Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD thesis, MIT, August 1998. 75 76 BIBLIOGRAPHY  Home/Search   Document Details and Download   Summary   Related Articles   Check

....used in this pipeline that I will refer to hereafter. In the following chapters, I will then focus on texture estimation and relief estimation for a facade building and describe new algorithms that reinforce the realism of a 3D urban model. 8 1.2. 1 The Pipeline As shown in Figure 1 1 after [Coo98] the automatic 3D pipeline consists of several distinct steps: Acquisition, Pose Refinement (Spherical Mosaicing and Mosaic registration) and Model Extraction (Vertical Facade Extraction and Texture and Relief Estimation) Graphics Workstation Mosaic Registration Vertical Facade ....

....images in one node. A spherical mosaicing procedure can thus be performed, which maps the images onto a sphere surrounding the camera. Another optimization process refines the camera position across the entire set of nodes, using feature correspondences to solve the registration among the nodes ( Coo98] AT00] a) b) Figure 1 3: a) hemispherical tiling used for node acquisition; b) textured hemispherical mosaic. 1.2.4 Facade Extraction The next step of the pipeline consists of extracting a vertical facade through the whole set of images located at di#erent positions that see it. The ....

[Article contains additional citation context not shown here]

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD thesis, Dept. of Electrical Engineering and Computer Science, September 1998. 49

....[8] These methods use only 2 D images and require a user to guide the 3 D model creation phase. This leads to lack of scalability wrt the number of processed images of the scene or to the computation of simplified geometric descriptions of the scene which may not be geometrically correct. Teller [6, 18, 26, 5] uses an approach that acquires and processes a large amount of pose annotated spherical imagery of the scene. This imagery is registered and then stereo methods are used to recreate the geometry. Zisserman s group in Oxford [11] works towards the fully automatic construction of graphical models ....

S. R. Coorg. Pose Imagery and Automated Three-- Dimensional Modeling of Urban Environments.PhD thesis, MIT, September 1998.

....Computer Science. The broad goal of the project is to create a method for rapidly acquiring computer models of three dimensional (3 D) urban environments. This representation could be used to create a virtual environment for numerous applications including games, virtual tours, and urban planning [Coo98]. Research is currently underway to create a fully automated process which uses images from an urban setting to produce a 3 D CAD model of the scene. This model includes not only geometric information, such as the shapes, locations, and orientations of building facades, but also textural ....

....In order to combine the images accurately, estimates must be made of their position and of the internal parameters of the camera. The information recorded by Argus is accurate to within about one degree, however 3 D reconstruction algorithms require accuracy one to two orders of magnitude greater [Coo98]. To provide this level of precision we use the pose refinement algorithm which is the subject of this thesis. The spherical mosaicing algorithm makes use of the data contained within the images to align and orient them with greater accuracy. This is accomplished through an iterative global ....

[Article contains additional citation context not shown here]

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD Thesis, MIT, August 1998. (http://graphics.lcs.mit.edu/pub_theses.html)

....availability of the sensor output, which we call pose imagery, makes possible fundamentally novel and powerful approaches to the model reconstruction problem. The sensor and algorithms are described elsewhere [ Chou and Teller, 1997, Coorg and Teller, 1997, Mellor et al. 1997, DeCouto, 1998, Coorg et al. 1998, Coorg and Teller, 1998, Coorg, 1998 ] 1.1 Related Work Existing computer vision algorithms [ Horn, 1986, Faugeras, 1993 ] usually have one or more significant limitations. Often, they make the assumption that input imagery is supplied already controlled, or geo registered. Achieving ....

....call pose imagery, makes possible fundamentally novel and powerful approaches to the model reconstruction problem. The sensor and algorithms are described elsewhere [ Chou and Teller, 1997, Coorg and Teller, 1997, Mellor et al. 1997, DeCouto, 1998, Coorg et al. 1998, Coorg and Teller, 1998, Coorg, 1998 ] 1.1 Related Work Existing computer vision algorithms [ Horn, 1986, Faugeras, 1993 ] usually have one or more significant limitations. Often, they make the assumption that input imagery is supplied already controlled, or geo registered. Achieving geo registration in existing photogrammetry ....

[Article contains additional citation context not shown here]

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. Ph.D. Thesis, MIT, 1998.

....accurate estimates of acquisition time, camera position, and camera orientation. The availability of the sensor output, which we call pose imagery, makes possible fundamentally novel and powerful approaches to the model reconstruction problem. The sensor and algorithms are described elsewhere [2, 6, 16, 9, 5, 7, 4]. 1.1. Related Work Existing computer vision algorithms [13, 11] usually have one or more significant limitations. Often, they make the assumption that input imagery is supplied already controlled, or geo registered. Achieving geo registration in existing photogrammetry systems is a manual ....

....facades of buildings in the Technology Square office park using a histogramming technique on edge orientations. Moreover, we can produce high quality estimates of building texture for each facade, using a simple technique based on weighted median statistics. Both algorithms are described fully in [7, 4]. The third salient aspect, our use of hemi spherical images, yields primarily engineering, rather than theoretical, advantages. We acquire imagery by rotating (panning and tilting) a camera into about 50 discrete orientations around a fixed optical center, or node. Automated optimization ....

[Article contains additional citation context not shown here]

S. Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments (to appear). PhD thesis, MIT Ph.D. Thesis, 1998.

....pose from (approximate) initial estimates. This paper describes our method for refining estimates of camera orientation for images taken around a single optical center. Recovering relative translations and orientations between mosaics acquired at di#erent optical centers is addressed elsewhere [Coo98, ATar, ATon]. Related Work Of fundamental interest in mosaic computations is the warp relating a pair of overlapping images. The simplest method to compute this warp uses four point correspondences between the two images [Hec89, Fau93] Several algorithms (e.g. ZFD97] use this approach. However, ....

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD thesis, MIT, 1998.

No context found.

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD thesis, MIT, August 1998. 75 76 BIBLIOGRAPHY

No context found.

Satyan Coorg. Pose Imagery and Automated Three-Dimensional Modeling of Urban Environments. PhD thesis, MIT, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC