Search

We present an extensible, generic, spatio-temporal trajectory simplification framework that modularises trajectory simplification into the stages of normalising, ranking, and reduction. We combine a range of ranking strategies and scoring heuristics – some from the literature and some new – into our framework modules and create a variety of spatio-temporal trajectory simplification methods. These trajectory simplification methods are experimented upon using real world and synthetic datasets, measuring running time, geometric displacement, and region-of-interest visitation. The results indicate that our proposed framework creates a number of efficient and effective spatio-temporal trajectory simplification methods. [ABSTRACT FROM PUBLISHER]

Abstract: Flickr presents an abundance of geotagged photos for data mining. Particularly, we propose the concept of extracting spatio-temporal meta data from Flickr photos, combining a collection of such photos together results in a spatio-temporal entity movement trail, a trajectory describing an individual's movements. Using these spatio-temporal Flickr photographer trajectories we aim to extract valuable tourist information about where people are going, what time they are going there, and where they are likely to go next. In order to achieve this goal we present our novel spatio-temporal trajectory regions-of-interest mining and sequential pattern mining framework. It is different from previous work since it forms regions-of-interest taking into consideration both space and time simultaneously, and thus produces higher-quality sequential patterns. We test our framework's ability to uncover interesting patterns for the tourism sciences industry by performing experiments using a large dataset of Queensland photo taker movements for the year 2012. Experimental results validate the usefulness of our approach at finding new, information rich spatio-temporal tourist patterns from this dataset, especially in comparison with the 2D approaches shown in the literature. [Copyright &y& Elsevier]