Wednesday, 21 Feb 2018
You are here:

Subscribe to our Newsletter

How many eyes has a typical person? (ex: 1)

Twitter Feed

EMail Print


Jens Hälterlein, Leon Hempel (TUB.CTS) visiting Prof. Touradj Ebrahimi (EPFL)
13.11.2014 - 15.11.2014
Internal Visits


This visit was a continuation of the collaboration between EPFL and TUB CTS within the mini-project on mini drones. The mini-project focusses on the evaluation of privacy protection tools in surveillance videos shot with a mini-drone.The main goal of this collaboration was to investigate the evaluations by various stakeholders, especially data protection professionals, law enforcement agencies and security service providers
The mini-project is divided in 5 topics
• Use cases and scenarios
• Datasets
• Tools for mini-drones

• Evaluation methodologies
• Benchmarking
The selection of use-cases took place as follows:
• Input by EPFL: description of use-case regarding the problems to be solved and the solutions envisaged
• Input by TUB CTS: which stakeholders are involved; which laws, norms and practices are affected
• Result: defining the benefits of using video analytics (mini-drones) in this context in general The ensuing selection of scenarios was carried out in an iterative process:
• Input by EPFL: providing a selection of scenarios and a narrative description for each scenario
• Input by TUB CTS: providing a methodology / guideline for these narratives
• Input by EPFL: providing reworked selections of scenarios
• Input by TUB CTS: analyzing scenarios regarding their comprehensiveness
As a result of this preparatory work the use-case “surveillance of temporary parking lots” was chosen. This use-case is a crucial security issue since within the context of major temporary event such as concert or sport event, it is not possible to setup a full-fledged surveillance system. Hence, mini-drones can be used to survey the area and help in managing parking spaces, crowd control, etc. For instance, to provide such a surveillance in a temporary parking lot, mini-drones can be used for scanning the scene and reporting some information, including the number of free lots, identified cars that have not moved for an extended period of time, suspicious behaviour, etc. Special care needs to be taken to insure personal privacy such as protection of license plate numbers and people.
Within this use-case the following scenarios were depicted as relevant:
Scenarios without abnormalities:
• Empty parking
• Half full parking
• Full parking
Scenarios with abnormalities: • Bad parking
• Suspicious behaviour
• Illicit behaviour • Obstacles
Subsequently, EPFL shot a number of video sequences (34 in total, 16-24 seconds each, 720p resolution, 30 fps) for the dataset, using a Phantom 2 Vision+ mini-drone. These video sequences refer to the chosen scenarios as they show activities such as attacking the driver, stealing a car, normal scenario during the day (circling drone), normal scenario during the night and stealing/attacking a pedestrian (13 normal sequences, 6 suspicious sequences, 19 illicit behavior in total). Each sequence was annotated focussing on the regions: face, person, accessory, vehicle and license plate. After that the privacy protection tools pixelization, blurring, masking, warping and morphing were applied to the these regions.
During the visit EPFL presented the filtered video sequences to TUB CTS. This was prefaced by Margherita Bonetto giving a presentation on "Investigation of privacy issues in mini-drones based video surveillance". After that possible evaluation methodologies were discussed. To inform TUB CTS about their experience with subjective evaluation methodologies, Pavel Korshunov gave a presentation on "Crowdsourcing based evaluations of privacy issues in surveillance video". Finally it was agreed on using the LinkedIn social network to contact stakeholders from the relevant backgrounds and professions. The visit has contributed to WP2, WP3 and WP7 activities.