Spatio-temporal Event Processing with StreamInsight, SQL Server Denali, and Bing Maps – Part 5

Yesterday, fellow Bing Maps MVP Nicolas Boonaert (blog) kindly tweeted a link to my series of posts about analysing spatio-temporal data using StreamInsight and Bing Maps. In addition to the generous words he said publicly, he also sent me a private message teasing me about the graphical presentation of the data. I believe the term he used was “moving pixelised potato on a map”…

Spurred on by Nicolas’ comments, I thought I’d take the opportunity to improve the visual impact of my StreamInsight output by making use of my own HeatMap library to plot the outbreak as a heatmap rather than as a convex hull. To do so required remarkably few changes to the code:

First, instead of creating a (Polygonal) convex hull around all the events in the Stream Insight window at a  given point in time, I added a UDA that would instead create a (MultiPoint) union of all those points.

public class MultiPointUnion : CepAggregate<string, string>
  {
    public override string GenerateOutput(IEnumerable<string> eventData)
    {
      var gb = new SqlGeographyBuilder();
      gb.SetSrid(4326);
      gb.BeginGeography(OpenGisGeographyType.MultiPoint);
      foreach (var d in eventData)
      {
        SqlGeography point = SqlGeography.Parse(d);
        gb.BeginGeography(OpenGisGeographyType.Point);
        gb.BeginFigure((double)point.Lat, (double)point.Long);
        gb.EndFigure();
        gb.EndGeography();
      }
      gb.EndGeography();

      return gb.ConstructedGeography();
    }
  }
}

Then, I amended the javascript that updated the map in the browser control. Instead of plotting a single polygon as a vector shape on the map, I imported the heatmap library and added a new HeatMapLayer as follows:

heatmapLayer = new HeatMapLayer(
  map,
  new Array(),
  { intensity: 0.4,
    radius: 25,
    colourgradient: {
      0.0: 'rgba(255, 255, 255, 0)',
      0.5: 'rgba(255, 255, 120, 100)',
      0.8: 'yellow',
      0.95: 'red',
      1.0: 'white'
    }
  }
);

(Colours were chosen somewhat arbitrarily – white hot spot right at the centre of each point, followed by red, and gradually diminishing yellow). Then, each time the WCF listener received a new event summary from StreamInsight, rather than plot it as a Microsoft.Maps.Polygon, it called a method that would split the WKT of the MultiPoint instance into an array of Microsoft.Maps.Locations, and supply these to the heatmap layer:

heatmapLayer.SetPoints(locationArray);

The result is shown in the video below, complete with my rather vague commentary (based on this evidence, I don’t think I’ll ever realise my childhood dreams of being a radio DJ… ). One thing to notice is that, while the convex hull approach I used before emphasises the overall geographic spread at any point in time, the heat map approach emphasises instead the intensity of events in certain areas.