Non-symmetry and Anti-packing Pattern Representation Model in Visual Tracking

Abstract

Based on the non-symmetry and anti-packing pattern representation model (NAM), we propose a new division model in object tracking, whose most important characteristic is asymmetry. For the target region, we divide it into homogeneous blocks of different sizes and irregular positions. Pixels in the same homogeneous block have similar values, so we can use mean value and area to represent the feature of this block. When tracking the object, the proposed algorithm updates the mean values of each block to adapt to the change of the object. The characteristics of the whole target consist of the feature information and the position information of each block. Compared with the traditional division models, the NAM has fewer divided blocks but more highlight target structure characteristics. The experimental results in this paper show that our proposed tracking algorithm based on the NAM is better than those based on the quadtree or the compression tracking (CT) in both accuracy and speed.