Segmentation of Geolocalized Trajectories using Exponential Moving Average
Résumé
Nowadays, large sets of data describing trajectories of mobile objects are made available by the generalization of geolocalisation sensors. Relevant information, for instance, the most used routes by children to go to school or the most extensively used streets in the morning by workers, can be extracted from this amount of available data allowing, for example, to reconsider the urban space. A trajectory is represented by a set of points (x; y; t) where x and y are the geographic coordinates of a mobile object and t is a date. These data are difficult to explore and interpret in their raw form, i.e. in the form of points (x; y; t), because they are noisy, irregularly sampled and too low level. A first step to make them usable is to resample the data, smooth it, and then to segment it into higher level segments (e.g. "stops" and "moves") that give a better grip for interpretation than the raw coordinates. In this paper, we propose a method for the segmentation of these trajectories in accelerate/decelerate segments which is based on the computation of exponential moving averages (EMA). We have conducted experiments where the exponential moving average proves to be an efficient smoothing function, and the difference between two EMA of different weights proves to discover significant accelerating-decelerating segments.