Map inference based on the crowd aggregated mountain GPS traces

Abstract

We analyzed aggregated by crowd GPS traces over mountain area, determined challenges and difficulties of map inferencing over mountain trails. Proposed specific solutions and thoughts about what should be considered while implementing algorithms to detect centerline over GPS mountain traces and its implications. In this research, mountain trails unique features were derived and explored major challenges of trails features while implementing map inference algorithms TC1 and KDE. Second, original new algorithm was developed to tackle mountain challenges and problems while implementing TC1 and KDE, called TC2. TC2 encompassed a wide approaches from existed algorithms. Extracting the most densest parts, by using kernel mean points, retrieving closest segments to those kernel points and iteratively clustering them with single-linkage methodic from TC1 algorithm based on mean directions. Third, existed ground truth map from national park was analyzed and it was concluded that ground truth made by survey person has inconsistencies. Such inconsistencies could be avoided using automatic map interference.