Extracting line string features from GPS logs

Geo data is an important foundation for any type of location-based service, but geo data often is expensive or distribution is limited by certain license restrictions. As a solution, open projects collect geo data from individuals and publish these data under a public licence. As a major drawback, the correction and integration of collected data is difficult and cost-intensive. This paper describes an approach to automatically derive linear map data (roads, paths, highways etc.) from GPS logs. People just have to passively carry inexpensive GPS loggers whenever they drive or walk outdoors. The raw GPS logs then are automatically merged to a map. The approach contains error correction and sensor data fusion mechanisms. Our algorithm takes into account the specific measurement characteristic of GPS and is based on a probabilistic model. It performs two steps: first, track parts that represent the same paths are identified. Second, identifiable parts are fused to a single path, considering the Gaussian distribution of the input measurements. We verified the approach with approx. 200 000 measurements in the area of Nuremberg that represents approx. 6 000 km driving distance. We show that an automatic processing of GPS logs to produce a road map is effectively possible. Extracting line string features from GPS logs Schriftenreihe der Georg-Simon-Ohm-Hochschule Nürnberg 81