Object Modeling from a Casual Collection of Images

Object modeling from a casual collection of images has found a range of applications including product advertisement over the web and visualization of museum objects. We have first analysed the current techniques in the context of a three-dimensional complexity space: Image collection cost, data storage requirements, and information processing complexity. It is clear that none of them is optimal in that space. Our goal is to develop an object modeling system that is flexible, robust and affordable. A preliminary version of our prototype system is presented in this paper. We discuss components of our system such as flexible camera calibration and robust surface extraction, and present initial couraging results.