Access Type

Open Access Dissertation

Date of Award

January 2014

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

First Advisor

Hao Ying

Abstract

Advanced driver assistance systems, such as unintentional lane departure warning systems, have recently drawn much attention and R & D efforts. Such a system will assist the driver by monitoring the driver or vehicle behaviors to predict/detect driving situations (e.g., lane departure) and alert the driver to take corrective action. In this dissertation, we explored utilizing the nonlinear binary support vector machine (SVM) technique and the time series of vehicle variables to predict unintentional lane departure, which is innovative as no machine learning technique has previously been attempted for this purpose in the literature. Furthermore, we developed a two-stage training scheme to improve SVM's prediction performance. Our SVMs were trained and tested using the experiment data generated by VIRTTEX, a hydraulically powered 6-degrees-of-freedom moving base driving simulator at Ford Motor Company. The data represented 16 drowsy drivers (about three-hour driving time per subject) and six control drivers (approximately 20 minutes driving per subject), all of which drove a simulated 2000 Volvo S80. More than 100 vehicle variables were sampled at 50 Hz. There were a total of 3,508 unintentional lane departure occurrences for the 16 drowsy drivers and 23 for four of the six control drivers (two had none). We optimized the performances of the SVMs by experimentally finding their best kernel functions and parameter values as well as the most appropriate vehicle variables as their input variables. Our experiment results involving the 22 drivers with a total of over 6.84 million prediction decisions demonstrate that: (1) the two-stage training scheme significantly outperformed the commonly used (one-stage) training scheme, (2) excellent SVM performances, as measured by numbers of false positives and false negatives, were achieved when the prediction horizon was set at 0.6 s or shorter, (3) lateral position and lateral velocity served as the best input variables among the nine variable sets that we explored, and (4) the radical basis function was the best kernel function (the other two kernel functions that we tested were the linear function and the second-order polynomial). We conclude that the two-stage-training SVM approach deserves further exploration because to the best of our knowledge, it has demonstrated the best unintentional lane departure prediction performance relative to the literature.

Download

Share

COinS