Is Posturography a Candidate for a Vigilance Test

Studies exhibit that between 20 and 40% of traffic accidents in Germany are related to driver’s hypovigilance. Hypovigilance, as stated by some authors, effects driver’s per formance in a similar way as alcohol consumption does. But un like blood alcohol level testing up to now there is no mobile, non invasive vigilance test with low test durations. Posturography -a method to assess the balance control system quantitatively -may provide the possibility for such a vigilance test. In this paper we will investigate the discriminatory abilities of posturography using data recorded in conjunction with overnight driving simulation experiments commissioned by Caterpillar Machine Re search. A total of 19 young adults volunteered to participate in balance assessment. Experiments included a driving simulation with repetitive assessment of subjective self-rating while driving using Karolinska Sleepiness Scale. As objective vigilance score the standard deviation of lateral position was chosen. Subjective and objective vigilance scores and the Time-Since-Sleep are utilized as labels for discriminant analysis. Two kinds of features were extracted from posturographic recordings. Among others, parameters of diffusion plot and sway density analysis were utilized as features in time domain. As spectral domain features power spectral densities were estimated and averaged in empirically optimized equidistant frequency bands. The usefulness of posturography as vigilance test was evaluated by the mean test set error of computational intelligence algorithms including artificial neural networks and Support-Vector Machines (SVM). These algorithms can be regularized between local and global decision finding. SVM using Gaussian kernel function achieve error rates of 8.8% (leave-one-out cross-validation). Consider ing some concerns regarding reliability and validity we conclude that it is possible to discriminate patterns of different vigilance levels using posturography.