Inferring Road Maps from Gps Traces : Survey and Comparative 1 Evaluation

1 Due to the availability of GPS sensors in a variety of everyday devices, GPS trace-data is becoming 2 increasingly abundant. One potential use of this wealth of data is to infer, and update, the geometry 3 and connectivity of road maps, using what are known as “map generation” or “map inference” 4 algorithms. These algorithms offer a tremendous advantage when no existing road map data is 5 present – rather than incur the expense of a complete road survey, GPS trace-data can be used to 6 generate entirely new sections of the road map at a fraction of the cost. Road map inference can 7 also be valuable in cases where a road map is already available. Here, they may not only help 8 to increase the accuracy of drawn road maps, but also help to detect new road construction and 9 dynamically adapt to road closures – useful features for digital road maps being used for in-car 10 navigation. Proposed algorithms have been accompanied by evaluations that are largely qualitative 11 in nature, and feature little or no comparison against prior work. To address this lack of quantitative 12 and comparative evaluation, this paper makes the following contributions: (a) a comprehensive 13 survey of the current literature on map generation. (b) the first method described for evaluating 14 automatically generated maps. (c) a qualitative, quantitative, and comparative evaluation of three 15 reference algorithms. (d) open-source implementations of the three algorithms, a 118 hour trace 16 data set and ground truth map for unrestricted use by the automatic map generation community. 17 Biagioni and Eriksson 3