Master of Science
Advanced Driver Assistance Systems, or ADAS, which can notify the driver of potential dangers or even perform emergency maneuvers in dangerous situations, have been shown to play a crucial role in accident prevention and driver feedback. An Intelligent ADAS, or i-ADAS, relies on information about the state of the driver, their behavior or condition, the vehicle and the environment. Understanding the behavior requires the development of driver models, which can help predict how a person may react in certain situations or help determine if the individual is not performing at their usual level of ability. A key element in building such models is the ability to detect and analyze common driving maneuvers, such as making turns, on an individual-by-individual basis. Thus algorithms are needed which can detect and characterize individual driving maneuvers.
In this research, we present a position-based turn detection algorithm for detecting turns from vehicle data and GPS coordinates. Based on a dataset of sixteen drivers involving 278 turns, the algorithm achieves an accuracy of 97.84%. The turn parameters detected by the algorithm are then averaged for each driver and clustered using K-Means. Turn parameters t - 5 seconds are also clustered prior to each detected turn and t + 5 seconds are clustered after each turn. The cluster centroids at each point in time determine particular driving behaviours which are summarized in four categories, and the cluster assignments are examined over time to categorize drivers into these behaviour categories. This analysis reveals two optimal times for analyzing driver behaviour. Our overall aim is to be able to build automated methods that can use this research to eventually determine characteristics of individual drivers during turns in order to build models of drivers for use with i-ADAS.
Knull, Jennifer Emily, "Turn Detection and Analysis of Turn Parameters for Driver Characterization" (2017). Electronic Thesis and Dissertation Repository. 4818.