TO FINISH

Contents (hide)

  1. 1. Stages
    1. 1.1 Viewshed Computation
    2. 1.2 Visibility Index Computation
    3. 1.3 Multi-observer Siting
  2. 2. Presentations
    1. 2.1 UniGIS
    2. 2.2 One summary slide
  3. 3. Selected Publications
    1. 3.1 SDH 2006
    2. 3.2 GIScience 2004
    3. 3.3 ISPRS 2004
    4. 3.4 SDH 2002
    5. 3.5 `SDH 1994
    6. 3.6 GIScience 2000 Savannah
  4. 4. Earlier Web Page
  5. 5. Software

1.  Stages

This theme, which is producing new algorithms and implementations relating to terrain visibility, comprises several stages.

1.1  Viewshed Computation

Given

  1. some terrain, i.e, elevation data, expressed as a matrix,
  2. an observer sitated at a given height above the terrain,
  3. a radius of interest out to which the observer can see, and
  4. potential targets at any other point of the terrain, each situated at some fixed height above the local terrain,

we compute the observer's viewshed, or the region visible from the observer. More precisely, it is the projection onto the xy plane of the targets that are visible. A novel aspect of our work is that we efficiently compute high-resolution viewsheds, not, say, 64x64 approximations.

1.2  Visibility Index Computation

The visibility index of a point in the terrain (for given observer and target heights and radius of interest) is the fraction of the circle out to the radius of interest that is visible to the observer at its center. We use a Monte Carlo approximation to determine the visibility index of every point in the terrain.

This work has produced some counterintuitive results. For example, for some terrain, there is no significant correlation between elevation and visibility index; on the average, higher is no better for observation.

1.3  Multi-observer Siting

The object of this stage is to find a set of observer sites which, together, can cover the whole area of interest, or as much as possible.

2.  Presentations

2.1  UniGIS

UniGIS, Vancouver, May 2006. This summarizes RPI's role in the GeoStar project. Read it first.

2.2  One summary slide

One summary powerpoint slide from May 2006 is: here.

3.  Selected Publications

Here are some of the more currently relevant of my siting publications. The others are listed on my main website. Read some of these next.

3.1  SDH 2006

. W. Randolph Franklin and Christian Vogt. In Progress in Spatial Data Handling: 12th International Symposium on Spatial Data Handling, pages 845-861.Springer, , 2006. ISBN 978-3-540-35588-5. (paper, talk).

Paper, Talk

3.2  GIScience 2004

WR Franklin & C Vogt Efficient Multiple Observer Siting on Large Terrain Cells, GIScience 2004 Third International Conference on Geographic Information Science University of Maryland Conference Center October 20-23, 2004.

3.3  ISPRS 2004

WR Franklin & C Vogt Multiple Observer Siting on Terrain with Intervisibility or Lo-Res Data, XXth Congress, International Society for Photogrammetry and Remote Sensing, Istanbul, 12-23 July 2004.

3.4  SDH 2002

WR Franklin, Siting Observers on Terrain, Symposium on Spatial Data Handling, Ottawa, July 2002. Once upon a time, the conference had a website at http://www.geomatics2002.org/index_e.html .

This paper presents an experimental study of a new algorithm that synthesizes separate programs, for fast viewshed, and for fast approximate visibility index determination, into a working testbed for siting multiple observers jointly to cover terrain from a full level-1 DEM, and to do it so quickly that multiple experiments are easily possible. Both the observer and target may be at a given fixed distance above the terrain. The process operates as follows. (1) An approximate visibility index is calculated for each point in the cell under consideration. (2) A set of tentative observers is selected from the highly visible points. (3) The actual observers are selected from them, so as to cover as much of the cell of possibe, using a greedy algorithm. Various experiments with varying parameters were performed on the Lake Champlain West cell, with observations such as the following. (1) Forcing tentative observers to be well spaced was more important than using the most visible tentative observers. (2) Most of the new observers added (because they covered the most unseen points) were already visible to an existing observer. (3) Randomly deleting many tentative observers before final selection didn't reduce the final area covered.
  • paper (this is an expanded colorized version of what was published).
  • talk

3.5  `SDH 1994

WR Franklin and C Ray. Higher isn't necessarily better: visibility algorithms and experiments. In Thomas C. Waugh and Richard G. Healey, editors, Advances in GIS Research: Sixth International Symposium on Spatial Data Handling, pages 751-770, Edinburgh, 5-9 Sept 1994. Taylor & Francis.

We describe several fast programs to compute viewsheds and weighted visibility indices for observation points in a raster terrain. These programs explore various tradeoffs between speed and accuracy. We have analyzed many cells of data; there is no strong correlation between a point's elevation and its weighted visibility index. However, the, very few, high visibility points tend to characterize features of the terrain. This work can form a basis for automatically locating observers jointly to cover a terrain region of interest.

3.6  GIScience 2000 Savannah

'*[wrf-appvis-00]*' WR Franklin, [[approxvis/index.html| <span class=title>Approximating visibility</span>]], '~GIScience 2000~' plenary talk, 30 October 2000, Savannah, Georgia.

Abstract: Determining the visibility indices of points in a terrain is an important application in cartography, but is very compute-intensive. This paper studies the relative importance of various factors in visibility computation, with a view to making the process more efficient. Preliminary results include the following. Scaling down elevations to 8-bits precision doesn't change the general visibility indices, but can cause artifacts, where the visibility index changes in steps. The height of the observers and targets above the terrain have little effect on the result. Changing the radius of interest completely changes the result. A larger ROI leads to finer details in the visibility index map. It also leads to a smaller average visibility index. In every test case but one the visibility indices appear Poisson distributed. We approximate the visibility index by running various numbers of lines of sight from each possible observer to many random targets. Ten targets per observer is insufficient. 30 targets per point appears to lead to a visibility map that resembles many more points. Nevertheless, there is still a visible improvement on increasing from 100 to 300.

Our goal is to use these properties in a visibility index program that will use industrial engineering sampling techniques to estimate the most visible points with the minimum number of observer-target tests.

4.  Earlier Web Page

Here is an earlier page on this topic: http://wrfranklin.org/Research/siting/ I am gradually moving things from that page to this. Read it next if you want even more info.

5.  Software

A testbed consisting of a sequence of interacting programs has been produced. To learn about it, read here:

  • sections 1 and 2 of the SDH2002 paper, linked above.
  • in point form, in the SDH2002 talk

If those are not satisfactory, then read here:

The source code, makefiles and sample data are here: http://wrfranklin.org/wiki/Main/siting/site.tgz. My development environment is linux and c++.

The source code, plus executables and sample computed viewsheds and images are in this 126MB tarball: http://wrfranklin.org/wiki/Main/siting/site-big.tgz.


<< Alternate Terrain Reps | Research Page List | Gridding Contours >>