A Crash Course on Vision Based Modeling

Опубликовано 25 августа 2016, 20:57

The ubiquity of digital cameras and the ease of collecting video data opens the possibility of modeling 3D scenes and actions from images and video. In this course, we will provide an overview of the problems and techniques in Vision Based Modeling, an active area of research in computer vision. We will describe the principles and methods used for analyzing multiple images of a 3D scene taken from different view points ("stereo"), and for analyzing a sequence of images from video ("motion"). Specifically, over a course of four lectures, the following topics will be reviewed : (i) techniques for computing visual motion and stereo correspondence, (ii) the interpretation of visual motion and its applications, (iii) the geometric relationships between multiple images, (iv) techniques for calibrating the camera and recovering the camera pose and motion, (v) techniques for recovering the 3D scene structure based on 2D image correspondences,