Geography Reference
In-Depth Information
5.2 Spatial Clustering
We used a subset of photos referring to the territory of Switzerland. 1 For dis-
cretizing the space we use a method (Andrienko and Andrienko 2011 ) that divides
the territory to non-overlapping polygons of given size in a way that reflects the
distribution of points. In brief, the generalization method groups points into spatial
clusters and uses the centroids of the clusters as generating points for Voronoi
tessellation (Okabe et al. 2000 ) of the territory. We applied spatial clustering to the
positions of the photos and built Voronoi cells (1,183 in total) with average
diameter of 2 km around the obtained clusters. The whole operation took between
3 and 5 s using sampling approach with about 20,000 points. The general steps of
the algorithm are described below:
Algorithm 1: Territory tessellation
Given: Sequence of positions of points P = {xi, yi} and desired radius r
Output: A set of Voronoi cells V
Description of the algorithm:
1. Group the points of P in spatial clusters with desired radius r
S = SpatialClusters(P; r)
2. Compute the centroids of the spatial clusters C = Centroids(S);
3. Generate Voronoi cells around the centroids
V = Voronoi Tessellation(C);
5.3 Time Series Analysis
For every cluster, we calculated frequencies of people's visits and the number of taken
photos aggregated by month and built a time series graph spanning 5 years (2005-
2009 inclusive). Figures 1 and 2 show a part of Switzerland with examples of different
temporal patterns of events for selected regions denoted as A, B, C, D, E. Figure 1
shows frequencies of people's visits while Fig. 2 displays frequencies of the taken
photos.
According to Fig. 1 , people visit the region labeled A in all seasons. In total, 71
people visited this region and took 1,721 photos. We can observe a sharp increase in
the number of photos taken in every year in January. A possible explanation is that
1 We do not present results of a complete analysis of the photos on the territory of Switzerland
but only provide several examples as illustrations of what can be detected. It is clear that many
more events occurred, and the challenge is to develop such methods that will find as many of
them as the available data permit.
Search WWH ::




Custom Search