[Python] How to merge many nearly parallel line segments into a single density-weighted...

I have a dataset of line segments derived from many GPS traces, where users have followed the same roads but their paths aren’t identical (think of thousands of cars driving along a road, each with slightly noisy GPS). I’ve already broken their traces into small line segments (using a 1/50 sampling rate), and now I want to aggregate these into a smaller set of representative segments with associated weights. The goal is to generate a road-like “heatmap” that can be rendered efficiently in a frontend (e.g. Mapbox) by drawing these weighted segments rather than drawing all possible lines and performing operations on the user device to render nice looking lines

My current segments look something like this in my code

segments = [
{'start_lat': 0.0, 'start_lon': 0.0, 'end_lat': 50.0, 'end_lon': 50.0, 'weight': 1},
{'start_lat': 0.1, 'start_lon': -0.2, 'end_lat': 49.8, 'end_lon': 50.3, 'weight': 1},
{'start_lat': -0.2, 'start_lon': 0.2, 'end_lat': 50.2, 'end_lon': 49.7, 'weight': 1},

]


I've tried a fairly naive comparison that checks if start and end points are within a distance threshold, then averages them, but am struggling to get that working and it seems particularly inefficient.

What’s the best approach in Python (Shapely, GeoPandas, or scipy.spatial, etc.) for:

• 
  Detecting line segments that are nearly parallel/overlapping within a tolerance.
  
  
• 
  Merging them into one “average” segment.
  
  
• 
  Accumulating a density/weight so I can later visualize these as heatmap lines (e.g., in Mapbox).
  

Performance is of course really important as this is being run in a geodjango backend so rapid https responses is good!

Any tips or links to solutions would be very much appreciated. It would also be good to know if this is entirely the incorrect approach for generating a heatmap (specifically one which captures these lines rather than gaussian blobs from individual points) from many disjointed linestrings and whether there is a more industry accepted approach.

Thanks

Continue reading...