δ)-differential privacy via Gaussian noise with a Normalized Control Penalty (NCP) and a Temporal Dynamic Conditional Random Field (DCRF) to enhance cross-frame trajectory consistency while ensuring strong privacy guarantees. The method is compatible with mainstream detectors and trackers such as YOLOv4 and DeepSORT, and achieves state-of-the-art performance on MOT16, MOT17, Cityscapes, and KITTI benchmarks, significantly outperforming existing approaches like NTPD and PPDTSA. It effectively reduces both KL divergence drift and tracking RMSE, thereby achieving a superior trade-off between privacy preservation and tracking utility.

Technology Category

• Computer Vision: Motion & Tracking
• Machine Learning: Privacy
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Security and Privacy: Tracking, profiling, and countermeasures against them
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Multi-object tracking in video often requires appearance or location cues that can reveal sensitive identity information, while adding privacy-preserving noise typically disrupts cross-frame association and causes ID switches or target loss. We propose TSDCRF, a plug-in refinement framework that balances privacy and tracking by combining three components: (i) $(\varepsilon,δ)$-differential privacy via calibrated Gaussian noise on sensitive regions under a configurable privacy budget; (ii) a Normalized Control Penalty (NCP) that down-weights unstable or conflicting class predictions before noise injection to stabilize association; and (iii) a time-series dynamic conditional random field (DCRF) that enforces temporal consistency and corrects trajectory deviation after noise, mitigating ID switches and resilience to trajectory hijacking. The pipeline is agnostic to the choice of detector and tracker (e.g., YOLOv4 and DeepSORT). We evaluate on MOT16, MOT17, Cityscapes, and KITTI. Results show that TSDCRF achieves a better privacy--utility trade-off than white noise and prior methods (NTPD, PPDTSA): lower KL-divergence shift, lower tracking RMSE, and improved robustness under trajectory hijacking while preserving privacy. Source code in https://github.com/mabo1215/TSDCRF.git

Problem

• multi-object tracking
• privacy preservation
• differential privacy
• identity switch
• trajectory consistency

Innovation

• Differential Privacy
• Multi-Object Tracking
• Time-Series CRF
• Normalized Control Penalty
• Privacy-Utility Trade-off