Apparatus

Sensing for the Patrol task is performed by two hat-mounted cameras (Figure 2). The larger of the two cameras points downwards to watch the hands and body. The smaller points forward to observe what the user sees. Each camera is fitted with the widest angle lens available. Figure 3 shows sample images from the hat. Both cameras require an attached ``camera control box'' placed in a backpack with a video mixer and a Hi-8 camcorder. The video mixer combines up to four channels of video into one NTSC signal by mapping each channel into one of four quadrants of the screen. The output is then recorded by the Hi-8 camcorder. While it is possible to provide enough on-body computation to run feature detection in real-time, the reference video tape is needed for experimental purposes. The resultant backpack is larger than is desirable for a daily-use wearable computer but allows enough maneuverability for the two to three hours of a Patrol session.

 

  

Figure: The two camera Patrol hat.

  

Figure: The downward- and forward-looking Patrol views.

An extended version of the apparatus is currently being tested. In this system, an identical hat is handed to a team-mate or opposing player. The two channels of additional video are sent wirelessly to the first player where they are recorded using the spare channels of the video mixer. Both instrumented players are fitted with noise-cancelling close-fitting microphones to provide two channels of audio. Eventually, this second perspective may provide interesting data on player interaction, but currently the harsh RF environment causes a significant amount of noise in the system.