Obtaining Dense Road Speed Estimates from Sparse GPS Measurements

A major challenge for traffic management systems is the inference of traffic flow in regions of the network for which there is little data. In this paper, GPS-based vehicle locator data from a fleet of 40-60 roving ambulances are used to estimate traffic congestion along a network of 20,000 streets in the city of Ottawa, Canada. Essentially, the road network is represented as a directed graph and a belief propagation algorithm is used to interpolate measurements from the fleet. The system incorporates a number of novel features. It makes no distinctions between freeways and surface streets, incorporates both historical and live sensor data, handles user inputs such as road closures and manual speed overrides, and is computationally efficient - providing updates every 5 to 6 minutes on commodity hardware. Experimental results are presented which address the key issue of validating the performance and reliability of the system.