The objective of FILTER2 (FILtrage negaTif des contEnus de videopRotection 2.0) is to provide the forensics investigators faced with huge digital CCTV files with automatic solutions guiding them in the optimized sequence to use when such files or segments thereof) need to be processed by video analytics. As illustrated by the recent events, the investigators face two types of situations: if dangerous suspects escaped, the vital priority is to concentrate on the pieces of evidence with the highest probability to locate the individual, even if further in depth analysis may be required later in the process; in the formal investigation phase, priority is to extract whatever evidence is available in the file and time is no more the priority.
This operational need is a real scientific and technical challenge. It has been identified within the COFIS Video-surveillance Demonstrator VOIE (a government – security industry joint venture), by PTS (the Government investigators) and formalized by Thales. Only option found was to submit it, through the FILTER2 project to three academic teams recognized in field, while taking advantage of the real data available in VOIE.
In front of a big mass digital video files, at best time-stamped and with the sources geo-localized, the scientific challenge is to develop an automatic high speed (more than 16 times fast replay) usability indexation tool (with the two use cases in mind) and to build thereon the most effective strategy to conduct the video content analysis.
Different agnostic (i.e. without any knowledge of the source) methods will be developed to work in the compressed domain and to be run without any human support at “fast speed”; metadata, typically as defined by ISO 22311, will be analyzed whenever available. Then, using the knowledge of the operational experts, decision trees will be developed, integrated with a human interface and carried on the VOIE Demonstrator.
This CoFIS video-surveillance demonstrator has been founded to be kept active by successive projects. The first one is FUI 19 VOIE, stated in June 2015 for 30 months. An ethical committee (where CNIL, the French Privacy Authority sits) is attached to it. The forensic investigation subproject (SP4) is hosted by Police Judiciaire and Prefecture de Police, both being FILTER2 partners. FILTER2 relies on these forensic experts and on research teams of IRIT (Toulouse), of XLIM (Poitiers) and of GREyC (Caen). The coordinator, THALES Communications & Security SA (TCS) takes care of the integration and of the interfacing with VOIE, preparing a quick exploitation of the results obtained.
The above partners are complementary, are used to work together and are recognized at the international level: TCS is a world leader in security systems and in cooperative research; it is the coordinator of FUI 19 VOIE and the ISO convener of ISO 22311 standard. IRIT, UMR CNRS 5505 is specialized in information extraction, analysis and modeling (especially in multimedia metadata management). XLIM-SIC, UMR CNRS 5272 is a recognized expert in quality evaluation of color imagery (especially for CCTV, where it collaborates with the UK Home Office and with the Westminster University on London busses). GREyC-IRF, UMR CNRS 6072 is specialized in imagery analysis and biometry (especially with regard to video quality assessment for persons identification and CCTV). The PTS (DCPJ and Identite Judiciaire) will bring their technical and operational knowledge of the requirements and will evaluate the solutions developed.
Madame Sonia Lejot (THALES COMMUNICATIONS & SECURITY SA)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Préfecture de Police Ministère de l'Intérieur
GREyC Groupe de Recherche en Info, Image, Automatique et Instrumentation de Caenrmatique
XLIM XLIM - Université de Poitiers
UPS-IRIT Université Toulouse III Paul Sabatier
TCS THALES COMMUNICATIONS & SECURITY SA
Help of the ANR 711,997 euros
Beginning and duration of the scientific project: October 2016 - 36 Months