Feature-Based Displacement Field Estimation for Visual Tracking By Using Coarse-to-Fine Block Matching

Abstract

Visual tracking is very important for many applications in computer vision, such as traffic control, intelligent surveillance, industry automation, human computer interaction, etc. In this paper, we propose a hierarchical framework for estimating displacement vectors at feature points with the coarse-to-fine block matching method. Feature points are chosen to be the image locations having large curvature, i.e., the corners. First, we calculate the pyramid for each image. At the toppest layer of the pyramid, corners and their displacement vectors are estimated with the block matching method. These displacement vectors are used to predict and narrow down the search region of the following lower layer of the pyramid. Early jump-out technique is also used to enhance the efficiency of block matching. Reliable displacement fields are obtained efficiently at the lowest layer of the image pyramid.