A basic problem in radar technology is that of tracking multiple targets in a radar scan. An unknown number of measurements are received at each radar scan, and each measurement must be associated with an existing or new target. In a cluttered environment or when targets are close, more than one measurement may fall into a track gate. This paper provides a clear introduction to one solution to this problem--a “track splitting” algorithm, in which a track is split into n tracks corresponding to the n measurements inside a track gate. A multiprocessor implementation is used for the many Kalman filters that are run when branching occurs.
The transputer is a single-chip processor with multiple bidirectional communication paths. The authors show how a network of transputers can implement a reduced form of the tracking algorithm. A somewhat less than linear speedup is obtained as the number of transputers is increased. Performance is improved by decoupling tasks on the different processors.
The authors briefly mention their recent work. It involves the tracking of more targets, missing measurements, and different target start times. The unsolved problem in radar target tracking is that as more ambitious objectives are addressed in the form of higher densities of detections, sooner or later the number of incorrect associations becomes overwhelming.