Real time Object Tracking based on Local Texture Feature with Correlation Filter
Abstract
Object tracking is a fundamental problem in computer vision, therefore attracts many researchers and there has been many influential algorithms in the list few years. A good feature descriptor is half of the success. In this paper, we propose an effective tracking algorithm based on texture feature with correlation filter. We improve the tracking precision by Haar-like feature and speed it up via discrete Fourier transform. The feature which is a good representation of the target can be obtained via dense sampling. What is more important is that the feature containing a great deal of information of the target appears only in the form of low-dimension. With the property of circluant matrix, we can translate the processes of training and testing into Fourier domain, which can cut the computation and reduce the time complexity. Our algorithm is proved to have better performance especially in deformation and rotation in the benchmark datasets.