Video cameras for conducting behavioral analyses

Purpose of the project

Le prmier étage du bâtiment MED
Figure 1. The first floor of the MED building

Video surveillance has several important applications. It can be used to study the layout of public spaces, monitor complex urban environments and analyze behavior in busy workplaces. And it has long been one of the central activities of several EPFL labs, including the Computer Vision Laboratory (CVLab) and Signal Processing Laboratories 2 and 5 (LTS2 and LTS5).Until this point, the methods that these labs developed were only applied in the lab setting. But an image acquisition system is now ready to be tried out in the real world, and the MED building has been chosen as the testing ground. Anonymized metadata, like building inflows and outflows, will be extracted from the video images in order to develop models of how the building is used and to determine how these models are affected if the furniture within the building changes. To ensure anonymity and protect users’ privacy, only groups and crowds will be studied rather than individuals. The images will be altered in order to guarantee the anonymity of building users.

The researchers will use this system to test lab-developed algorithms in the real world and then to compare and combine them in order to improve their performance. In addition to producing better algorithms, this project will lead to closer collaboration between the aforementioned labs and to cross-disciplinary projects with other labs interested in areas such as behavioral modeling, sociology and the management of large quantities of data.

How the system works

The image acquisition system is made up of 32 synchronized cameras that cover the ground floor and first floor of the MED building, as shown in Figure 1. The red lines in Figure 2 below show how the cameras will be connected to each other. Figure 3 shows which part of the building the cameras will cover. The system is not an archive, and only a small number of people in charge of maintaining it will be able to physically access it. The files will be destroyed before any external work is carried out on the system. At the end of each week, all images stored on the system’s hard disk will be deleted. Anonymized images needed for research purposes will be extracted and stored as an archive on an EPFL hard disk. Only the researchers will have access to these images, and they will have to log in each time they wish to use them.
Plans du rez-de-chaussée
Figure 2. Ground floor and first floor plans of the atrium in the MED building. The red dots indicate where the cameras can be connected to a power source and the internet.

Measures to ensure anonymity

Informing users when the cameras are on, and providing alternate routes

A light will come on at the building’s main entrances to indicate when the cameras are on. Alternate routes will be indicated so that users can reach all parts of the building without having to pass through areas covered by the cameras.

Anonymized images

Images will be encrypted by the system. If images will be used for research purposes at EPFL, an automatic algorithm will blur the faces before the images can be viewed.
If any images will be used in external publications or in exchanges with other universities, the individuals in the images must first formally consent to the use of the images for research purposes.

Storing images

Images will be saved on a dedicated server to which only a limited number of authorized researchers will have access. A logbook will be used to keep track of who accesses the files and when.

Request to delete images

Encrypted images will be set aside and not used for at least two weeks. During this period, users can send a request to the following address if they would like certain images to be deleted:[email protected]
The email must include the date and time that the images in question were taken. The batch of images will then be deleted without being viewed.