Large Naturalistic Driving Studies (NDS) of driver behaviour are a quite recent development in traffic safety research. They aid construction of advanced driver assistance systems (ADAS), improve driver training programs, and serve as input into traffic safety legislation. They result in gigantic databases containing sensor, radar, GPS, and video recordings of many thousands of hours of normal driving. The data is used to estimate risk of accidents caused by complex interaction between driver behaviour, traffic situation and vehicle function. The complexity of the data, and problems to avoid sampling bias, poses difficult challenges for risk estimation ? and wrong estimates may lead to ADAS or training which increases accident risk. In this project we will adapt methods from epidemiology (smart subsampling, weighted risk estimation); extreme value statistics (extrapolation from incidents to real crashes); and high dimensional statistics to provide unbiased and efficient accident risk estimates from NDS data. Our estimates will be used in practical traffic safety research, and in a longer perspective contribute to diminishing the number of accidents. The project is a collaboration between researchers from SAFER, an internationally leading traffic safety research center, and from Mathematical Sciences in Gothenburg. It includes expertise from frontline research in traffic safety, statistical epidemiology, statistical extreme value theory, and high-dimensional statistics.
Professor Emeritus at Chalmers, Mathematical Sciences, Applied Mathematics and Statistics
Funding Chalmers participation during 2012–2015