Department of Computer Engineering
S E M I N A R
Deformable Shape-from-Motion and Interventional Laparoscopy
Professor Adrien Bartoli
Advanced Laparoscopy and Computer Vision Research Group
Computer Vision deals with the general problem of reasoning about the physical world given visual data provided in images and videos. The data may be acquired by a broad variety of imaging sensors, ranging from consumer grade cameras to professional medical sensors. One of the fundamental goals in Computer Vision is to reconstruct the 3D shape of objects viewed by the image sensor. A wide variety of methods has been developed over the years. This presentation will focus on what is known as Shape-from-Motion. This aims to recover 3D shape using multiple images taken by a moving camera. Shape-from-Motion is unsolvable without some a priori shape knowledge. One assumption that has been used with great success is that many objects are rigid. This is widely applicable in for example man-made environments and has led to successful, stable and mature results both in terms of the theory of visual geometry and in terms of practical effective algorithms.
The world however is also made of non-rigid, deformable objects. Shape-from-Motion in this case is a far more challenging problem. Indeed, the state of research has been left fairly opened since deformable Shape-from-Motion was first attempted about a decade ago. The central topic of this presentation will be to present the specific difficulties when attempting deformable Shape-from-Motion and the theoretical and practical advances that have been made so far. The two problems of deformable image registration and 3D shape recovery from registered images will be discussed. The presentation comprises numerous experimental results, with an emphasis on advanced methods for surfaces of materials including paper and cloth. The application and extension of these techniques in the specific context of interventional laparoscopy will be presented and discussed.
Bio: Adrien Bartoli has been a Professor of Computer Science at Université d’Auvergne Clermont 1 since fall 2009. He does research in fundamental and medical computer vision and runs the ALCOV (Advanced Laparoscopy and Computer Vision) research group at ISIT jointly with Professor Michel Canis. Previously he was a permanent CNRS research scientist at LASMEA since 2004 and a Visiting Professor at DIKU Copenhagen between 2006-2009. He was a postdoctoral fellow at the University of Oxford between 2003-2004 and a PhD candidate at INRIA between 2000-2003. He received the 2004 Grenoble-INP PhD thesis prize, the best paper award at CORESA’07 and the 2008 CNRS Bronze Medal.
DATE: 25 April, 2011, Monday @ 10:40