Real-time detection of moving crowds using spatio-temporal data streams

Abstract

Over the last decade we have seen a tremendous change in Location Based Services. From primitive reactive applications, explicitly invoked by users, they have evolved into modern complex proactive systems, that are able to automatically provide information based on context and user location. This was caused by the rapid development of outdoor and indoor positioning technologies. GPS modules, which are now included almost into every device, together with indoor technologies, based on WiFi fingerprinting or Bluetooth beacons, allow to determine the user location almost everywhere and at any time. This also led to an enormous growth of spatio-temporal data.

Being very efficient using user-centric approach for a single target current Location Based Services remain quite primitive in the area of a multitarget knowledge extraction. This is rather surprising, taking into consideration the data availability and current processing technologies. Discovering useful information from the location of multiple objects is from one side limited by legal issues related to privacy and data ownership. From the other side, mining group location data over time is not a trivial task and require special algorithms and technologies in order to be effective.

Recent development in data processing area has led to a huge shift from batch processing offline engines, like MapReduce, to real-time distributed streaming frameworks, like Apache Flink or Apache Spark, which are able to process huge amounts of data, including spatio-temporal datastreams.

This thesis presents a system for detecting and analyzing crowds in a continuous spatio-temporal data stream. The aim of the system is to provide relevant knowledge in terms of proactive LBS. The motivation comes from the fact of constant spatio-temporal data growth and recent rapid technological development to process such data.