You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
48 lines
5.6 KiB
48 lines
5.6 KiB
\chapter{Introduction} |
|
\label{ch:01_introduction} |
|
|
|
|
|
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
|
\section{Context and motivation} |
|
\label{sec:xxx} |
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
|
|
|
|
|
Accurate prediction of the individual driver behavior is essential for various automotive future technologies. Forecasting applications addressing the driver behavior are for instance risk assessment of traffic scenarios for active and passive safety, automated driving, driving route prediction (forscasting) and control functions or energy management and strategies [1]–[3]. With increasing and more accurate information of vehicle and environmental data ensuring precise predictions, advanced driver assistance systems (ADAS) combined with driver models guarantee high reliability. |
|
|
|
|
|
Using the back end server-side with online data, modern navigation systems provide the ADAS and driver model with a high amount of forecast information of the vehicle environment. Based on the ADAS application, the prediction horizon can be classified into short-range and long-range predictions. The two categories differ mainly in the range of the input data. While short-term predictions affect to the vehicle sensors ranging a few hundreds meters in front of the ego-vehicle motion, long-range prediction uses navigation and road based data approaches minimizing the time dependency in the forecast. |
|
|
|
The work focuses on long-range prediction horizons ranging from a few kilometers ahead to the final destination. Automotive applications using the long-range prediction horizon are for instance efficiency systems, which are optimizationstrategies of hybrid electric vehicles (HEVs) and range calculation for battery electric vehicles (BEVs). Thereby the precision of the upcomming power and velocity prediction is instrumental for the overall vehicle power consumption of HEVs and influences the range forecast of BEVs [4]. |
|
|
|
|
|
Focusing the range calculation and energy forecast of BEVs, main influences of predictive energy consumption are given by the vehicle parameters, which are for instance the vehicle mass together with the driving resistance coefficients and the driver depended vehicle dynamics given by the individual driving velocity and its time derivatives [5]. With individual velocity prediction the ADAS can give premature feedback to the driver enabling a satisfactory and reliable systems for the customer. |
|
|
|
Long-range prediction of the individual velocity is challenging though, since the probability of the predicted speed occurrence decreases with an increasing prediction horizon [6]. The prediction output quantity is influenced by several uncertain factors, which are for instance traffic jams, time dependent road surfaces and weather conditions. In addition the driver behavior can change over time, which leads to a complex system that is hard to grasp with pure physical relations. To model the driver behavior machine learning algorithms are used. Latest research analyzed neural networks, non-parametric regressions or stochastic prediction models [4], [6]–[9]. |
|
|
|
While [4], [6], [9] do not take environmental features and [8] does not take the predicted continuous vehicle velocity into account, this approach investigates modeling the driver individual behavior with a stochastic process given the environment. The work examines the modeling procedure in combination with the influences on the driver behavior based on environmental factors, e.g. road curvature. The parameters and distributions of this process are learned based on individual naturalistic driving data. Thereby, linearity and Gaussian distribution assumption for underlying densities that lead to the Kalman Filter and Rauch-Tung-Striebel Smoother algorithms showed to be too limited. The use of non-parametric distributions, making numerical methods like particle-based algorithms essential, showed far more promising results for modeling driver individual velocity behavior. |
|
|
|
The stochastic algorithms together with the online traffic speed source and a neural network approach given [7], are evaluated by using 1500 km real world naturalistic driving data. The study data refers to to 39 driving tracks and includes 8 different drivers. On the basis of the data set, a total |
|
of 5 different velocity prediction algorithms are trained and validated. The stochastic models show high accuracy and |
|
precise prediction result. Following the remainder of this paper is organized. In Sec. II, the problem formulation with the modeling procedure |
|
from the driving surrounding to the vehicle dynamic including driver behavior is presented. Additionally conditional stochastic |
|
independence is defined guaranteeing the use of hidden Markov models. Furthermore the database of the study with |
|
a classification method are described. The four stochastic prediction models are explained in Section III, starting with |
|
the linear and Gaussian assumption, Sec. III-A. The nonparametric stochastic model is investigated in Subsection III-B. |
|
Finally, the results of the prediction algorithms are shown in Section IV with the subsequent conclusion in Sec. V. |
|
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
|
\section{Survey of related work} |
|
\label{sec:0101_xxx} |
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
|
|
|
|
|
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
|
\section{Research objectives and thesis outline} |
|
\label{sec:xxx} |
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |