Human-vehicle interaction analysis

Abstract

The present work is the Thesis work of Vehicle assessment comparison from a smartphone reference with different approaches, to pursue the Master on Science on Manufacturing Systems. The automotive industry is continuously evolving by implementing top-edge technologies to improve comfort, safety, and driving experience to the users. In the context of Industry 4.0 and the Smart Cities paradigm, the concept of Intelligent Transportation System has become a research topic in the last few years. In the race for autonomous driving, researchers and industry have stressed the importance of monitoring drivers and passengers to determine the driving style, safety, and fuel efficiency, among other essential features. Despite all the work that has been done to monitor drivers, some approaches consist of vehicle-fixed devices or personalized devices that do not allow for the reproduction of experimentation to other vehicles. This instrumentation limits enormously the possibility to monitor any type of vehicle and collect information to develop intelligent algorithms that can predict driver and vehicle features such as driving behavior, energy consumption, fatigue, or vehicle’s element prognosis. Current researches focus on analyzing the interaction as a system from the vehicle’s point of view or driver’s point of view. Nevertheless, they have not been observed on both sides. To overcome these issues, an experimental setup is proposed on this work. The importance of this project is the easiness and replicability of the experimental setup; it is then validated by analyzing the logged data and the correlations between variables. Besides, state-of-the-art algorithms are compared to validate and select the best performance. This thesis integrates an experimental setup easy to use and implement with available commercial devices. Then, to validate the setup, a selection of algorithms based on a literature review were replicated and fed with the data logged from the experimental setup. A set of analyses of the resulting dataset is done to observe the interaction of vehicle and driver signals’ performance on how these signals are correlated. The first part of this work is devoted to the experimental setup definition and testing. Here the process was iteratively done by generating a procedure. Then, the next step consisted of exploring the logged data with a statistical tool to determine a possible correlation between signals and to reduce the dataset order but preserving most of the information. Later, state-of-the-art algorithms and data-driven identification models were identified and validated for specific key performances of vehicles and drivers. The vehicle’s key performances boarded on this thesis are the driving style, energy consumption, and emissions. Besides, the driver’s key performances are the heart condition, temperature, electrodermal activity, and heart rate. These features are highly studied when evaluating vehicle’s or driver’s state. The result of evaluating these performances with the selected algorithms shown that the driving style had 77% of correct classification, energy consumption, and emissions had around 16% and 11% of relative error, respectively. The results of this project show how vehicle and driver interact by analyzing its key performances. The Principal Component Analysis technique helped to find correlation among the raw data and also reduced the features from 57 to 27 without significant losses on the information. Besides, it demonstrated the correlation between vehicle’s and driver’s key performances by analyzing PCA plots and the covariance matrix.