Ionic sweat is more conductive than dry skin, which causes an increase in bulk conductivity, proportional to the amount the glands have filled[Llo59]. An example of a measurement of the skin conductivity response is shown in Figure 2 where seimens is the unit of measurement for conductivity, equivalent to . The typical startle response has three components, as shown, a period of latency following the stimulus (approximately one to three seconds in duration), a rising arm, and a decay.
The magnitude of the response is related to the magnitude of the stimulus. Stronger stimuli give stronger responses[SF90]. Responses can be triggered by events including turning on a light, hearing a sound or anticipating a loss in a card game [Dam94]. However, these responses will be smaller than those caused when the wearer's survival is threatened. By setting two thresholds in the detection algorithm, StartleCam can respond differently to these two classes of events, perhaps only storing events that trigger small responses in our video archives and transmitting to our SafetyNet only those events that trigger large responses.
Although well correlated with emotional events in studies, the skin conductance response is still not entirely predictable. Even for stimuli which are supposed to be universally startling, such as white noise bursts, some subjects will show a strong response and others will habituate rapidly, not responding when the stimulus is repeated[SF90]. Examples of these reactions to the same stimulus are shown in Figure 3. The same subject may show either reaction on variables which are not fully understood, but which may depend on the time of day and their stress level. It may also happen that a wearer exhibits a startle response when there is no apparent stimulus[SF90]. Although the frequency of unstimulated responses may be an indicator of overall stress level in the wearer, they are not indicators of particularly noteworthy events. This would cause the StartleCam system to fail with a false negative result even when the detection algorithm works perfectly. In a real world application, StartleCam would detect naturally elicited responses.