Visual tracking using Locality-constrained Linear Coding and saliency map for visible light and infrared image sequences

摘要

As a development of sparse coding, while retaining the advantage of sparse coding in classification, Locality constrained Linear Coding(LLC) greatly improves the time efficiency of appearance modeling. However, in order to further promote the performance of real-time and develop a tracking algorithm that can be applied to both visible light images and infrared images, this paper proposes a tracking algorithm using LLC and saliency map under the framework of particle filtering. It is universally acknowledged that number of particles determines the accuracy of tracker under the framework of particle filtering. Unfortunately, the increase in the number of particles leads to the augment of computational burden. Therefore, the basic idea of the proposed algorithm is to reduce the computational number of observation vectors while keeping the effective number of particles and achieve the goal of strengthening the real-time performance of tracker. The proposed algorithm firstly uses spectral residual to obtain a saliency map of the current frame and then computes the saliency score of each particle. Secondly, several particles are eliminated directly according to the difference between the saliency score of the particle in the current frame and the target score in the previous frame. Thirdly, LLC is used to compute the observation vector for the rest particles and complete tracking tasks. Both quantitative and qualitative experimental results demonstrate that the proposed algorithm performs favorably against the nine state-of-the-art trackers on twelve challenging test sequences including six visible light sequences and six infrared sequences. In addition, related experimental results reveal that the proposed algorithm decreases the computational complexity and has the better tracking performance compared with the tracker just using LLC in the framework of particle filtering.