Consecutive Detection of Extreme Central Fatigue
In order to establish fatigue monitoring technologies a valid method for automatic detection of extreme central fatigue is needed. At present, acquisition of biosignals and their analysis by computational intelligence methods are most promising. We present experiments during which 10 volunteers drove overnight in our real-car lab following a partial sleep deprivation design. Based on several biosignals (EEG, EOG) recorded during microsleep events a classifier was constructed. We have shown earlier that spectral power densities of EEG and EOG averaged in narrow bands performed best as signal features and that carefully parameterized Support-Vector Machines perform best for classification. Afterwards, classification of approximately 1.5 million consecutively segmented biosignals was performed in order to check utility for real detector application. The independent validation of this step is shown to be crucial. Two different methods based on a subjecttive and an objective measure are presented. A methodological problem remains open in how to proceed with suspicious periods where some behavioral signs point to extreme fatigue, but driving seems still to be possible.
KeywordsEEG EOG Support-Vector Machines Fatigue Microsleep
Unable to display preview. Download preview PDF.
- 1.Gründl M (2005) Fehler und Fehlverhalten als Ursache von Verkehrsunfällen und Konsequenzen für das Unfallvermeidungspotenzial. PhD Thesis (in German), Regensburg, Germany, 2005.Google Scholar
- 3.Overview at http://www.microsleep.deGoogle Scholar
- 4.Golz M, Sommer D, Holzbrecher M, Schnupp T (2007) Detection and Prediction of Driver’s Microsleep Events, Proc 14th Int Conf Road Safety on Four Continents; Bangkok, Thailand, 2007Google Scholar