5.8  Concluding Remarks

Visual comparison of the high frequency activity illustrated using the spectrogram with the low frequency activity, clearly illustrates that changes in the high frequency activity have a direct correlation with the sleep onset period while changes in the low frequency activity are not consistent.  The high frequency signals have the potential of providing information useful in incipient drowsiness detection (if the signal processing methods are chosen appropriately).  The spectrograms used in this chapter only provide a crude look at the high frequency activity and more sophisticated analysis can be used to further explore the observed shifts in energy in terms of: sensitivity, stability, consistency, etc.  It also appears from this preliminary analysis that high frequency activity correlates with stage 1 sleep and microsleeps, which provides the basis for developing an algorithm that will be able to track transients in alertness faster than previous methods and may lend itself to simpler implementation than the traditional sleep staging rules.

Also, because this activity is unexplored in the literature, there may be more information content in this portion of the EEG spectrum than simply drowsiness tracking or sleep correlation.  To gain a fuller understanding, this activity must be explored under a variety of test conditions.

A natural question is whether these frequency shifts are the result of eye closure and not necessarily of drowsiness.  This is not inconceivable in light of the relationship between alpha activity and eye state.  In chapter 6 we will explore this issue in more detail to determine the dependency of the high frequency activity on the state of the eyes.

In this chapter, we have examined the higher sampling rate used in the visual stimulus experiment and some of the implications related to processing the EEG signals.  We examined why other researchers used high sampling rates and how the additional information was used for different reasons (improved signal bandwidth for capturing transients and improved time domain analysis such as evoked potentials, this work was not directed at looking for new and higher frequency sustained activity).  Methods of data handling and extraction were covered, along with preliminary data analysis methods.  We explored how the data collection methods used in this work change the signal characteristics of the data being analyzed and looked at the various implications of these changes.

Small changes in signal power at higher-than-usual frequencies were observed as a direct result of the signal processing methodology used in this work.  Also, order-of-magnitude changes in high frequency energy between good test performance and failures on selected data over multiple tests and test subjects were discovered.

Spectrograms were introduced and applied to much larger data segments (from 2 second clips to 10 minute segments).  Image enhancement techniques were used to enhance the visual information content of the images to reveal that the changes previously observed in smaller data segments also exist in the larger transitional segments.  We visually observed that power drops from the highest frequencies downward as failures increase in duration and returns to normal levels after performance returns to normal.

The object of this research is to develop a drowsiness detection and tracking system.  Up to this point, we've reviewed the laboratory experiment and data collection methods.  Preliminary analysis of the data reveals a relationship between changes in spectral content of the EEG signal in frequencies well above those typically studied in the EEG literature, and these newly discovered frequencies have a direct correlation with task performance.  The next chapter will build upon the initial findings of this chapter to develop a preliminary algorithm which uses these higher EEG frequencies discovered for drowsiness detection.  The algorithm will provide a quantitative measure of the observed high frequency activity where, up to this point, the analysis has focused primarily on qualitative changes and their relationship to behavioral characteristics during a visual stimulus task.

<PREVIOUS

NEXT>