Recovery of 3D Shape from Images
The goal of this project is to create a comprehensive C++ library for recovering 3D information from images. The library is intended to be a toolbox for a variety of different applications. One example of the use of 3D data is to derive information from real scenes for Image Based Rendering. Some of the work in this project has investigated how to synthetically change the lighting effects in a photograph, giving rise to changes in the appearance of objects and new shadows. This is of value for manipulating snapshots whose appearance is unsatisfactory, which is an important area given the replacement of film-based cameras with digital cameras in the home consumer market over the next few years. It is also of use in a professional setting for the design and advertising markets.
Background & Objective: Recovery of 3D shape is not a single algorithm, rather it is a battery of different techniques. At one end of the spectrum are sparse feature approaches detecting points, straight line segments or curves and recovering their 3D coordinates, at the other end of the spectrum are optical flow approaches which can be used to generate dense range maps, while further techniques include shape-from-focus, shape-from-texture, and shape-from-shading. The objective of the work is to provide a toolbox of components related to 3D shape recovery, for use in a variety of applications.
Technical Discussion: The basic functionality of the library is for 3D reconstruction from an uncalibrated image sequence. Because the system works with uncalibrated image data, there is no need for prior information about the camera intrinsic parameters or camera motion. Thus camera calibration, and the requirement for controlled motions of the camera when taking images, are avoided. The sparse feature reconstructions have been used for synthetic relighting of images, and are being investigated for the construction of a 3D model of an urban environments. Newer components in the library deal with the creation of volumetric descriptions for an object (3D occupancy grids), using a calibrated system and many views around the object.
Technology Areas:
Computer Vision
Graphics
Modification Date: June 13, 2008