Section: Contracts and Grants with Industry

DGA: CrowdChecker (ENS and E-vitech)

Participants : Jean Ponce, Josef Sivic, Ivan Laptev.

CrowdChecker (DGA) is a joint DGA project with industrial partner E-vitech. This contract belongs to our video understanding research program. It aims at real-time characterization of a crowd seen from a camera mounted 3 to 10 meters over the ground. It includes segmentation of the crowd, clustering by movement, detection of abnormal behaviors (persons, for instance, crossing the crowd flow, or having unusual speed), tracking people. Several parts of computer vision and machine learning are involved: crowd optical flow estimation, image processing, crowd feature extraction, statistical learning from video database, etc.