Curve-Based Multiview Pose Estimation and Scene Reconstruction
Multiview geometry for absolute and relative pose estimation and scene reconstruction has largely been based on RANSAC-based selection of point-based feature correspondences leading to a scene representation as an unorganized cloud of 3D points. However, in many scenes such point correspondences are not present in sufficient quantity or do not lead to distinguishable correspondences. In such cases, however, the same scenes frequently depict image curve structure which can be used to build a 3D curve drawing which can act as a scaffold on which surfaces can be reconstructed. The talk presents a framework whereby edge/curve structure can be used as the basic primitive for scene reconstruction as well as pose estimation.
Benjamin Kimia received the B. Eng., M. Eng. and Ph.D. degrees from McGill University, Montreal, Canada. He has been a faculty member at Brown University since 1990 in the Electrical and Computer Engineering area of School of Engineering. His work has broadly centered on the mathematical, computational, psychophysical and neurophysiological aspects of understanding human and computer vision. His early work focused on understanding shape for object recognition and later evolved to integrating a coherent notion of shape in a variety of computer vision applications. In collaboration with colleagues at Brown University he pioneered and developed REVEAL, a system for use in the field by archaeologists. He has also held a steady interest and engaged in medical imaging problems throughout his research. He is most recently interested in multiview geometry problems, assistance devices for the navigation of the blind, indexing into large databases by proximity graphs, automated animal behavior analysis, and image-guided ablation of tumors.