Evaluation of an OpenCL-Based FPGA Platform for Particle Filter

Abstract

Particle filter is one promising method to estimate the internal states in dynamical systems, and can be used for various applications such as visual tracking and mobile-robot localization. The major drawback of particle filter is its large computational amount, which causes long computational-time and large power-consumption. In order to solve this problem, this paper proposes an Field-Programmable Gate Array (FPGA) platform for particle filter. The platform is designed using the OpenCL-based design tool that allows users to develop using a high-level programming language based on C and to change designs easily for various applications. The implementation results demonstrate the proposed FPGA implementation is 106 times faster than the CPU one, and the power-delay product of the FPGA implementation is 1.1% of the CPU one. Moreover, implementations for three different systems are shown to demonstrate flexibility of the proposed platform.