Group members

Sub-Groups structure


Group members: 72

The analysis of video material collected by national and regional Law Enforcement Agencies (LEA) from many varied sources, including CCTV and handled cameras, has become a critical component in investigations related to major criminal acts and terrorist attacks. The amount of video data of interest for legal investigations is continuously increasing with the rapid deployment of video equipment, surveillance cameras in public and private areas, body-worn cameras of police forces, smartphones or digital cameras used by bystanders recording crime scenes, or even the criminals/terrorists themselves posting videos on the Web that they recorded during the act.

After a terrorist attack or a major criminal act, each hour is precious to find clues that can help to capture the offenders and prevent further acts, and accelerate and secure (regarding the exposure of the police forces) the operations led to capture the offenders to save lives and prevent further damage.

Video investigations are as old as video recordings of crime scenes exist. It consists in LEA investigators viewing the content of all available videos to search for clues that can help them to identify offenders and gather information about the circumstances of the crime, or legal evidences for a lawsuit. In early days (e.g. president Kennedy shooting) this could be easily done because the available video recordings could be counted in minutes, whilst nowadays it must be counted in 1,000s of hours. In spite of the huge growth of available video material from a variety of sources, the whole video investigation work is still mostly done manually by LEA officers. They must prepare each video record, i.e. load the video, convert the video into a standard format, and then view (and possibly review) the records one by one until the human eye detects some elements of interest. This is obviously very time consuming and extremely demanding in human resources and capabilities. Some concrete figures highlight the level of this challenge:  after the Toulouse and Montauban shootings in France in 2012 (known as the Merah case), 35 Terabytes of video footages were collected from various sources representing 10,000 hours of recordings to analyse. The effort required to view all the available material represented 2 person-years. With the proliferation of video registration over the last decades, it has hence become impossible for LEAs to analyse all the available video material: a survey among French police officers revealed that less than 10% of all collected videos of interest are analysed.

There is hence an urgent need for efficient tools that can assist the LEAs in their daily video investigation tasks, to help them to process the huge volume of video material with less effort, and to find clues and evidences faster. The urgency of addressing this need has become particularly clear after the recent vague of terrorist attacks that hit European countries.

The core technology for such tools is video analytics. Video analytics is a technology from the computer vision domain, aiming at finding and recognising objects, persons, scenes, etc. Many previous projects have developed video analytics for other applications, which are also relevant for video investigation tools. However, the current TRL of these analytics is only sufficient for applications in a controlled environment, i.e. were camera position and parameters, lighting conditions, image quality, etc., are suitable. For example, object recognition, face recognition, number plate recognition can be exploited for real world applications in a controlled environment.

The TRL of these analytics is however too low for using them in video investigation tools, where the environment conditions of the videos vary a lot depending  on the source of videos: the collected video material comes from a variety of sources in the public domain (e.g. from surveillance systems in streets, airports, subway or train stations, etc.), from private institutions (banks, hotels, gas stations, etc.), but also video registrations provided by bystanders, or posted on the Web. As a consequence, the analytics used for video investigation must be robust to different environment conditions such as the camera characteristics, image quality, viewing angles, moving cameras (handheld cameras or smartphones), etc.

In the video investigations operational environment, today's video analytics technology generates error rates that are too high for LEA investigators to gain trust in such tools. The research and innovation efforts dedicated to the video investigation tools was yet too limited to get beyond the prototype stage, and hence no product has been recognized as robust enough by the LEAs. As a consequence, currently the main tool available to LEA investigators for gaining time in video investigations is to accelerate (by typically X4) the viewing speed.

One of the main roadblocks for increasing the TRL of video analytics in the legal investigation domain is the lack of representative data, needed for research and development. Most of the real world data used in LEA investigations cannot be provided to the technology developers and tools vendors for legal reasons.

Additionally, the few video investigation tools available are generally closed systems, limited to the video analysis technology mastered in-house by the vendor (e.g. motion detection and limited object recognition), and not open to integrate other video analytics technologies that could be provided by third parties. This limitation hinders the development of innovation activities leading to new products that could offer to end-users enhanced features and/or performance, and flexibility to better fit specific user requirements, and globally the need of video investigation systems to adapt to evolving user needs and the continuous progress of technology. Innovation in video investigation tools remains hence limited and no real products are widely used by the LEAs community yet.

The absence of suitable and LEA trusted technical solutions to support video investigations leads to a threefold problem for LEAs:

  1. missing crucial or useful information to lead rapidly optimally targeted actions against the offenders or to be able to solve a case,
  2. spending a huge amount of time on unproductively sifting through massive amounts of video content,
  3. using tools that are error prone and thus counterproductive.

There is hence an urgent need for targeted research for developing accurate and robust video analytics, and it can succeed based on the following facts:

  • LEAs have now a good vision of the features they need and would expect from video investigation tools from operational view, and can specify them,
  • The multiplication of digital cameras and generalisation of high resolution video recordings will contribute to reach higher analytics accuracy,
  • New system architectures using big-data concepts will be able to accelerate the processing of the huge and steadily increasing video data volumes,
  • The use of deep-learning technologies for the analytics can significantly improve the robustness of video analytics - provided they can be fed with representative data,
  • Video investigation is a post-event activity, i.e. it is carried out on a set of video files; consequently video analytics in this context has no real-time constraints which reduces somewhat the technical difficulties faced in video-surveillance (real time).

There is yet a clearly increasing demand for efficient tools to improve LEAs' video investigation capabilities in their daily practice. The video investigation tool market is yet still in its early stages, but market experts forecast a rapid market growth from 2019 (25%/year), leading to market of 140 M€ in 2022. There is a unique opportunity for European companies to take a leading position in this upcoming market, provided that they join their forces based on a set of strong common concepts.


    The analysis of multimedia content collected by Law Enforcement Agencies (LEAs) has become a critical component in investigations related to major criminal acts and terrorist attacks.

    The amount of video data relevant for legal investigations is continuously increasing with the rapid deployment of video equipment.

    In spite of this growth, the whole video investigation work is still mostly carried out manually by the LEA officers and need unreasonable time.

    The VICTORIA Project funded by the European Union’s Horizon 2020 research and innovation programme under grant agreement N°740754 is addressing this need. Its consortium of 14 partners including LEAs, research groups, academics and industrial companies, world-leaders in security markets have joined for a common aim: deliver an ethical and legally compliant video analysis platform to accelerate the video analysis tasks of LEAs.


      LowRes Concept


        More files


          Within the 3 year project, VICTORIA develops:

          • A robust and accurate TRL-6 video analysis technology, that will boost LEA’s investigative capacity;
          • Increased usability of video analysis tools;
          • Cost and time reduction for video investigation;
          • Crimes solved faster, by using all LEA’s available videos;
          • Prevention of further terrorist attacks;
          • Better citizens protection;
          • Trained LEA investigators in the use of video investigation tools;
          • Integration of ethical considerations and the EU legal framework, including the General Data Protection Regulation;
          • Grounds for further innovation potential.

          VICTORIA Public Website