To infer a relationship between features of the physiological signals and the affective state of the driver, information about the context of the driving situation was collected and synchronized with the signals from the physiological sensors. Several methods were considered for this task. The idea of using an observer was initially rejected because it was predicted that this would create confounding social interactions and inhibit natural expression of emotion toward driving events. However, pilot studies were conducted both with and without observers as passengers, considering the value of the observer's annotations. The method of directly prompting a response from the subject using either a keypad or audio recording diary was not used because it might have distracting the driver's attention at a critical time and endangered their safety.
In the first study, the context of the driver's state was captured using a digital camera focused on the driver and an audio recording. The physiological data was correlated with the digital images from the camera by a time stamp from the wearable computer. The audio tape was coded and correlated separately. Problems with this method of capturing context included the poor quality of the digital images (shown in Figure 1), the lack of documentation of driving events and the ambiguity of the driver's commentary.
In the second study, more reliable recording methods were used to capture context. To capture facial expression, a small video camera was mounted on the steering column, behind the release for the airbag, as shown in Figure 3. A second camera was placed on the center of the dashboard pointing out towards the road, using a .42 wide angle lens to capture road and traffic conditions (as shown in Figure 4). Also, an observer rode as a passenger to record driving events. The video signals from the cameras were synchronized with the video signal from the physiological monitoring system using a quad splitter, as shown in Figure 2. This allows for an unambiguous record of physiological response to driving events. A fourth input to this system can be used either to display annotations from the observer or to show input from a video camera capturing the road behind the car.