Rapid modeling of cones and cylinders from a single calibrated image using minimum 2D control points

摘要

Three-dimensional modeling from a single image is a very useful technique in applications, such as augment reality, architecture, urban digitalization, heritage preservation, etc. While the modeling of lines and planes has been well studied, a convenient way of modeling cones and cylinders, which are commonly encountered objects, is not available yet. Although fully automatic approaches exist, they cannot guarantee the precision demanded by a real application, and are not appropriate for modeling a complex environment, where users need full control of the modeling process. In addition, approaches relying on the detection of full object contours cannot be directly transformed to simple interactive modeling procedures. This paper presents novel approaches to modeling cone and cylinder objects based on a set of minimum two-dimensional control points. The approaches enable intuitive, fast and accurate modeling from a single image. Both synthetic and real data experiments demonstrate the advantages of the proposed approach: desired accuracy with minimal user interactions.