Visual Scene Understanding français

CEA LIST develops visual scene understanding technologies able to

  • Perceive and model the environment
  • Detect, track and describe objects of interest
  • Re-identify people
  • Analyse people's behaviour and activity
  • Recognise observed events

Research projects

Oject detection

Real-time 2D/3D robust people detection

DeepMANTA: MANy TAsk Deep Learning for object detection, 3D localisation, accurate pose estimation in a single image

Railway sign detection with deep neural networks and boosting

Fine-grained object recognition

2D/3D object fine categorization using deep features

Deep metric learning and sparse representations for person re-identification

Scene modelling

3D obstacle detection in stereovision

3D scene segmentation and ground surface analysis in stereovision

Semantic image segmentation with deep neural networks

Object tracking

People visual tracking in camera networks with probabilistic models

Online multiple object tracking using sparse representations

Activity recognition and behaviour analysis

Unsupervised anomaly detection in a crowd

Violent event detection by weakly supervised learning of unstructured motion

Daily living activity detection and recognition

Daily Home Life Activity Dataset (coming soon)

Sensor calibration

Camera automatic calibration


  • Security
  • Advanced driver assistance systems
  • Smart environment
  • Advanced manufacturing


CEA LIST has a strong background in various vision sensors: colour cameras, infrared cameras, PTZ, fish-eye cameras, stereovision, 3D cameras, and camera networks.