Privacy-preserving surveillance: an interdisciplinary approach

Intelligent video surveillance is an active and lively field of research, predominantly in the domains of image exploitation and situation assessment. The availability of privacy-invasive system functionality such as real-time object tracking and automatic extraction of biometric features is becoming reality. Not surprisingly, video surveillance generates an increasing interest among information security and privacy researchers. A categorical argument against video surveillance targets the chilling effect of such systems, which arguably is in conflict with the fundamental right to free development of the individual. When faced with surveillance cameras, we cannot know whether we are currently observed or not. However, the mere possibility of being observed tends to change the way we behave, which usually is considered an undesired phenomenon in free societies and therefore addressed by legislation. The principle of proportionality, as laid down in Articles 8(2) and 52(1) of the Charter of Fundamental Rights of the European Union, demands a careful weighing of the purpose of a surveillance measure, ie, the legally protected interest to be defended, against the legitimate interests of people affected by the surveillance measure. However, we do observe that video surveillance is spreading rapidly, even though the proportionality of privacy invasion and utility may not always be justified. In addition, even if we consider video surveillance to be necessary in particular cases, the question of how and to which extent privacy of the people concerned can be preserved must be evaluated. Given that modern video surveillance technology works at the level of abstracted objects rather than raw video streams, we argue that the computer vision capabilities of such systems can also be exploited for improving the selectiveness of surveillance measures. Intelligent video surveillance systems are capable of fusing information extracted from video streams into abstracted objects, including attributes such as IDs by face recognition, location, or certain activities. Hence, we can analogously incorporate an authentication mechanism as an information source, which enables the system to determine (group) identities of people who are a priori known to be concerned by the surveillance measure, eg, employees of an airport as an environment,