NON-INTRUSIVE DRIVER DROWSINESS MONITORING VIA ARTIFICIAL NEURAL NETWORKS

Open Access
Author:
Culp, Jonathan
Graduate Program:
Mechanical Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
None
Committee Members:
  • Moustafa El Gindy, Thesis Advisor
  • Md Amanul Haque, Thesis Advisor
Keywords:
  • RPROP training
  • driving simulation
  • artificial neural networks
  • driver drowsiness
  • Radial Basis Networks
Abstract:
A completely non-intrusive method of monitoring driver drowsiness is described. Because of their abilities to learn behavior and represent very complex relationships, artificial neural networks are the basis of the method presented. Four artificial neural networks are designed based on the hypothesis that the time derivative of force (jerk) exerted by the driver at the steering wheel and accelerator pedal can be used to discern levels of alertness. The artificial neural networks are trained to replicate non-drowsy input, and then tested with unseen data. Data sets that are similar to the training sets will pass through the network with little change, and sets that are different will be changed considerably by the network. Thus, the further the driver’s jerk profile deviates from the non-drowsy jerk profile, the greater the error between the input and output of the network will be. The changes in network error with drive time are presented from testing the networks with simulated driving data, and the performance of the artificial neural network designs are compared.