I have a large database of gps traces (long, lat, timestamp) and would like to see if I can detect when congestion occurs on roads from that data. Any idea's or pointers on good algorithms to use? I've seen loads of stuff using fixed point collectors (ANPR systems, video cameras, infra-red detectors, etc) but not from GPS traces.
Automated Traffic Incident Detection with GPS Equipped Probe Vehicles
"This paper discusses the results of a proof-of-concept study in which a transit vehicle fleet was used as the probe system for traffic incident detection."
Results: "The results indicate highly successful Detection Rate (DR) with low False Alarm Rate (FAR). The techniques also yield an average Time to Detect (TTD) of 30 seconds to 3 minutes depending on the probe vehicle population size, incident characteristics, and GPS performance"
Credit to Chaminda Basnayake
See the 'Travel Time' algorithm on page 4.
I have been doing something similar lately with AIS vessel tracks and modelling navigational risk. The best place to start is to consider your software and whether you want to use GIS or analyse these potential collisions using some SQL analysis instead. What is it you want to eventually produce?...traffic DENSITY is a good indicator for congestion based in time, maybe create some 90th percentiles or my colleagues have been VERY impressed with using the animation within ArcScene to display pinch points and movement.