Clustering algorithm for transputer implementation of a target tracker

In multiple target tracking (MTT), track splitting algorithms are computationally intensive because several filters have to be run when branching occurs, and the outputs of these filters have to be compared for similarity calculations. Within a transputer network implementation of a tracking algorithm, each processor performs these calculations making it necessary to circulate track information around the network. During this pipeline-like process, the similarity calculations across the transputers tend to produce a system bottleneck. The author overcomes this problem by presenting a clustering algorithm which avoids the need for similarity calculations between transputers