Highly-Parameterizable Data Generation and Modularized Real-Time Data Stream Analysis in Forest Fire Fighting Scenarios (Master Thesis, Finished)


Marko Obradovic


In the last years, sensor devices have been steadily improved, shrunk, and cheapened. Moreover, according to the IoT vision, today many sensor devices are able to connect either directly or via a smart phone or a smart watch to the Internet and thus to emit their measurements as continuous data streams. In the course of this development, the usage of sensor devices for the purpose of monitoring humans and their environment consensually has heavily increased. This is especially true if the humans act as members of a team who collaborate in order to achieve a common objective such as in disaster management scenarios or sports competitions.

In these scenarios the coordinator of the team (i.e., the officer-in-charge or the coach) can receive the sensor data streams that are emitted by sensor devices attached to the team members and visualize them. However, the full advantage of the monitoring first emerges if the raw sensor data streams are filtered, processed, and analyzed beforehand by a real-time data stream analysis system. More precisely, we argue that it would be very beneficial for the team coordinator if there was a real-time data stream analysis system that consumes the raw sensor data streams and produces more meaningful and useful output data streams that can be received by and/or visualized for the team coordinator. In course of this Master Thesis, such a real-time data stream analysis system that is able to perform complex team analyses in a forest fire fighting scenario should be developed.

However, for the purpose of developing such a real-time data stream analysis system, one first needs an evaluation dataset in order to be able to evaluate the correctness and performance of the system. More precisely, one requires a set of stored data streams (one for each sensor device) that can be re-played in order to generate data streams which can be analyzed in real-time by the analysis system. In sports scenarios, it is possible to capture real sensor data streams and store them as datasets for evaluation purposes. For instance, it is possible to install the components of a measuring system at the equipment of the players of two soccer teams, inside the ball and around a soccer field and to capture the sensor data streams that are emitted during a test match as done in [1]. In contrast, it is impossible or at least neither morally acceptable nor allowed to engender a natural disaster such as a forest fire in order to capture the data streams that are emitted by the firefighters who try to extinguish the fire.

Therefore, the first part of this Master Thesis is to develop a parameterizable simulator that generates a realistic evaluation dataset for the forest fire fighting scenario in which dozens or even hundreds of firefighters collaborate to extinguish the fire and rescue civilians. Assuming that each of the firefighters is equipped with a multitude of sensor devices that measure his/her position, health state (e.g., heart beat rate) and environment (e.g., temperature and wind speed), the simulator has to generate a data stream for each of the sensor devices.

The second part of this Master Thesis is to develop a distributed real-time data stream analysis system for this scenario. The main purpose of the analyses should be to equip the officer-in-charge in real-time with an up-to-date and complete overview of the scenario, to warn him/her as fast as possible of current and future dangers and optimally even to suggest actions such as optimal evacuation routes. This requires to detect not only “simple” complex events (such as a fire) or to generate “simple” aggregates (such as the average temperature measured by a specific or all firefighters in a given interval) but to perform complex team analyses, i.e., to detect semantically-rich team events (such as that if there is a fire at the center and a wind gust in south direction measured at the north then there is a potential danger for firefighters and civilians in the south) and to generate complex team aggregates (such as a live map of the whole scenario that combines the latest measurements, historical data and interpolations), in real-time.

The real-time data stream analysis system has to be developed on top of OSIRIS-SE [2]. OSIRIS-SE is a stream-enabled workflow management system developed by the Databases and Information Systems (DBIS) research group of the University of Basel. OSIRIS-SE enables combining powerful workers that consume (input) data streams, perform a part of the analysis workload, and produce output data streams in real-time to a sophisticated analysis workflow. Hence, OSIRIS-SE can be regarded as a worker-based complex event detection (CED) system. Since the workers of an analysis workflow have to be implemented in Java and thus in a Turing-complete programming language it is possible to construct analysis workflows that perform arbitrary complex team analyses.

Another important requirement is that the workers have to be implemented in a highly modularized and re-usable way. More precisely, every worker should consist of (one or) multiple internal components that are performed in parallel and that do not share state. Each of these internal components should consist of multiple component building blocks (CBBs) that kind of form an internal component workflow. In course of this Master Thesis, the student has to define this workflow building model in more detail and to specify requirements and limitations of the model. Moreover, the student has to elaborate a set of parameterizable CBBs that are not only applicable for this but for all possible team collaboration or at least disaster management scenarios and facilitate scenario-specific CBBs only if it is indispensable.

The resulting real-time data stream analysis system has to perform some interesting and complex analyses that demonstrate the power of the worker-based CED approach not only for computer scientists but also for other interested people. For this purpose, a platform-independent GUI that visualizes the analysis results in an appealing way should be implemented.

In course of the Master Thesis preparation, the forest fire fighting scenario as well as the goals and tasks have to be defined in more detail in a written proposal. The proposal should also comprise limitations of the planned system and a concrete working plan (Gantt diagram).


The following Master Thesis is proposed in the context of the StreamTeam project of the DBIS Group at the University of Basel. In this thesis, we provide a scenario-specific approach to Real-time Complex Team Analyses. In particular, we focus on Complex Team Analyses in Forest Fire Fighting scenarios.

We argue that in order to properly perform Real-time Complex Team Analyses three matters need to be addressed: data, analysis and visualization. Hence, we decompose the overall problem into three subproblems: the Data Problem, the Analysis Problem and the Visualization Problem. For the Forest Fire Fighting scenario we provide F3DataGen, F3Analyzer, and F3Visualizer as solutions for the proposed subproblems.
F3DataGen is a parameterizable and well founded Forest Fire Fighting Scenario Simulation, Sensor Data Generation and Stream Simulation System which provides sensor datasets by realistically simulating Forest Fire Fighting scenarios. Moreover, it simulates real-time Sensor Data Streams using the generated sensor datasets.

F3Analyzer is a distributed and modularized Real-time Complex Team Analysis Workflow on the basis of a Real-time Data Stream Analysis System called PAN. It provides evidence for the applicability of Real-time Complex Team Analyses to Forest Fire Fighting scenarios. Besides being suited for Forest Fire Fighting scenarios, it is equipped with generic analysis components that can be used for Complex Team Analyses in arbitrary scenarios. Thereby, we make a first small step towards the generic Real-time Data Stream Analysis System for scenario-independent Complex Team Analyses which is envisioned in the context of the StreamTeam project.

F3Visualizer is a web-based scenario-specific visualization GUI and alert system which provides meaningful illustrations of the Complex Team Analysis results in real-time, and thus aims for improved incident management and monitoring in Forest Fire Fighting scenarios.

In the course of our evaluations, we provide nine different Forest Fire Fighting scenario simulations and the corresponding sensor datasets to confirm the quality of F3DataGen. Besides that, we examine the performance of the simulations in terms of runtime and memory consumption. Moreover, we prove the quality of the performed Complex Team Analyses by providing a statistical examination in terms of sensitivity, i.e., hit rate, of the proposed Complex Team Event Detections.

Start / End Dates

2016/04/14 - 2016/10/13


Research Topics