The StartleCam System

StartleCam is a wearable computer system with a digital video camera, physiological sensors and a wireless connection to the Internet. The wearable computer used in this prototype is the MIT Media Lab's Lizzy design¹ using a 100MHz 486 processor with a two gigabyte hard drive [Sta95]. The Lizzy computer is used to run the real-time signal processing algorithm to detect the startle response, to hold a rotating buffer of video images from the digital camera, and to control the flow of video information to the hard disk or the Internet.

  

Figure: The StartleCam system consists of a skin conductivity sensor (GSR) which is sampled by an analog to digital converter attached to a wearable computer. A digital camera and digital modem are also attached to the computer. Images are captured by the digital camera and stored in a buffer in memory. When the computer algorithm detects a startle response, the buffer of images is downloaded and transmitted wirelessly back to the Internet. Figure (a) shows the details of the system and Figure (b) shows the system as worn with skin conductivity sensors on the hand (Alternatively, the skin conductance sensors can be placed on the foot.)

The digital video camera shown here is a black and white Connectix QuickCam, which can capture up to 15 frames per second of 320 by 240 pixel images in 6-bit grey. The camera has a fixed focus, from eighteen inches to infinity, and automatic white-balance calibration. A compression script stores images from the QuickCam in a JPEG format, which requires only 1-3 kilobytes of memory per image.

Images from the camera are stored in a virtual buffer until the StartleCam detection algorithm is triggered. This system maintains a five-image buffer of images recorded at one frame per second. When a trigger is detected the entire buffer is saved as a single image and can be both downloaded to the hard drive or sent over the Internet to a remote server.

To transmit information back to the Internet, two options are available with this system: a two megabits per second wireless ethernet connection using Digital's RoamAbout/DS and a 14.4 kbps CDPD connection using a Sierra wireless modem. The RoamAbout connection is preferred, but it is unfortunately limited to indoor applications. The slower CDPD connection is necessary for outdoor monitoring.

The skin conductance is measured by applying a small voltage to two electrodes and measuring the resultant current conduction of the skin. This analog signal is then sampled at 20 samples per second using the ProComp system from Thought Technologies. Although the electrodes shown here are placed on the index and middle fingers of the hand (as shown in Figure 1) the eccrine sweat glands that produce this response cover most of the body, allowing sensors to be located comfortably in a number of places. The palms and soles are preferred only because they have the highest concentration of eccrine glands[SF90]. Electrodes measuring these areas could be worn as a pair of rings or embedded into the insoles in shoes [PH97].

  

Figure: The matched filter is derived from a typical skin conductivity response. Figure (a) shows a response to an audio stimulus. The three phases of the response, a latency following the onset of the stimulus, a sharp rise and a consequent decay are labeled. Figure (b) shows the filter, a time reversed version of the rising edge of the response.