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

Bo Ma; Jinsong Wu; Weiqi Yan

cs.CVMar 14, 2026

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

Authors: Bo Ma, Jinsong Wu, Weiqi Yan

Organizations: School of Mathematics and Computer Engineering, Auckland University of Technology,Auckland 1024, New Zealand · Department of Electrical Engineering, University of Chile, Santiago, Chile

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

Figures & tables

Fig. 1: Segmentation and tracking results. Top: three consecutive frames from MOT16. Bottom: segmentation and trajectory after target identification.

Fig. 3: Flowchart of the proposed framework: detector $\rightarrow$ tracker $\rightarrow$ privacy module $\rightarrow$ TSDCRF refinement.

Fig. 4: Motion trajectory prediction with the CRF (Fig. 4). The time-series CRF enforces temporal consistency and reduces position deviation after privacy noise.

Fig. 5: The performance of the proposed methods. (a) Defense under trajectory hijacking attack; (b) defense under blind target object detection attack.

Fig. 6: Comparison of sensitive area distributions

Fig. 7: Comparison of degree distribution and related parameter $k$ .

Fig. 8: Comparison of degree distribution and sampling steps

Fig. 9: Privacy retrieval vs. path/structure: comparison of different methods (see text for metric).

Fig. 10: Privacy retrieval frequency vs. privacy budget $\epsilon$ .

Fig. 11: Privacy appearance frequency vs. parameter $k$ .

Fig. 12: Privacy appearance frequency vs. sampling step.

Fig. 13: Root-mean-square deviation of tracking bounding box

Method	$\varepsilon=0.5$	$\varepsilon=1.0$	$\varepsilon=1.5$
NCP+Gaussian (Ours)	0.18	0.21	0.24
PPDTSA	0.32	0.41	0.48
NTPD	0.52	0.61	0.68
White noise	2.18	2.52	2.78

TABLE I: RMSE of tracking bounding box (from Fig. 13 ), averaged over MOT16, MOT17, Cityscapes, and KITTI. Lower RMSE = better tracking accuracy.

Method	$\varepsilon=0.25$	$\varepsilon=0.5$	$\varepsilon=0.75$	$\varepsilon=1.0$
NCP+Gaussian (Ours)	21.2	24.1	26.3	28.0
White noise	17.8	19.5	20.9	22.1
NTPD	19.6	22.4	24.5	26.2
PPDTSA	20.1	23.0	25.0	26.8

TABLE II: PSNR (dB) of perturbed vs. original bounding-box coordinates (sensitive detections), averaged over MOT16, MOT17, Cityscapes, and KITTI. Higher PSNR = less distortion. Computed from MSE via

\mathrm{PSNR}=10\log_{10}(\mathrm{MAX}^{2}/\mathrm{MSE})

Dataset	$\varepsilon=0.25$	$\varepsilon=0.5$	$\varepsilon=0.75$	$\varepsilon=1.0$
MOT16	21.5	24.3	26.5	28.2
MOT17	21.0	23.9	26.1	27.8
Cityscapes	20.8	23.8	26.0	27.7
KITTI	21.4	24.2	26.2	27.9

TABLE III: PSNR (dB) for NCP+Gaussian (Ours) per dataset at different

\varepsilon

. Explains dataset-wise differences seen in Figs. 7 , 6 , and 13 .

Multi-Object Tracking

Multi-object tracking (MOT) links detections across video frames to maintain object identities through motion, occlusion, and changing viewpoints.

See all 193 papers

PPEDCRF: Privacy-Preserving Enhanced Dynamic CRF for Location-Privacy Protection for Sequence Videos with Minimal Detection Degradation
Mar 2, 2026Bo Ma, Jinsong Wu, Weiqi Yan +2Conditional Random Fields
Auckland University of Technology, New Zealand. · the Department of Electrical Engineering, Universidad de Chile, Chile.
PPEDCRF: Dynamic-CRF-Guided Selective Perturbation for Background-Based Location Privacy in Video Sequences
Apr 18, 2026Bo Ma, Weiqi Yan, Jinsong WuConditional Random Fields Adversarial Perturbations
Resideo Technologies Inc., Scottsdale, Arizona, USA · Auckland University of Technology, Auckland, New Zealand · Guilin University of Electronic Technology, Guilin, China
Drift-Resilient Temporal Priors for Visual Tracking
Apr 3, 2026Yuqing Huang, Liting Lin, Weijun Zhuang +2Visual Object Tracking
Harbin Institute of Technology, Shenzhen · Pengcheng Laboratory · Lero, the Research Ireland Centre for Software, University of Limerick +1

Mar 2, 2026J/K move · Enter open · S save

PPEDCRF: Privacy-Preserving Enhanced Dynamic CRF for Location-Privacy Protection for Sequence Videos with Minimal Detection Degradation

Bo Ma, Jinsong Wu, Weiqi Yan, Catherine Shi+1

Auckland University of Technology, New Zealand. · the Department of Electrical Engineering, Universidad de Chile, Chile.

Conditional Random Fields

Open paper AskPDF arXiv

Dashcam videos collected by autonomous or assisted-driving systems are increasingly shared for safety auditing and model improvement. Even when explicit GPS metadata are removed, an attacker can still infer the recording location by matching background visual cues (e.g., buildings and road layouts) against large-scale street-view imagery. This paper studies location-privacy leakage under a background-based retrieval attacker, and proposes PPEDCRF, a privacy-preserving enhanced dynamic conditional random field framework that injects calibrated perturbations only into inferred location-sensitive background regions while preserving foreground detection utility. PPEDCRF consists of three components: (i) a dynamic CRF that enforces temporal consistency to discover and track location sensitive regions across frames, (ii) a normalized control penalty (NCP) that allocates perturbation strength according to a hierarchical sensitivity model, and (iii) a utility-preserving noise injection module that minimizes interference to object detection and segmentation. Experiments on public driving datasets demonstrate that PPEDCRF significantly reduces location-retrieval attack success (e.g., Top-k retrieval accuracy) while maintaining competitive detection performance (e.g., mAP and segmentation metrics) compared with common baselines such as global noise, white-noise masking, and feature-based anonymization. The source code is in https://github.com/mabo1215/PPEDCRF.git

Abstract

Multi-Object Tracking

Explore similar work

PPEDCRF: Privacy-Preserving Enhanced Dynamic CRF for Location-Privacy Protection for Sequence Videos with Minimal Detection Degradation

PPEDCRF: Dynamic-CRF-Guided Selective Perturbation for Background-Based Location Privacy in Video Sequences

Drift-Resilient Temporal Priors for Visual Tracking