Resumo:
Vehicles are becoming more intelligent and connected due to the demand for faster, efficient,
and safer transportation. For this transformation, it was necessary to increase the
amount of data transferred between electronic modules in the vehicular network since it is
vital for an intelligent system’s decision-making process. Hundreds of messages travel all
the time in a vehicle, creating opportunities for analysis and development of new functions
to assist the driver’s decision. Given this scenario, the dissertation presents the results of
research to characterize driving styles of drivers using available information in vehicular
communication network.
This master thesis focuses on the process of information extraction from a vehicular network,
analysis of the extracted features, and driver classification based on the extracted
data. The study aims to identify aggressive driving behavior using real-world data collected
from five different trucks running for a period of three months. The driver scoring
method used in this study dynamically identifies aggressive driving behavior during predefined
time windows by calculating jerk derived from the acquired data. In addition, the
K-Means clustering technique was explored to group different behaviors into data clusters.
Chapter 2 provides a comprehensive overview of the theoretical framework necessary for
the successful development of this thesis. Chapter 3 details the process of data extraction
from real and uncontrolled environments, including the steps taken to extract and refine
the data. Chapter 4 focuses on the study of features extracted from the preprocessed data,
and Chapter 5 presents two methods for identifying or grouping the data into clusters.
The results obtained from this study have advanced the state-of-the-art of driver behavior
classification and have proven to be satisfactory. The thesis addresses the gap in the
literature by using data from real and uncontrolled environments, which required preprocessing
before analysis. Furthermore, the study represents one of the pioneering studies
conducted on commercial vehicles in an uncontrolled environment.
In conclusion, this thesis provides insights into the development of driver behavior classification
models using real-world data. Future research can build upon the techniques
presented in this study and further refine the classification models. The thesis also addresses
the threats to validity that were mitigated and provides recommendations for
future research.