Variables extraction and trajectory reconstruction for modelling driver behaviour

Abstract

Active safety and autonomous driving are two booming words in the automotive industry of today. Most, if not all, companies are in a race to provide a safe environment for driving and are thriving to reach their goals in producing autonomous vehicles. The general goal of this project is to develop a human driver model for critical lane-changing manoeuvres. Annotation of such data was done manually on videos acquired from dash-cam footage from a naturalistic driving study (NDS) by identifying di erent events in time, such as time of lane crossing and relative distance between vehicles. A semi-automatic tool to aid the annotation of the NDS footage was developed and several methods for range calculation are discussed such as the pixel width method and the triangulation method, which are also used for calculating the heading angle and the lateral o set of the lead vehicle. The results were found to be relatively accurate considering the simplicity of the tool. As for the tracking of the lead vehicles and the lanes, the most accurate method was found to be through interpolating between user inputs.