after establishing the flow the ideas in the paragraph are enhanced to stay on the topic. 

The envisioned multi-sensor tracker captures an object’s position and orientation, which is called pose. The tracker is useful to applications that need to know an object’s pose. An auto stereoscopic display is one such application. It displays 3D scenes to a viewer depending on his head pose. As the viewer moves about, the pose changes and the display changes accordingly. To enable the display to operate properly, the tracker provides an accurate pose of the viewer’s head, frequently enough to update the display. 

The challenge in designing a tracker, which outputs the object’s pose frequently and accurately, is to minimize the delay in computing the object’s pose. Computing the object’s pose involves processing the data from the tracker’s input sensors. The sensor data captures features of the object’s pose. Each sensor’s data, however, only describes a limited dimension of the object’s pose relative to it; for example, only the distance from the tracked object to the sensor is known. The tracking algorithm computes the object’s complete pose by mathematically combining the data from all sensors. The computation time has to be minimal to frequently provide the tracked object’s pose. 
 
SCAAT (Single constraint at a time) is a tracking algorithm used to compute the object’s pose from the tracker’s sensor readings. SCAAT was developed at UNC and is successfully applied in the HiBall tracker’s tracking algorithm. SCAAT’s contribution is to compute the object’s pose with every individual sensor reading, although each sensor only has partial information of the object’s complete pose. The benefit of SCAAT over other tracking algorithms is low latency to compute the object’s pose. 

SCAAT prevents the tracker from being scaled to track in larger volumes. Increasing the tracking volume requires more sensors. SCAAT requires that all the sensor readings be processed at a central processor. The processing resources at the central processor limit the maximum number of sensors supported by the tracker.

Distributing the SCAAT computation enables the tracker to scale. The computation will be performed across several processors. A processor can be associated with each of the sensors. Adding more sensors adds more processors to manage the increased computation. The parallel computation at each processor will reduce the overall latency of computing the tracked object’s pose. The processors will need to communicate through a network to exchange partial and final results. 

The concepts from directed diffusion can be used to design a network to connect the processors of the distributed SCAAT. Directed diffusion was developed at USC [itag 2000]. It is a protocol that enables small-interconnected devices to discover each other and establish a communication link. In particular, it is intended to manage network resources for a collection of sensors working to perform a common task.

The multi-sensor tracker is a concept that comes from combining the theory of SCAAT and directed diffusion. The multi-sensor tracker will consist of a network of several small sensors. The sensors acquire data of the object’s pose from reference points on the object. Each sensor is connected to an independent processing unit. The directed diffusion protocol manages the communication network connecting the processors. The distributed version of SCAAT algorithm will compute the user’s pose on the sensors’ individual processors.

The effectiveness of the multi-sensor tracker will be evaluated by its three heuristics – scalability, tracking stability, and the human effort to initialize the system. The scalability of the tracker is evaluated by how many sensors can be added to the system without exceeding the bandwidth limitation of a network link. Tracking stability requires the system to continue to accurately track even when there are obstructions to the sensors. Human effort to initialize the system should be minimal to make the system reasonable to deploy.

The multi-sensor tracker’s design specifications depend on the requirements for an application to perform properly. The auto stereoscopic display, as explained earlier, is an example application. Perlin developed a prototype of the display at New York University [Perlin 1999]. The specifications of the display can be used to solve for the accuracy and sampling frequency required of the multi-sensor tracker. 

This paper explores the design issues of the multi-sensor tracker, currently only a concept, in the following five sections. The background material in section 2 compares the multi-sensor tracker to other trackers and explains the details of SCAAT and directed diffusion. Section 3 illustrates how the tracker’s performance requirements can be obtained from an application using it, in this case, the auto stereoscopic display. A description of how the multi-sensor tracker works is in section 4. Section 5 discusses the design choices of the network using the directed diffusion protocol to provide the communication mechanism for the SCAAT tracking algorithm. The conclusion and future work are in section 6.