SeTra: A Smart Framework for GPS Trajectories' Segmentation

Abstract

The main search engines nowadays are able to quickly perform searches on texts, images, audio or video. Recently a growing interest was devoted to the searching in the field of a new type of data resulting from the storage of GPS tracks: the trajectories. Therefore, we need to consider new searching strategies and methods fit to these data structures. The idea is to design a search engine aimed at the indexing and retrieving trajectories in specific repositories. This paper describes a starting framework for semantic analysis of trajectories data. This framework aims at detecting and extracting features from GPS raw data. These elements, in turn, populate new feature vectors in a MIRS Multimedia Browsing Indexing and Retrieval System, to support location-based queries. For several reasons, the feature extraction process starting from raw GPS data is a multiple steps process. The raw GPS data are affected by noise and inaccuracies, the trajectories are not uniform and homogeneous data because, for example, they are drawn by repeatedly changing the means of locomotion. Furthermore, the usual sections of trajectory ('trip segments'), describing a motion with certain kinematic parameters, could be alternated with slowly strolling around in confined areas ('activity') such as a mall, a gym or a small market. The reasons listed above require the pre-processing of the raw GPS data in order to split the paths into uniform and homogeneous subsets. The pre-processing step removes the noise from the trajectories and separates the 'trip segments' from the 'activities'. Next, each 'trip segment' is classified using a sequence of two detection techniques based on fuzzy logic and SVM. Finally, the synthetized trajectory components are stored in feature vectors, ready to start location-based queries for retrieval systems. The indexed trajectories will allow the queries in a database, for example, on both on-road routes and off-road trails. The model proposed in this framework provides the basis module for support retrieving of pre-compiled trajectories but instead not supports the streaming of trajectories if the trajectory is not yet fully completed.