3-D Modeling from Video

"Three-dimensional (3D) models of buildings, objects, and terrain are in high demand for a wide variety of military and commercial applications. Unfortunately, the availability of these models is limited and generating them takes considerable effort both intime and expense. Our goal on this project is to develop the additional techniques required to make such a capability both inexpensive and ubiquitous. The most successful recent approaches for recovering 3D geometric models from image sequences are basedon the use of the fundamental matrix and the trifocal tensor. These methods however rely on feature tracking algorithms that typically cannot track features over a wide range of viewing directions, limiting the accuracy of the recovered structure. Inaddition, the tracked features are generally sparse, resulting in a sparse 3D model inadequate for most modeling applications. The first issue will be addressed with a tracking algorithm that takes into account the local 3D structure around feature points.The second issue will be addressed by incorporating a second pass of the structure estimation to recover the complete structure using a stereo matching approach. The 3D structure will be represented using a 3D modeling and structure integration approachdeveloped at the Colorado School of Mines for stereo image sequences. The proposed approach will provide a solution to the remaining practical problems in the development of a 3D modeling