Coordination of the observers

In general, the coordination of the observers is a complex task since it is composed of at least the following two contradictory requirements: (i) precise position determination for each tracked object, and (ii), monitoring the empty parts of the scene for appearance of new objects. The optimal coordination strategy is necessarily application specific, since it depends on many parameters such as the counts of observers and objects of interest, the priority of individual objects, whether all observers can ``see'' the entire scene, etc. The following terms may prove useful in the design of a strategy:

• an object tracked by exactly one observer is defined as weak;
• an observer tracking a weak object is defined as bound;
• an observer tracking an object which is tracked by exactly one additional observer is defined as important;
• an observer tracking 0 objects is defined as idle;
• an observer which is neither important nor bound nor idle, is defined as free.

It seems that the real time performance will remain the most challenging requirement for quite some time, so that the communication protocol between the coordinator and the observers should not prescribe waiting for confirmation messages. This can be achieved by scheduling the activation of the coordination procedure in regular time intervals (e.g. 2 seconds). In such an arrangement, the procedure in each invocation examines the situation on the blackboard, issues one or more control messages to the observers, and returns to the blackboard control component (no answer from the observer is required). The following minimalistic strategy has been devised for robust (although suboptimal) coordination in partially occluded scenes containing a small number of objects.

1. observers in seeking operating mode, bound, important, free and idle observers are assigned priorities of 0, 0, 1, 2 and 3, respectively;
2. if there are no observers with a non-zero priority, no action is performed;
3. otherwise, the highest ranked observer (round robin scheme is used to choose among observers with the same priority) is chosen and is denoted as $O_C$;
4. if there is a weak object $A$ positioned outside the field of view of $O_C$ (otherwise, $A$ is occluded from $O_C$), $O_C$ is assigned the tracking of $A$;
5. otherwise, only if $O_C$ is not important, it is switched to the seeking mode.