The improvement of visual object tracking using a combination between particle and correlation filter

One of the many challenging problems from computer vision is constructing robust object tracking from various problems and object conditions. Difficulties in tracking are more complex when an image sequence has many types of tracking problems such as background clutter, fast motion, occlusion, illumination variance, scale variance, motion blur, non-rigid object pattern, and another. Meanwhile, the task to build a novel approach which efficiently handle all tracking problems still in development step, especially supported by low computing time. We combine adaptive particle filters (APF) and discriminative correlation filters (DCF) to overcome more complex tracking problems so it can improve overall tracking accuracy. Our combination employs APF as the main framework and embeds DCF in it to estimate motion vector. However, this will make a double-edged knife where the utilization of two approaches together will make enlarged computing time. So, we also utilize leap-tracking as an alternative solution to reduce high computation time. Our testing uses 15 sets of benchmarks tracking datasets and 3 sets by ourselves with various tracking problems. The results show that our combination significantly improves tracking accuracy. Thus, the reducing of computational procedures efficiently will increases tracker speed without reduces accuracy and avoids identification errors.