Object Tracking Using Particle Filter Based on Color Correlogram

Abstract

This paper proposes an approach for object tracking using particle filter based on an improved color correlogram. The improved color correlogram for representation object contains not only color information but also spatial information, which makes feature more distinctive. Instead of using the whole correlogram matrix as feature, we construct feature vector based on elements of the upper triangular matrix, which easily computes similarity of features and reduces memory consumption and computational complexity. Target is divided into non-overlapping parts, which improve robustness of the feature by incorporating global and local target information in a single mode. Experiments show that the approach achieves good performances in real scenes.