Identification of driving manoeuvres using smartphone-based GPS and inertial forces measurement

Engelbrecht, Jarrett (2015-03)

Thesis (MEng)--Stellenbosch University, 2015.

Thesis

ENGLISH ABSTRACT: Road accidents are a growing concern for governments and is rising to become one of the leading causes of death in developing countries. Aggressive driving is one of the major causes of road accidents, and it is therefore important to investigate ways to improve people's driving habits. The ubiquitous presence of smartphones provides a new platform on which to implement sensor networks in vehicles, and therefore this thesis focuses on the use of smartphones to monitor a person's driving behaviour. The framework for a smartphone-based system that can detect and classify various driving manoeuvres is researched. As a proof of concept, a system is developed that specifically detects lateral driving manoeuvres and that classifies them as aggressive or not, using a supervised learning classification algorithm. Existing solutions found in research literature are investigated and presented. The best existing solution, a dynamic time warping classification approach, is also implemented and tested. We use an aggressive driving model that is based on the angle of a turn, the lateral force exerted on the vehicle and its speed through the turn. The tests and results of the implemented manoeuvre detection and classifcation algorithms are presented, and thoroughly discussed. The performance of each classifer is tested using the same data set, and a quantitative comparison are made between them. Ultimately, a lateral driving manoeuvre detection and recognition system was successfully developed, and its potential to be implemented on a smartphone was substantiated. The suitability of supervised learning classi ers for classifying aggressive driving, in comparison to dynamic time warping classifcation, was successfully demonstrated and used to validate our aggressive driving model. Conceivably, this work can be employed in the future to develop an holistic smartphone-based driver behaviour monitoring system, which can be easily deployed on a large scale to help make the public drive better. This would make our roads safer, reducing the occurrence of road accidents and fatalities.

AFRIKAANSE OPSOMMING: Padongelukkige is 'n groeiende bekommernis vir regerings en is een van die hoof oorsake van sterftes in ontwikkelende lande. Aggressiewe bestuur is een van die grootste oorsake van padongelukke, en dit is dus belangrik om ondersoek in te stel oor hoe mense se bestuurgewoontes verbeter kan word. Die alomteenwoordigheid van slimfone bied 'n nuwe platform waarop sensor netwerke geïmplementeer kan word in voertuie. Daarom fokus hierdie tesis op die gebruik van slimfone om 'n persoon se bestuurgedrag te moniteer. Die raamwerk vir 'n slimfoon-gebaseerde stelsel wat verskeie bestuurbewegings kan opspoor en klassifiseer is nagevors. As 'n bewys van die konsep, is 'n stelsel ontwikkel wat spesifiek laterale bestuurbewegings opspoor en dan klassifiseer of dit aggressief is of nie, met behulp van 'n klassifikasie algoritme wat onder toesig geleer is. Bestaande oplossings gevind in navorsingsliteratuur word ondersoek en aangebied. Die beste bestaande oplossing, 'n dinamiese tyd buiging klassifikasie benadering, word ook geïmplementeer en getoets. Ons gebruik 'n aggressiewe bestuurmodel wat gebaseer is op die hoek van 'n draai, die laterale krag wat uitgeofen is op die voertuig en sy spoed deur die draai. Die toetse en die resultate van die geïmplementeer beweging opsporing en klassifisering algoritmes word aangebied, en deeglik bespreek. Die prestasie van elke klassifiseerder is getoets met behulp van dieselfde stel data, en 'n kwantitatiewe vergelyking is tussen beide gemaak. Oplaas is 'n laterale bestuurbeweging bemerking en herkenning stelsel suksesvol ontwikkel en sy potensiaal om geïmplementeer te word op 'n slimfoon is gestaaf. Die geskiktheid van die onder-toesig-geleerde klassifiseerders vir die klassifikasie van aggressiewe bestuur, in vergelyking met dinamiese tyd buiging klassifikasie, was suksesvol gedemonstreer en gebruik om ons aggressiewe bestuurmodel te bewys. Hierdie werk kan in die toekoms gebruik word in 'n holistiese slimfoon-gebaseerde bestuurdergedrag monitering stelsel, wat maklik op groot skaal ontplooi kan word om te help verseker dat die publiek beter bestuur. Dit sal ons paaie veiliger maak, en die voorkoms van padongelukke en sterftes verminder.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/96597
This item appears in the following collections: