Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ComPrivDet: Efficient Privacy Object Detection in Compressed Domains Through Inference Reuse

Yunhao Yao, Zhiqiang Wang, Ruiqi Li, Haoran Cheng, Puhan Luo, Xiangyang Li

Abstract

As the Internet of Things (IoT) becomes deeply embedded in daily life, users are increasingly concerned about privacy leakage, especially from video data. Since frame-by-frame protection in large-scale video analytics (e.g., smart communities) introduces significant latency, a more efficient solution is to selectively protect frames containing privacy objects (e.g., faces). Existing object detectors require fully decoded videos or per-frame processing in compressed videos, leading to decoding overhead or reduced accuracy. Therefore, we propose ComPrivDet, an efficient method for detecting privacy objects in compressed video by reusing I-frame inference results. By identifying the presence of new objects through compressed-domain cues, ComPrivDet either skips P- and B-frame detections or efficiently refines them with a lightweight detector. ComPrivDet maintains 99.75% accuracy in private face detection and 96.83% in private license plate detection while skipping over 80% of inferences. It averages 9.84% higher accuracy with 75.95% lower latency than existing compressed-domain detection methods.

ComPrivDet: Efficient Privacy Object Detection in Compressed Domains Through Inference Reuse

Abstract

As the Internet of Things (IoT) becomes deeply embedded in daily life, users are increasingly concerned about privacy leakage, especially from video data. Since frame-by-frame protection in large-scale video analytics (e.g., smart communities) introduces significant latency, a more efficient solution is to selectively protect frames containing privacy objects (e.g., faces). Existing object detectors require fully decoded videos or per-frame processing in compressed videos, leading to decoding overhead or reduced accuracy. Therefore, we propose ComPrivDet, an efficient method for detecting privacy objects in compressed video by reusing I-frame inference results. By identifying the presence of new objects through compressed-domain cues, ComPrivDet either skips P- and B-frame detections or efficiently refines them with a lightweight detector. ComPrivDet maintains 99.75% accuracy in private face detection and 96.83% in private license plate detection while skipping over 80% of inferences. It averages 9.84% higher accuracy with 75.95% lower latency than existing compressed-domain detection methods.

Paper Structure

This paper contains 17 sections, 1 theorem, 8 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Setup
  4. System Design
  5. I-Frame Detection
  6. P/B-Frame Detection
  7. Inference Reuse Mechanism
  8. Experiments
  9. Experiment Setup
  10. Ablation Experiments
  11. Evaluating Threshold Variations
  12. Evaluating Detection Latency Components
  13. Comparative Experiments
  14. Comparison with Existing Pixel-Domain Methods
  15. Comparison with Existing Frame-Level Methods
  16. ...and 2 more sections

Key Result

Theorem 4.3.1

Let $p_I$ and $p_{P/B}$ denote the probabilities of I- and P/B-frame occurrence, $p_{ab}$ the probability of an abnormal P/B-frame, and $p_{new}$ that of a new privacy object. We define $C_{I}$ and $C_{P/B}$ as the capability of $\mathcal{F}^I_\mathcal{OP}$ and $\mathcal{F}^{P/B}_\mathcal{OP}$, wher $\blacktriangleleft$$\blacktriangleleft$

Figures (6)

  • Figure 1: The System Overview of ComPrivDet.
  • Figure 2: Examples of Accumulated Motion Vectors and Accumulated Residuals.
  • Figure 3: Evaluation of Abnormality Determination Latency Ratio.
  • Figure 4: Comparison with Existing Pixel-Domain Detectors
  • Figure 5: Comparison with Existing Compressed-Domain Frame-Level Detectors
  • ...and 1 more figures

Theorems & Definitions (5)

  • Definition 4.2.1: Accumulated Motion Vector $MV^{acc}_i$
  • Definition 4.2.2: Accumulated Residual $R^{acc}_i$
  • Definition 4.3.1: Abnormal Predicted Frame
  • Theorem 4.3.1: Capability of ComPrivDet
  • proof