POPPerception On Purpose
1. Institut National de Recherche en Informatique et Automatique (INRIA), France
Two research groups are involved in the POP project — the computer vision group (PERCEPTION) and the statistics group (MISTIS). The computer vision group will provide the scientific and technological expertise needed to integrate various vision modules into a working system, i.e., everything from camera acquisition and camera calibration software, to methods for recovering depth using stereo, motion segmentation methods, and object tracking methods. The statistics group will provide theoretical insights for the development of robust statistical methods. These methods will be used for the 3-D reconstruction and tracking algorithms that must operate in the presence of noisy, bad, or partially missing data.
The expertise on real-time 3-D reconstruction of humans from image contours will be used within the context of the active stereo sensor to be developed by the POP partners. The immense expertise on camera calibration using generic camera models will be used for the calibration of the ative stereo head.
Both the statistics and the vision groups will contribute to the development of a theoretical and methodological model of perceptual attention, in particular they will investigate together the most robust and efficient way to model interactions between high-level knowledge and low-level sensorial information. The statistics group is particularly competent in the domain of Bayesian decision theory and therefore will have a major contribution to model the integration of auditory and visual cues.
Our experimental and development platform/laboratory will be made available to POP researchers. One in-house development engineer (Hervé Mathieu, member of the vision group) will supervise the software development activities of POP in order to integrate and build the POP’s experimental platform.