Parallel algorithms for 2D Kalman filtering

Abstract

Methods for implementing 2-D reduced-update Kalman filtering using parallel machines are described. Various types of parallel architectures can be used for the implementation. Algorithms are described and compared for implementation on the Sequent Balance 21, a multiprocessor system with shared memory, CLIP4, a 96*96 SIMD processor array, and the Connection Machine, an SIMD array of 64 K processors. All the machines show great improvement in performance. The advantage with the Connection Machine is the individual processor's ability to write and read from different locations. This allows multiple-image filtering and a great increase in the efficiency of processor usage.<>