Abstract
This work details the setup of a 3D vision component, itself par of a larger framework developped at the LIRMM and dedicated to motion compensation on the beating heart. Our goal was to track the motion of the heart at real-time and evaluate the 3D position of the surface of the organ. These informations are obtained from video data from one or two cameras. A first contribution is a method to choose texture-based descriptors to robustly characterize the surface of the organ and track it. The other main contribution is a tracking method, based on both these descriptors and pattern matching, that showed good results on in-vivo experiments. This method is applied on real-life images to compute 2D and 3D motion, and its results are compared with the ones of other existing approaches.