Fouille et indexation de dOcuments compleXes et multimedia, Multimédia, Images, Indexation, REconnaissance
The Fox-Miire team is working on two main themes. The first one consist in human behavior retrieval from video streams. New motion descriptors extending current descriptors (like Lukas-Kanade) are studied to detect and track human behaviors. Semantic characterization of the multimedia contents is also studied by using approaches based on the bag of words technique. Four large fields of validation are considered: the gaze, the event, the emotion, and people recognition. These studies are performed under two different scales: the person and the crowd. The second theme focuses on statistical 3D shape analysis and its applications in a wide range of fields such as: (1) segmentation, matching, and retrieval of 3D data, (2) face and facial expression recognition from data acquired with 3D scanning devices, and (3) 3D-data acquisition. The research group is developing new mathematical and algorithmic solutions by combining several domains of expertise such as geometry, computational topology, statistics, machine learning, pattern recognition, computer graphics, and computer vision.