[Paper Review] A Review on Visual Privacy Preservation Techniques for Active and Assisted Living

Siddharth Ravi, Pau Climent-Pérez+1 more|arXiv (Cornell University)|Dec 17, 2021

Video Surveillance and Tracking Methods4 citations

TL;DR

This paper proposes a novel taxonomy for classifying visual privacy preservation techniques in active and assisted living (AAL) environments, emphasizing perceptual obfuscation to protect bodily privacy and machine obfuscation to defend against unauthorized AI analysis. It links these methods to privacy-by-design principles and identifies critical research gaps in environmental privacy, non-standard cameras, and user acceptance.

ABSTRACT

This paper reviews the state of the art in visual privacy protection techniques, with particular attention paid to techniques applicable to the field of active and assisted living (AAL). A novel taxonomy with which state-of-the-art visual privacy protection methods can be classified is introduced. Perceptual obfuscation methods, a category in the taxonomy, is highlighted. These are a category of visual privacy preservation techniques particularly relevant when considering scenarios that come under video-based AAL monitoring. Obfuscation against machine learning models is also explored. A high-level classification scheme of the different levels of privacy by design is connected to the proposed taxonomy of visual privacy preservation techniques. Finally, we note open questions that exist in the field and introduce the reader to some exciting avenues for future research in the area of visual privacy.

Motivation & Objective

• To address the growing need for visual privacy in active and assisted living (AAL) environments where video monitoring supports elderly care but risks violating bodily and identity privacy.
• To identify and classify existing visual privacy preservation techniques, especially perceptual obfuscation, which protects privacy through human perception rather than cryptographic means.
• To connect low-level technical methods to high-level privacy-by-design principles, ensuring privacy is embedded from the system's inception.
• To highlight under-researched areas such as environmental privacy, gait anonymization, and privacy preservation in non-standard cameras (e.g., fisheye, thermal).
• To explore social and legal dimensions, including user trust, the privacy paradox, and the acceptability of reversible obfuscation in forensic or legal contexts.

Proposed method

• Proposes a novel four-category taxonomy for visual privacy preservation: intervention methods, blind vision, secure processing, and data hiding, with a focus on perceptual obfuscation.
• Classifies techniques based on their mechanism: perceptual obfuscation (e.g., blurring, pixelation) to protect bodily privacy, and machine obfuscation (e.g., adversarial perturbations) to evade ML models.
• Integrates the taxonomy with the privacy-by-design framework, aligning technical solutions with systemic, proactive privacy engineering principles.
• Analyzes the limitations of current methods in handling non-standard cameras (e.g., omnidirectional, thermal) due to image distortion and lack of model compatibility.
• Evaluates perceptual obfuscation’s side effects on environmental privacy, noting that while small objects (e.g., credit card numbers) may be obscured, larger objects may still leak sensitive information.
• Calls for interdisciplinary research combining technical, social, and legal perspectives to assess user perception, acceptance, and the ethical implications of reversible obfuscation.

Experimental results

Research questions

• RQ1 How can visual privacy preservation techniques be systematically classified in the context of AAL systems?
• RQ2 To what extent do perceptual obfuscation methods protect bodily privacy without compromising system utility in AAL monitoring?
• RQ3 What are the limitations of current privacy-preserving techniques when applied to non-standard camera types such as fisheye or thermal cameras?
• RQ4 How do social and legal factors, including the privacy paradox and user trust, influence the adoption of obfuscation techniques in real-world AAL deployments?
• RQ5 What are the implications of reversible obfuscation methods for legal and forensic use, particularly regarding data integrity and information loss?

Key findings

• Perceptual obfuscation methods, such as blurring and pixelation, are highly relevant for protecting bodily privacy in sensitive AAL environments like toilets and bedrooms.
• Machine obfuscation techniques, including adversarial perturbations, are effective in protecting identity from unauthorized machine learning models but are underexplored in AAL contexts.
• Environmental privacy—protecting sensitive objects like credit cards and address labels—is an under-researched area, despite being critical for comprehensive privacy protection.
• Non-standard cameras such as omnidirectional and thermal cameras pose significant challenges due to image distortion and lack of compatibility with existing detection and obfuscation pipelines.
• Reversible obfuscation methods raise legal concerns, as reconstructed images may contain stochastic noise, potentially undermining their admissibility in forensic or judicial settings.
• The privacy paradox—where users express concern but act contrary to their stated preferences—undermines the validity of self-reported user acceptance studies, necessitating more robust preference measurement methods.