Computer Vision at Intersections: Explorations in Driver Assistance Systems and Data Reduction for Naturalistic Driving Studies

Philipsen, Mark Philip ; Jensen, Morten Bornø

4. term

Vision, Graphics and Interactive Systems, Master

2015

2015-06-03

This report constitutes a long master thesis in Vision, Graphics, and Interactive Systems. It details the work done during almost two semesters abroad at UC San Diego. The work has been research oriented, the report is therefore structured in separate parts in contrast to a standard linear product development flow. The scope of the research has primary been traffic light detection, but also the development of a vehicle detector at intersection used for naturalistic driving studies. A comprehensive survey of traffic light recognition (TLR) systems inducing both academic and industrial research has been done. From this it is clear that TLR research lack challenging public datasets and standardized evaluation methodology. Thus, the largest public dataset in the world with around 110,000 annotated traffic lights has been created and made available. Finally a comparative analysis of a learning-based versus heuristic model-based approaches in relation to traffic light detection is done. The work has resulted in submission of one conference paper to Intelligent Vehicles Symposium 2015, two conferences papers to Intelligent Transportation Systems Conference 2015, and finally a journal paper for Transactions on Intelligent Transportation Systems.

Computer vision ; driving assistance systems ; machine learning ; intelligent vehicles

Download
View record in AAU Student Projects