![]() The map fragment portrays the central part of Berlin. When a density map is built, every photo gets its normalized color value. Thus, every photo gets a weight based on the proximity of other photos around. For every photo in a cluster, the cumulative sum of the influences of the other photos is calculated using the Gaussian. We propose to use the notion of influence func- tion introduced in. A drawback is that the informa- tion about the locations of the individual photos is lost. A frequently used approach to visualizing densities on a map is color coding of the counts of points inside areas, as in and in Figure 1 (back- ground shading). (2) The resulting clusters are not restricted to a specific shape or size and properly reflect the natural distribution of the data. The benefits are the following: (1) regions with density below some threshold are counted as noise and wont be reflected in visualization. We propose to use density-based clustering algorithms (DBCA). Appropriate clustering methods are needed to find the natural areas of high concentration without imposing any artificial division of the territory. Therefore, the resulting map may inaccurately show the places of concentration. A disadvantage of the grid- based approach described in is that grids dont reflect the natural data distribution. Density map is a good technique to visualize concentrations of events. ![]() For instance, a peculiarity can be noted in the area of Munich (marked in the figure): the number of photographers is very high from August till October, which may be related to the famous Oktoberfest (beer festival). The com- mon and distinct features of the seasonal variations in different places are easily seen. ![]() The color filling serves to improve the visibility of the diagrams and the differentiation between the positive and negative values (yellow and blue, respectively). The vertical dimension represents the de- viation from the local mean value. The horizontal axis of a diagram represents 12 months from January to De- cember. The counts resulting from the aggregation have been transformed into the differences from the local mean values normalized by the standard deviations. The diagrams represent the yearly variation of the number of photographers. Dy- namic filtering has been applied so that only the cells visited by at least 250 photographers are visi- ble. The shading of the cells portrays the total number of different people who made their photos in these cells. The map fragment shows the south of Germany. The data have been aggregated spatially by grid cells and temporally by months, irrespective of the years. Figure 1 gives an example of how aggregation results can be visualized and explored. The results of the aggregation are time series of counts related to the spatial compartments (number of photos, number of different people) and, possibly, other statistics such as the mean number of photos per person. For the temporal aggregation, the time may be treated as linear or as cyclic and, respectively, di- vided into consecutive intervals or into intervals of the daily, weekly, or yearly cycle. ![]() For the spatial aggregation, the territory is divided into suit- able compartments, for example, using a regular grid. Spatial and temporal aggregation of movement data viewed as independent events can be done by means of database queries. Aggregation helps an analyst to cope with large amounts of data. The methods we present do not take into account the contents of the photos but only their spatial and temporal references. The example dataset we use for the presentation consists of about 590 000 photos made on the territory of Germany during the period from Janutill March 30, 2009. The tasks and methods are summarized in Table 1. For the defined tasks, we aim at developing appropriate methods. We distinguish space-centered tasks, where the data are used to study the properties of the space and places, from agent-centered tasks targeting at the properties and behaviors of the people (in general, moving agents). We define possible types of analysis tasks related to the views of the data as events and as trajectories. take a systematic approach to the analysis of event-based movement data.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |