Transcript Video Understanding for Activity Monitoring

Video Understanding for Activity Monitoring
François Brémond – PULSAR – INRIA – Sophia Antipolis
[email protected]
Keywords: Scene understanding, event recognition, behaviour analysis, multi-sensor fusion, real-time systems, system evaluation,
adaptable system
Video Understanding Approach
Global approach integrating all video understanding
functionalities
Objective: Real-time Interpretation of videos from pixels to events
Segmentation
Classification
Tracking
Scenario Recognition
* Alarms
* Access to
forbidden area
Metadata
 Cognitive vision: 4D analysis (3D + temporal analysis)
 Artificial intelligence: explicit knowledge
 Predefined scenario models
 Context of the 3D environment, camera calibration
 Model of expected moving objects
 Software engineering:
 Easy generation of dedicated systems
 Reusable & adaptable platform
Extract and structure knowledge (invariants & models) for
 Perception for video understanding
 People detection (posture and gesture recognition) using background subtraction
and feature computation (e.g. Haar, HOG)
 Maintenance of the 3D coherency throughout time
 Short/long term tracker, crowd tracker
 Sensor information fusion
 Event recognition
 Predefined models based on an ontology
 Learnt primitive and composite events
 Evaluation, control and learning
 Evaluation framework
 Adaptation to dynamic environment
 Offline clustering, data mining
3D scene model
Scenario models
(A priori Knowledge)
Activity Monitoring Results
Parked aircraft monitoring in Toulouse
Unloading Front Operation
Subway monitoring
Jumping over barrier
Control access in Paris subway
Disturbing people in a train scenario
Subway monitoring
Tailgating detection
Blocking an exit
Subway monitoring
Elderly monitoring
Bank agency monitoring in Paris for bank attack prevention
Fighting
Subway monitoring
Eating in the living room
Vandalism against
vending machine
Elderly monitoring
Crowd behavior
Opposite direction
in a demonstration
Fainting
Dissemination
François Brémond
http://www-sop.inria.fr/pulsar/personnel/Francois.Bremond
75 Papers in international conferences and journals on
video understanding available at the web site
12 PhD thesis
Keeneo
Start-up (intelligent video surveillance) created in 2005.
http://www.keeneo.com
25 projects in safety/security:
• 7 European projects: Esprit, ITEA, FP6, FP7:
PASSWORDS, ADVISOR, AVITRACK, SERKET, CARETAKER,
CANTATA, COFRIEND
• 1 DARPA: VSAM
• 10 French projects: ANR, DGE, Prédit, TechnoVision,
PACA, CG06
• 7 industrial research contracts: Bull, Vigitec, SNCF,
RATP, ALSTOM, STMicroElectronics, Thales