Accelerating the Improved Version of Particle Filter Object Tracking Algorithm by Graphic Processing Unit

Particle filter is one of the robust algorithms in the object tracking that tracks moving objects correctly even in the complex scenes, occlusion and can work with both linear and non -linear and Gaussian and non-Gaussian distributions. Despite of its ability in tracking, this algorithm is very slow and it’s not suitable in the real-time applications. One of the strategies to increase the speed of Particle filter is parallelization by GPU. In this paper, an improved model of particle filter object tracking based on processing of important particles has been parallelized on the GPU. In the improved particle filter algorithm, unimportant particles has been eliminated, so the number of particles has decreased. Although this model of improvement has increased the speed of algorithm by 39%, it’s not sufficient in the real-time applications. Based on determined results, this algorithm when is implemented by GPU is more than 6 times faster than its CPU implementation.

Keywords:

Object Tracking , Particle Filter , Parallel Processing , Graphic Processing Unit