Technologies d’Analyse Vidéo dans le domaine de la sécurité Séminaire MEITO - Analyse vidéo et reconnaissance d'image 01 Avril 2010
Jean-Yves Dufour Responsable du laboratoire Thales / CEA-LIST « Vision Lab » Thales Services / ThereSIS Campus Polytechnique,  1, avenue Augustin Fresnel 91767 Palaiseau Cedex France Tél : 01 69 41 59 64 Mail :  [email_address]
Plan de la présentation Video Surveillance :  généralités Principaux domaines d’application Video Surveillance Intelligente Quelles fonctionnalités ? Transfert recherche / industrie Labo Vision de Thales / CEA
What is Video Surveillance ? Surveillance of an infrastructure (store, parking lot, airport, …) through images provided by video cameras observing “strategic” areas.  two key purposes : Recording evidence  for investigative purposes  Providing a human operator with images enabling him to analyse  the situation  generally : to detect and  to react Huge expansion of Video-Surveillance : Due to early 2000 events :governments have made personal and asset security a priority in their policies.     Deployment of large CCTV systems.  Example, London Underground and Heathrow Airport have more than 5000 cameras each (2004).     Major research domain and commercial market or computer vision  :  Many collaborative funded projects, Many scientific  international workshops  / seminars / papers … Hundreds of companies created
Main application domains Mainly oriented towards security (citizens, assets)  Transports Railway / underground transportation (stations, tracks , tunnels, wagon) Airports,  Urban and motorway road networks, Maritime transportation Large urban security (roads, streets, public place, …)  Public events :  sports, leisure,  pilgrimage,  … Industrial environments factories, pipe-line, energy power  plants, … Border surveillance Military applications (remote surveillance, protection of installations)
Smart Video Surveillance Issues : Huge amount of image data (thousands of sensors in large systems) Human limitations  :  lack of attention span: Human operator cannot maintain an acceptable level of attention for long time Sifting through large collections of surveillance videos is tedious and error prone for a human investigator.    Manual surveillance becomes impractical     Needs for “situational awareness”     “ Smart video surveillance” ? Assist people in surveillance task: the computers watch the video and produces  Real-time alerts  …  without fully replacing them : People still have to deal with the (false) alarms Assist people for post-analysis tasks :  Indexing video to ease research Providing tools to accelerate research
Which functionalities ? Detect intrusion Presence of somebody in forbidden zone Needs to detect moving object in protected area Somebody enters a protected area through forbidden way Needs to spatially analyse trajectory of people (tracking)
Which functionalities ? Loitering : Somebody stays too long time in an “sensible” area Counting (people, vehicles, lorries, …) : how many objects pass through a given area    Detection of moving objects    Classification (people / vehicle or track / lorry, …)    Tracking to ensure that same person is not counted more than once) Source : Thales Italy and MICC (University of Florence)
Which functionalities ? Abandoned / removed object    Detection of changes
Which functionalities ? Analyse flow (crowd, vehicles) Counting, speed estimation  Detect abnormal behaviour At individual level  (eg loitering,  jumping over turnstiles, U-turns, …) At group level  (eg fighting) At  crowd level (crush)
Which functionalities ? Recognise : Vehicles Number Plates People
Which functionalities ? Other (not exhaustive) Detect people crossing / jumping / going / falling on rails Surveillance of tube doors closing (in station doors, wagon door) General railways protection (cars blocked on a crossing) Tunnel entrance or exit surveillance Analysis of behavior inside wagon / car (multi-sensor capacities) Smoke / Fire detection Traffic Management Luggage belt surveillance
From research to industry Existing gap between commercial systems and academic research :  Commercial systems  :  “simple functions”, dedicated to application and mono-sensor.  Main processing tasks : intrusion detection motion detection  detection of packages  Plate identification Academic research  : generating more accurate and robust algorithms in : object detection, recognition  and tracking,  human activity recognition ,  Scene automatic interpretation from video activity Multi-sensor fusion database exploration,  Main issues for technological transfer : Difficulty to accurately and exhaustively qualify performance of algorithms  HW Processing / communication performance Signal degradation : Frame rate Compression
Thales/CEA Vision Lab Mission : Provide Thales “Business lines” with video analysis functionalities  Efficient  Innovative With proved and mastered performances Activities (1/2) 1- Collect / develop a set of building blocks: State of the art video analysis algorithms Developed by  CEA/LIST  Completed  with technologies  provided by : Thales technical  teams  (“Vision cluster”) Selected Partners (academic labs / SMEs)
Thales/CEA Vision Lab Activities (2/2): 2- Setup an environment and a methodology to develop and test processing chains  Heterogeneous network of sensors > 30 cameras, indoor and outdoor, still or PTZ. SW platform Plug-in architecture Complete processing chain from sensors to HMI HW platform High Computing power:  150 physical bi-proc servers,    4-core Xeon 2,5 GHz, 32 GB Ram for each server High Storage capacity ( 10 TB) Wide image  database (Standard Corpus , Video provided by customers) Testing and evaluation tools and procedures  Use of image simulation technologies to : create scenarios difficult to encounter in real world (but which define the alarms) Fully controlled scene  and image characteristics
Main actors of the domain Commercial systems and components 3VR SECURITY, ACIC, AEXEON ANALYTIX, AGENT VI, AIMETIS, ARALIA, AXIS, BLUE EYE VIDEO, BOSCH, CERNIUM, CISCO, CITILOG, DYNAPEL, DVTEL, EADS, EMITALL, EVITECH, FOXSTREAM, GE SECURITY, GENETEC, GEUTEBRUCK, GEOVISION, HONEYWELL, IBM-ISS, INDIGOVISION, INTELLIO, INTELLIVIEW, INTELLIVISION, IOIMAGE, IOMNISCIENT, IPSOTEK, IVISIOTECH, KAOLAB, KEENEO, LENEL, MAGAL, MANGO DSP, MARCH, MATE, MILESTONE SYSTEMS, MIOVISION, MOSAIC TECHNOLOGY, TRAFFICON, NEXVISION, NICE VISION, OBJECTVIDEO, OPAX, ONSSI, PIVOTAL VISION, PRAETORIAN, PYRAMID VISION – SARNOFF, SAGEM, SECUSCAN, SIEMENS, SONY, SPIKENET, THALES, VERINT, VIDEALERT, VIDEOMINING, VIDIENT, VIGILANT TECHNOLOGY, VISIOWAVE, VIVOTEK ….. This list is not exhaustive !!! Collaborative programs (EU) CROMATICA, PRISMATICA, ADVISOR, CARETAKER,  VIEWS, … Main research teams Partners of previous programs

Thales

  • 1.
    Technologies d’Analyse Vidéodans le domaine de la sécurité Séminaire MEITO - Analyse vidéo et reconnaissance d'image 01 Avril 2010
  • 2.
    Jean-Yves Dufour Responsabledu laboratoire Thales / CEA-LIST « Vision Lab » Thales Services / ThereSIS Campus Polytechnique, 1, avenue Augustin Fresnel 91767 Palaiseau Cedex France Tél : 01 69 41 59 64 Mail : [email_address]
  • 3.
    Plan de laprésentation Video Surveillance : généralités Principaux domaines d’application Video Surveillance Intelligente Quelles fonctionnalités ? Transfert recherche / industrie Labo Vision de Thales / CEA
  • 4.
    What is VideoSurveillance ? Surveillance of an infrastructure (store, parking lot, airport, …) through images provided by video cameras observing “strategic” areas. two key purposes : Recording evidence for investigative purposes Providing a human operator with images enabling him to analyse the situation generally : to detect and to react Huge expansion of Video-Surveillance : Due to early 2000 events :governments have made personal and asset security a priority in their policies.  Deployment of large CCTV systems. Example, London Underground and Heathrow Airport have more than 5000 cameras each (2004).  Major research domain and commercial market or computer vision : Many collaborative funded projects, Many scientific international workshops / seminars / papers … Hundreds of companies created
  • 5.
    Main application domainsMainly oriented towards security (citizens, assets) Transports Railway / underground transportation (stations, tracks , tunnels, wagon) Airports, Urban and motorway road networks, Maritime transportation Large urban security (roads, streets, public place, …) Public events : sports, leisure, pilgrimage, … Industrial environments factories, pipe-line, energy power plants, … Border surveillance Military applications (remote surveillance, protection of installations)
  • 6.
    Smart Video SurveillanceIssues : Huge amount of image data (thousands of sensors in large systems) Human limitations : lack of attention span: Human operator cannot maintain an acceptable level of attention for long time Sifting through large collections of surveillance videos is tedious and error prone for a human investigator.  Manual surveillance becomes impractical  Needs for “situational awareness”  “ Smart video surveillance” ? Assist people in surveillance task: the computers watch the video and produces Real-time alerts … without fully replacing them : People still have to deal with the (false) alarms Assist people for post-analysis tasks : Indexing video to ease research Providing tools to accelerate research
  • 7.
    Which functionalities ?Detect intrusion Presence of somebody in forbidden zone Needs to detect moving object in protected area Somebody enters a protected area through forbidden way Needs to spatially analyse trajectory of people (tracking)
  • 8.
    Which functionalities ?Loitering : Somebody stays too long time in an “sensible” area Counting (people, vehicles, lorries, …) : how many objects pass through a given area  Detection of moving objects  Classification (people / vehicle or track / lorry, …)  Tracking to ensure that same person is not counted more than once) Source : Thales Italy and MICC (University of Florence)
  • 9.
    Which functionalities ?Abandoned / removed object  Detection of changes
  • 10.
    Which functionalities ?Analyse flow (crowd, vehicles) Counting, speed estimation Detect abnormal behaviour At individual level (eg loitering, jumping over turnstiles, U-turns, …) At group level (eg fighting) At crowd level (crush)
  • 11.
    Which functionalities ?Recognise : Vehicles Number Plates People
  • 12.
    Which functionalities ?Other (not exhaustive) Detect people crossing / jumping / going / falling on rails Surveillance of tube doors closing (in station doors, wagon door) General railways protection (cars blocked on a crossing) Tunnel entrance or exit surveillance Analysis of behavior inside wagon / car (multi-sensor capacities) Smoke / Fire detection Traffic Management Luggage belt surveillance
  • 13.
    From research toindustry Existing gap between commercial systems and academic research : Commercial systems : “simple functions”, dedicated to application and mono-sensor. Main processing tasks : intrusion detection motion detection detection of packages Plate identification Academic research : generating more accurate and robust algorithms in : object detection, recognition and tracking, human activity recognition , Scene automatic interpretation from video activity Multi-sensor fusion database exploration, Main issues for technological transfer : Difficulty to accurately and exhaustively qualify performance of algorithms HW Processing / communication performance Signal degradation : Frame rate Compression
  • 14.
    Thales/CEA Vision LabMission : Provide Thales “Business lines” with video analysis functionalities Efficient Innovative With proved and mastered performances Activities (1/2) 1- Collect / develop a set of building blocks: State of the art video analysis algorithms Developed by CEA/LIST Completed with technologies provided by : Thales technical teams (“Vision cluster”) Selected Partners (academic labs / SMEs)
  • 15.
    Thales/CEA Vision LabActivities (2/2): 2- Setup an environment and a methodology to develop and test processing chains Heterogeneous network of sensors > 30 cameras, indoor and outdoor, still or PTZ. SW platform Plug-in architecture Complete processing chain from sensors to HMI HW platform High Computing power: 150 physical bi-proc servers, 4-core Xeon 2,5 GHz, 32 GB Ram for each server High Storage capacity ( 10 TB) Wide image database (Standard Corpus , Video provided by customers) Testing and evaluation tools and procedures Use of image simulation technologies to : create scenarios difficult to encounter in real world (but which define the alarms) Fully controlled scene and image characteristics
  • 16.
    Main actors ofthe domain Commercial systems and components 3VR SECURITY, ACIC, AEXEON ANALYTIX, AGENT VI, AIMETIS, ARALIA, AXIS, BLUE EYE VIDEO, BOSCH, CERNIUM, CISCO, CITILOG, DYNAPEL, DVTEL, EADS, EMITALL, EVITECH, FOXSTREAM, GE SECURITY, GENETEC, GEUTEBRUCK, GEOVISION, HONEYWELL, IBM-ISS, INDIGOVISION, INTELLIO, INTELLIVIEW, INTELLIVISION, IOIMAGE, IOMNISCIENT, IPSOTEK, IVISIOTECH, KAOLAB, KEENEO, LENEL, MAGAL, MANGO DSP, MARCH, MATE, MILESTONE SYSTEMS, MIOVISION, MOSAIC TECHNOLOGY, TRAFFICON, NEXVISION, NICE VISION, OBJECTVIDEO, OPAX, ONSSI, PIVOTAL VISION, PRAETORIAN, PYRAMID VISION – SARNOFF, SAGEM, SECUSCAN, SIEMENS, SONY, SPIKENET, THALES, VERINT, VIDEALERT, VIDEOMINING, VIDIENT, VIGILANT TECHNOLOGY, VISIOWAVE, VIVOTEK ….. This list is not exhaustive !!! Collaborative programs (EU) CROMATICA, PRISMATICA, ADVISOR, CARETAKER, VIEWS, … Main research teams Partners of previous programs