Identification of driver baselines

Abstract

This thesis aims to answer whether it is possible to produce one or more baselines based on naturalistic driving data collected over a period of 8 months. The baseline is based on variables extracted from the drivers action, such as acceleration and gaze vectors, along with variables extracted from the nature of the trip, such as time of day or road type. If shown that baselines can be deduced from these variables, it can be used to improve existing ADAS in terms of both safety and comfort. The analysis was made using statistical analysis, visual representations of data distributions, Gaussian mixture model clustering and time series clustering. The results suggests that it is not possible to determine only one baseline, or what that baseline might be for each driver. Instead the results suggests (at two occurrences) multiple baselines for each driver based on the nature of the trip. This may mean that multiple baselines could be established for different scenarios the driver might find itself. However as the data is limited, these findings may not be representative of a larger population. To confirm the findings in this paper further research has to be conducted on a larger set of data containing more drivers.