Parallel Implementation Of A Target Detection And Tracking System On A Graphics Processing Unit

In these days, with spreading imaging systems, target detection and tracking become an important utility of the systems. In literature, detecting and tracking targets that have various different features in military or civilian applications are common research are of the community. The main purpose on target detecting is detecting objects that can be disassociated from background. They can be stationary or moving. After detecting targets, state vectors of the targets that can include position, velocity, magnitude etc. are predicted via tracking algorithm. In this paper, for detecting targets, salient region extracting is aimed so, Boolean Map Salient, (BMS), approach is used. For tracking targets, Bayesian approach based particle filter is used. Particle filter tries to find probability density function, (PDF), using particles with different weights. The main purpose of choosing particle filter is nonlinear behaviors can be represented. There are limited time for finding salient objects and tracking them in an embedded system. For this reason, to decrease processing time of the application, it is decided to use graphics processing unit, (GPU). It is observed that parallel implementation on a graphics processing unit can provide significant improvements on processing time.