Pitch based vehicle localization using time series subsequence matching with Multi-scale Extrema Features

Non-GPS localization of vehicles on roadways has received considerable attention in recent years and a number of solutions have been proposed, with most solutions addressing local tracking. This paper presents an algorithm that achieves global localization within very large road networks using pitch information. A key contribution is the development of the Multi-scale Extrema Feature that provides a number of advantages over traditional time-series subsequence matching methods in order to implement the above scheme. The algorithm´s results in localizing a vehicle´s position without initialization within a road network spanning 6000 Km are also presented.