Implementation analysis for a hybrid particle filter on an FPGA based smart camera

Abstract

Design and development of embedded devices which perform computer vision related task presents many challenges, many of which stem from attempting to fit the complexity of many higher level vision algorithms into the constraints presented by programmable embedded devices. In this paper, we follow a simulation-based methodology in order to develop an architecture which will allow us to implement a mixed Particle Filter/Markov Chain Monte Carlo tracking algorithm in an FPGA-based smart camera, using tools such as SystemC and Transaction LevelModeling (TLM). Use of these tools has allowed us to make some preliminary predictions as to the memory usage and performance of the system, which will be compared to the results of more detailed simulations obtained in the way towards implementing this system.