Section: Contracts and Grants with Industry

Cifre grant with Thomson-R&D on human pose estimation and action recognition

Participants : Ivan Laptev, Patrick Pérez.

duration 36 months

This contract started in December 2008 is associated with the supervision of R. Kumar's thesis funded by a Cifre grant. It addresses the problem of human pose estimation and human action recognition in still images. The work follows a novel data-driven approach aiming at exploiting the space of human images from many examples. Given many (video) samples of people spanning the “person-image” space, the goal is to explore the structure of this space and use it to pose constraints on the interpretation of new images. Estimation of human pose is useful e.g. in the context of graphics and video editing but also as a cue for interpretation of human actions. Human action recognition in still images is required for the search, indexing and automatic annotation of large image and video collections.