Adressing the Need for Map-Matching Speed: Localizing Global Curve-Matching Algorithms

Tracking data becomes an increasingly available sensor data resource that can be used in a range of applications related to traffic assessment and prediction. Of critical importance in this context is the amount of (i) historic and (ii) current tracking data available (data per spatial area). Using data of varying quality (sampling rate and accuracy) requires sophisticated map-matching algorithms. Having current data requires fast algorithms. Currently only fast Incremental and slow Global algorithms exist that produce low and high-quality results, respectively. This work proposes a fast map-matching algorithm that at the same time gives quality guarantees for the result. While employing a global matching strategy, we exploit auxiliary knowledge about the tracking data to improve the computation speed. The classical global matching approach is to find among all possible paths in the road network one path that is the most similar to the curve represented by the tracking data. Our novel Adaptive Clipping algorithm exploits worst-case error measures associated with the tracking data to limit the portions of the road network that need to be considered in the matching process (output sensitive algorithm), while using the weak Fréchet distance to measure similarity between curves. An experimental evaluation shows that the Adaptive Clipping algorithm runs as fast (and often faster) as the Incremental Algorithm, and produces high quality matching results comparable to those of Global algorithms.