Parallelized Particle Filter With Efficient Pipelining on FPGA for Real-Time Ballistic Target Tracking

Abstract

In this paper, the problem of a real-time ballistic target tracking is addressed. Because the tracking process can be affected by the uncertainty in models and disturbances, probabilistic filters have been applied to solving the ballistic target tracking problem. Particle filters (PFs) can be used as a probabilistic filter because of their advantages of dealing with non-Gaussian models. However, since the PFs require too much computational costs, it is hard to apply the PFs to real-time systems for ballistic target tracking. The Graphic Processor Unit (GPU) can be considered to accelerate the PFs in the parallelization manner. But, the GPU-based parallelization requires too much power consumption, which means that it is difficult to be applied to the ballistic target tracking system due to the heat problem caused by the excessively increased power consumption. This paper proposes a practical approach which is a parallelized particle filter based on a heterogeneous processing system including a field-programmable gate array (FPGA) to accelerate the ballistic target tracking process by parallelizing the PFs while avoiding the heat problem. The evaluation results with a real heterogeneous processing system showed that the proposed approach could successfully conduct the ballistic target tracking and accelerate the PFs over four times.