Analysis of lateral vehicle position using naturalistic driving data

Abstract

Traffic crashes have a significant impact on humans lives and the world economy. Understanding crash causation is an important step to prevent crashes from happening. One way of gaining information about crashes and human driver behaviour is through the analysis of naturalistic driving data. Often the challenge arises to generate the maximum information from a given database to answer a certain problem or research question. The thesis covers the research question, of how to analyze large quantities of video material in a quantifiable and efficient way to gain information about the lateral position of a vehicle. The research addresses that problem by developing a tool to extract the lateral vehicle position of a front view video and using reference data to evaluate its performance and sensitivity. A reliable set of CAN-Bus data serves as reference data for the vehicle position. The vehicle position is extracted from the video through multiple image processing steps. Evaluations are performed to assess precision and accuracy of the method. A sensitivity analysis is conducted to determine the impact of providing correct camera parameters. The method is applied to a new data set to ensure versatility. The results show that the method is reliable but highly sensitive to the orientation and position of the camera relative to the vehicle. The overall estimation of the precision shows a systematic error in the measurements caused by the camera parameters. The sensitivity evaluation shows that the height and the roll of the camera have a significant impact on the precision of the method. The developed tool can be used to determine the lateral vehicle position. Tuning of the tool is needed to further improve accuracy and precision of the tool.