Ethical Training – Concept and Modules
Concept
Since the Centre for Technology and Society at the Technical University Berlin (TUB CTS) has become a member of the VideoSense Consortium a concept of ethical training has been developed that consists of a number of modules. The basic idea to provide these training modules is to enable the processes of engineering to access critical issues and to broaden their perspective.
The primary goal of the training modules is to promote social learning among engineers developing video surveillance applications. To appreciate socio-technical co-evolution means thus to understand and recognize the different roles and conflicting objectives of participating actors and communities, but also the emerging irreversibilities and their potential social impacts on these communities. Thus, learning is needed in order to identify as many perspectives and requirements of diverse stakeholders as possible. The early consideration of these can heavily determine the outcome of how a measure is designed and what implications it will have and impacts it will make when implemented.
For the development of the modules, mini projects respectively use cases from the VideoSense project are used. This ensures a practical bottom-up approach which takes into account the working and learning conditions of engineers.
Training modules
The goal of Training module 1 was for engineers to become more aware of the distance between the engineering practice, the invention of system architectures, the coding of software etc. on the one hand and situations of a real use beyond the mere use case on the other hand. It is thus aimed to increase anticipation and reflexivity in regard to possible impacts and conflicts that may occur, but were not thought of during the engineering process.
The exercise tasks were to firstly describe the use cases in terms of all the important processing steps from the raw video signals to the final information output and secondly to specify in an open manner ethical concerns about the individual process steps.
Partners were asked to indicate whether they see any ethical relevant concern or not and then specify why and finally to add possible solutions whether these are on the regulative (legal), technical or organisational level. As a framework they were asked to use a classification system (dimensions of social impact – scope, physical intrusiveness, distribution and coerciveness – and freedom infringement types – bodily integrity, equal treatment and non-discrimination, freedom of movement, freedom from unlawful detention, presumption of innocence, fair trial and due process, privacy and data protection). Every partner then provided a 15 presentation that was given during the VideoSense meeting in Berlin on August 22nd.
The goal of Training modules 2 and 3 is to increase anticipation and reflexivity in regard to possible privacy and/or social impacts that may occur when algorithms become part of the system architecture of a smart CCTV system. The exercise tasks of module 2 were first to describe one use-case in which the work done by the partners (object tracking, pattern recognition, privacy filters etc.) is implemented into a smart CCTV system. The description should focus on social/spatial context (e.g. train station, shopping mall, administration building), system architecture, data processing, data storage, data access etc. It was up to the partners to refer either to their individual work or to a mini-project.
Even although, concrete use-cases are not predetermined in VideoSense, the partners were asked to imagine a use-case that was (most) likely to become a reality. Cooperation between partners in doing this was encouraged. The second task was to answer 14 questions which refer to the classification system of training module 1, e.g. “What is the specific impact of the smart CCTV system on the freedom of movement? How do these infringements of the freedom of movement impact the person?” As support for this task, the dimensions of social impact were indicated for each question. In this respect the second module continued the training provided by module 1.
In training module 3 the goals of the first two modules were further pursued while at the same time an example was given as to how reflexivity can be implemented into the practice of Video Analytics in Security. Since the TUB CTS is also participating in the EU FP7 project "Security Impact Assessment Measures (SIAM)" it was made possible to invite the VideoSense Consortium to the SIAM User-forum II, which took place in Berlin on October 31st.
The purpose of the SIAM project is to develop criteria and tools for the evaluation of security-relevant aspects on aerodromes as well as public transportation systems. In the SIAM approach to assessment ethical and social impacts play an important role (for further information please visit the SIAM website: (www.siam-project.eu). During the User Forum II the "alpha version" of the SIAM Assessment tool was presented and interactively tested. This application had been developed for administrators of Security Measure Technologies within the public transportation sector.
MediaEval 2013
MediaEval is a benchmarking initiative dedicated to evaluating new algorithms for multimedia access and retrieval. It emphasises the 'multi' in multimedia and focuses on human and social aspects of multimedia tasks.
VideoSense organized the Visual Privacy Task within the MediaEval Grand Challenge 2013 which took place in Barcelona, Spain, on 18-19 October 2013.
This task aims to find ways to protect the privacy of people in video surveillance data. Participants were invited to propose methods to obscure elements in the image which carry information about the identity of subjects. Some examples include faces, ethnicity, gender, and accessories.
The evaluation was carried out by considering three complementary axis, namely the level of privacy protection guaranteed by the filter, the intelligibility (amount of retained information left after applying the filter in order to keep the video understandable) and whether the result was visually acceptable.
Each participant was given pre-annotated sequences from the PEViD dataset, provided by EPFL and annotated by all VideoSense partners.
The proceedings of the workshop are available online: http://ceur-ws.org/Vol-1043/
Here there are some examples of the proposed approaches.
The Second VideoSense Summer School 2013
Following the previous successful Summer School in Dublin, the Second VideoSense Summer School on Ethically-guided and Privacy-respecting Video Analytics was held on 30 June 2013 at the DSP2013 conference, Santorini, Greece.
The VideoSense Summer School aimed at bringing together young researchers working in the field of privacy-respecting and ethnically-aware video analysis in security, offering:
- Lectures delivered by experts from academia and industry providing a clear and in-depth summary of state-of-the-art research in the field.
- An enjoyable and stimulating environment in which participants benefited not only from the formal and practical sessions but also from informal and social interactions with established researchers and the other participants in the School.
To further facilitate communication and feedback all attendees presented a poster on their research.
Invited Speakers:
- Professor Atta Badii (University of Reading, UK) - Lecture on “Integrated Context-aware Tools and Architecture for Delivering and Evaluation of Privacy-Preserving Video-Analytics”.
- Professor Ebrahimi Touradj (Swiss Federal Institute of Technology (EPFL), Switzerland) - Lecture on “Security and Trust in social media networks”
- Prof. Jean-Luc Dugelay (EURECOM, France) - Lecture on “Facial Image Processing in Security Applications”
- Christian Fedorczak (Thales, France) - Lecture on “Video-surveillance and Urban Security, the Mexico city example”
- Dr. Ivo Keller (TU Berlin, Germany) - Lecture on “State-of-the-art and trends in video surveillance