I did not have an idea for what the intro should include. Hence the ideas are muddeled.

Computer human interaction enhances the user’s computer experience. The interaction can be enhanced when the computer adapts information according to the user’s pose, which it is tracking. The distributed multi-sensor tracker is intended to be a tracker users can comfortably wear for several hours at a time while the user interacts with the computer. In virtual environments the computer system immerses the user in the virtual world by displaying the user’s point of view. CAD and computer animators can better visualize their concepts by viewing them in three dimensions rather than viewing the concepts on a two dimensional screen. Auto stereoscopic displays display in 3D based on the user’s head pose. In addition, the user would benefit from the computer system tracking them in their office and displaying useful information, such as calendar and notices, on appropriate screens depending on were the user is in office. 

Existing tracking systems, such as the Intersense [Foxlin 2000] and HiBall [UNC 2000] trackers exist but require the user to wear tedious active electronic equipment. The tracker function with beacons in the real world and the active sensors on the user detect the beacons to identify the person’s position. The electronics, even small, is heavy and unnatural for the user to carry. Instead a passive system is preferable in which the user wears, for example inactive fiducials, such as used to capture motion in the movies. The user could easily wear fiducials on the rims of their glasses all day without inconvenience. The distributed multi-sensor tracker tracks the fiducials on the user with sensors that are distributed throughout user’s room. Unique, to the multi-sensor track is that each of the sensors has its own processing unit. The tracking algorithm will process the sensor data at each sensor and then combine all the results into the user’s tracked pose, which is the user’s position and orientation.

The advantage of using simple sensors to detect the fiducials is that they would acquire less data per sample than acquired from a more complicated sensor such as a camera. Processing the sensor data will be fast as data captured is the essential information that identifies the fiducial. Less data and faster processing will give the simple sensors a higher sampling frequency and contribute to more accurately capturing the user’s motion.

The tracking system can be scalable with parallel and distributed computation of the sensor data. Although all the sensor data could be processed at a central location, such a system would not scale well as more sensors are added to the system. The additional data will load the communications network and central processor. A more scalable solution is to attach each sensor to a dedicated processing unit that acquires data, extracts the fiducial information, and updates the user’s tracking estimate. The sensor’s tracking estimate would be communicated through a communications network, such that the complete tracked pose of the user can be updated with the other sensors.

Using simple sensors with parallel and distributed computing introduces an interesting computational paradigm that will be further discussed in this paper. Concepts from SCAAT (Single Constrain At A Time) [Welch 1997] theory introduce how to compute tracking estimates with information from individual sensors. Ideas in directed diffusion [Intanagonwiwat 2000] will contribute to the design of the communication network that interconnects the multiple sensors. There are various issues involved in developing the tracking algorithm. Each sensor and processor should individually extract as much tracking information as possible. The individual tracking information needs to be combined into one tracking estimate for the user’s pose. The communication network needs to efficiently transmit the data with the least amount of latency possible. The sensor network should have enough tracking information to automatically calibrate itself with minimal human intervention. The system should be fault tolerant to sensor noise and other expected environment disturbances that prevent the tracking system from acquiring tracking information.

The distributed multi-sensor tracker is explored in the following sections. Section 2 uses the auto stereoscopic display as an example to illustrate the specifications for a tracker used in interactive computing. Section 3 describes the multi-sensor tracker, SCAAT and directed diffusion. The issues involved in the tracking algorithm are discussed in section 4. The conclusion and future work complete the discussion in section 5.