PrivHAR-Bench: A Graduated Privacy Benchmark Dataset for Video-Based Action Recognition

2026-04-01

AI Summary

Existing research on privacy-preserving human activity recognition predominantly relies on a binary paradigm using either clear videos or a single privacy transformation, lacking standardized evaluation of the fine-grained trade-off between privacy strength and recognition utility. This work proposes PrivHAR-Bench, the first benchmark dataset encompassing nine incrementally stronger privacy levels—from lightweight spatial blurring to encrypted block permutation—augmented with background-removal variants. The benchmark provides fully annotated data, standardized train/validation/test splits, and an integrated evaluation toolkit. Experiments using R3D-18 reveal that recognition accuracy declines from 88.8% to 53.5% as privacy intensity increases, while cross-domain performance drops precipitously to 4.8%, thereby systematically uncovering, for the first time, an interpretable privacy-utility degradation curve.

Abstract

Existing research on privacy-preserving Human Activity Recognition (HAR) typically evaluates methods against a binary paradigm: clear video versus a single privacy transformation. This limits cross-method comparability and obscures the nuanced relationship between privacy strength and recognition utility. We introduce \textit{PrivHAR-Bench}, a multi-tier benchmark dataset designed to standardize the evaluation of the \textit{Privacy-Utility Trade-off} in video-based action recognition. PrivHAR-Bench applies a graduated spectrum of visual privacy transformations: from lightweight spatial obfuscation to cryptographic block permutation, to a curated subset of 15 activity classes selected for human articulation diversity. Each of the 1,932 source videos is distributed across 9 parallel tiers of increasing privacy strength, with additional background-removed variants to isolate the contribution of human motion features from contextual scene bias. We provide lossless frame sequences, per-frame bounding boxes, estimated pose keypoints with joint-level confidence scores, standardized group-based train/test splits, and an evaluation toolkit computing recognition accuracy and privacy metrics. Empirical validation using R3D-18 demonstrates a measurable and interpretable degradation curve across tiers, with within-tier accuracy declining from 88.8\% (clear) to 53.5\% (encrypted, background-removed) and cross-domain accuracy collapsing to 4.8\%, establishing PrivHAR-Bench as a controlled benchmark for comparing privacy-preserving HAR methods under standardized conditions. The dataset, generation pipeline, and evaluation code are publicly available.

Problem

• Privacy-Utility Trade-off
• Human Activity Recognition
• Video Privacy
• Benchmark Dataset
• Action Recognition

Innovation

• Privacy-Utility Trade-off
• Graduated Privacy Benchmark
• Video-based Action Recognition
• Human Activity Recognition
• Privacy-preserving Evaluation