Thursday, 21 Sep 2017
You are here:

Subscribe to our Newsletter

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

Twitter Feed

EMail Print

Event 

Title:
Mathieu Einig (UoR) visiting Christian Fedorczak (Thales)
When:
18.03.2013 - 25.03.2013
Category:
Internal Visits

Description

This visit was to evaluate Thales’ surveillance system architecture in terms of necessary components for enabling privacy protection technologies.
The aim of this placement was to receive formal training from Thales on their large-scale video surveillance systems in order to ensure that algorithms developed as part of the VideoSense could be deployed in the industrial R&D context. Their framework can handle the processing of video streams in both an internal (program developed with their API for communicating) and external (executable communicating with webservices) fashion. Events can be created from the video processing stage and can then be used to update a scriptable state-machine, and be stored into a database. The video and its metadata are then synchronised and sent to a video wall, which can display several videos with their overlay.
The integration of algorithms involved straightforward C++ modules analysing frames and sending events or objects that can be added through their API. This was evaluated by developing a test algorithm based on OpenCV that detected human faces, counted them, and triggered an alarm depending on their number. However, in the current state of the framework, the video stream cannot be modified and only basic overlays such as lines and boxes can be displayed on top by the video player, meaning that advanced privacy protection filters cannot be integrated at the moment. Two possible solutions were discussed: sending pre-filtered raw bitmaps as overlays, or sending special objects that would trigger the filtering on the video-player side.
.Current surveillance systems do not have privacy requirements in their design, making it difficult to include mitigation technologies.
Even although the experiments on algorithm integration were quite successful, some extra work will be needed on the surveillance system architecture in order to allow the filtering of the videos displayed on the video wall. This would provide an opportunity for further collaboration between UoR and Thales.