On IO-efficient viewshed algorithms and their accuracy

• Authors: Herman Haverkort;Laura Toma;Bob PoFang Wei
• Venue: Proceedings of the 21st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
• Year: 2013

Given a terrain T and a point v, the viewshed or visibility map of v is the set of points in T that are visible from v. To decide whether a point p is visible one needs to interpolate the elevation of the terrain along the line-of-sight (LOS) vp. Existing viewshed algorithms differ widely in which and how many points they chose to interpolate, how many lines-of-sight they consider, and how they interpolate the terrain. These choices crucially affect the running time and accuracy of the algorithms. In this paper our goal was to obtain an IO-efficient algorithm that computes the viewshed on a grid terrain with as much accuracy as possible given the resolution of the data. We describe two algorithms which are based on computing and merging horizons, and we prove that the complexity of horizons on a grid of n points is O(n), improving on the general O(nα(n)) bound on triangulated terrains. Our finding is that, in practice, horizons on grids are significantly smaller than their theoretical worst case bound, which makes horizon-based approaches very fast. To measure the differences between viewsheds computed with various algorithms we implement an error metric that averages differences over a large number of viewsheds computed from a set of viewpoints with topological significance, like valleys and ridges. Using this metric we compare our current approach, Van Kreveld's model used in our previous work [7], the algorithm of Ferreira et al. [6], and the viewshed module r.los in the open source GIS GRASS.