Traffic Flow Estimation using Probe Vehicle Data

Traffic sensing has been revolutionized with the commoditization of GPS technology. Smartphone navigation applications ubiquitously track vehicles as samples of the overall traffic. This so-called Probe Vehicle Data (PVD) has replaced traditional road-side sensor technologies, such as induction loops and microwave sensors, given its relative low cost, good coverage, and reliability. However, while PVD allows us to assess speed and by extension the overall traffic condition in a road network, this sample-based approach does not provide us with traffic flow, i.e., the number of vehicles passing through an edge of the road network. This paper bridges this gap by proposing and evaluating a range of methods to infer traffic flow for a road network that is ubiquitously observed using probe data but having traffic flow measurements only in very road-side sensor locations. We create Road Segment Archetypes that relate PVD speeds to flow from road-side sensors for these locations. These archetypes are then extended to the entire network covered only by PVD based on similar traffic characteristics. Using these archetypes we augment and experimentally evaluate different traffic flow estimation models using real-world traffic data. Experimental results show that the Road Archetype flow estimation is comparable to the accuracy of prediction models that would be based on actual road-side sensor flows.