Visual context identification for privacy-respecting video analytics

Abstract

With the growing need for privacy-aware and privacy-respecting CCTV systems, it becomes crucial to develop workflows and architectures that can support and enhance privacy protection. Recent advances in image processing enable the automation of many surveillance tasks, increasing the risks of privacy infringements.Fortunately, image processing and pattern recognition techniques can also be used for automatically evaluating the context in which video surveillance takes place, and can therefore be employed for applying context-specific privacy rules. This paper describes how Bag-of-Visual-Words algorithms as well as human tracking and gait analysis cane used for recognizing specific sub-contexts that necessitate the application of particular privacy protection rules in usage contexts such as ambient assisted living, public or workspace surveillanceWe explain how the data of a multi-modal surveillance system should be handled in order to avoid unnecessary processing of sensitive information through Image Quality Descriptors that will support visual classifications by computing reliability measures relating to the image quality such as noise or problems with respect to ambient conditions.

A rule-based methodology and assessment for context-aware privacy

“…As a response to the above, Badii et al in [22] proposed a set of techniques to automatically identify the visual context for the purpose of devising privacy-respecting video analytics. The list of the highlighted techniques includes object recognition, human tracking, and gait analysis as supportive information to identify the presented sub-context and eventually infer the overall scenario and environment.…”

Holistic Privacy Impact Assessment Framework for Video Privacy Filtering Technologies

“…[7], which require more careful deployment to exclusively conceal the privacy-sensitive elements within given video-frames. As a result, the widely-shared view of the stakeholders, as expressed by Andrew Senior [3], for example, has been to call for advances in effectiveness measurement of the privacy protection level afforded by video surveillance solutions; this is the challenge to which the work reported in this and relevant other work [1,4,5] have responded through the pioneering methodologically-guided UI-REF-enabled approach to negotiation-centric co-design as may be applied to context-aware socio-technical personalisation of requirements e.g. for privacy and its evaluation and holistic impact analysis.…”

“…The above evaluation criteria can be similarly applied to privacy-filtered audio-segments if an audio-track is present. These criteria constitute a sub-set of the requirements and evaluation framework as set out within the UI-REF Privacy-by Co-Design Methodology for user-centred, negotiable, context-aware co-evolution and holistic assessment of the impacts of privacypreserving video surveillance as set out in Badii 2012 [4]; these include:…”

“…2. Accordingly the establishment of prototypical templates to delineate an agreed context hierarchy within which various video-analytics operations can be co-specified for specific agreed contexts-objectives; for example, Low-Level Video-Analytics (LLDA), Shallow Vide-Analytics (SVA), Deep Video-analytics (DVA), and, Context-Aware Video-Analytics (CAVA) as sub-spaces of deployment of surveillance technologies [4].…”

“…Although some advances have been recently reported in the application of video privacy filtering techniques, for example as in [3]; the negotiation-centric co-innovation, co-evaluation and thus optimisation of situated privacy filtering has only just begun. This requires participative definition of both video capture context and security context for responsive, trace-able and accountable video-content category judgments supported by a framework such as UI-REF to provide high resolution requirements prioritisation and forensic designation of the Holistic Privacy Evaluation and Impact Assessment Metrics [4,5].…”