Given a digital elevation model (DEM) of a terrain and an arbitrary location, one question often asked is: "If I were to be at this location, what would I be able to see?" This is known as the viewshed of that viewpoint. Viewshed calculation is one of the fundamental computations on DEMs and is a vital component to many fields including telecommunication, strategic navigation and land use planning.
There exist algorithms for computing the viewshed of a point "accurately". However, these algorithms are computationally intensive and often not feasable for practical use, espacially on large DEMs and in cases where repeated viewshed computations are required. Many researchers have proposed algorithms to dramatically reduce computation time while sacrificing accuracy. While attention was given to analyzing the gain in computation time, results comparing accuracy of the new algorithms are lacking.
In this project we developed a method for comparing the accuracy of viewshed algorithms. Given a test algorithm, a reference algorithm, and a DEM, we pick a sample of points in the terrain and compute viewsheds using the two algorithms. We compute an overall accuracy score by keeping track of the number of "false visible" and "false invisible" points in the viewshed generated by the test algorithm compared to the one generated by the reference algorithm, over all sample viewpoints. The sample viewshed points are picked so that they represent a collection of topographically important features of the terrain, as these are points where it is important for the viewshed to be accurate. To this end, we select the channels and ridges in the terrain, where small and large viewsheds are expected, respectively.
We have implemented some of the commonly used viewshed algorithms (precise and approximate) and analyzed their accuracy according to the metric above.
|A terrain DEM||A sample of the ridges and channels|