PARTICLE FILTER-BASED OBJECT TRACKING USING JOINT FEATURES OF COLOR AND LOCAL BINARY PATTERN HISTOGRAM FOURIER

Abstract

Object tracking is defined as the problem of estimating object location in image sequences. In
general, the problems of object tracking in real time and complex environtment are affected by
many uncertainty. In this research we use a sequensial Monte Carlo method, known as particle
filter, to build an object tracking algorithm. Particle filter, due to its multiple hypotheses, is known
to be a robust method in object tracking task.
The performances of particle filter is defined by how the particles distributed. The role of
distribution is regulated by the system model being used. In this research, a modified system model
is proposed to manage particles distribution to achieve better performance.
Object representation also plays important role in object tracking. In this research, we combine
color histogram and texture from Local Binary Pattern Histogram Fourier (LBPHF) operator as
feature in object tracking.
Our experiments show that the proposed system model delivers a more robust tracking task,
especially for objects with sudden changes in speed and direction. The proposed joint feature is
able to capture object with changing shape and has better accuracy than single feature of color or
joint color texture from other LBP variants.