A pedestrian network construction algorithm based on multiple GPS traces

Currently, pedestrian networks are not as widely available as road networks are for many areas; this is partly due to the high demand for applications such as car navigation. However, this situation is rapidly changing as GPS-enabled mobile devices are increasingly becoming commonplace for pedestrian applications and as they are potential for volunteered data collection. For this reason, the development of efficient and cost-effective techniques appropriate for collecting and constructing pedestrian network databases is needed. This paper presents an algorithm for automatically identifying geometries of pedestrian path segments and constructing a pedestrian network from multiple GPS traces. The algorithm was experimented by using two sets of data sources, self-collected walking traces and OpenStreetMap (OSM) walking traces. The experiment with self-collected data was validated by comparing the constructed pedestrian network with a network baseline and calculating the quality measurements in three environment settings. The experiment with OSM traces was validated by visually inspecting maps overlaid on high-resolution images. Evaluation results show that the algorithm can automatically construct pedestrian networks.

► Of the 10 tiles in the study area, the average geometrical correctness is 57.55%.
► Of the 10 tiles in the study area, the average topological correctness is 76.90%.
► The algorithm performs well in open sky environments.
► More repeated GPS traces results in more accurately generated segments.
► The first five repeated GPS traces improved the accuracy the most.