Video Object Tracking Via Visual Prior and Context Information
GUAN Hao1，XUE Xiang-yang1，AN Zhi-yong1,2
1（School of Computer Science, Fudan University,Shanghai 200433,China）2（School of Computer Science and Technology,Shandong Institute of Business and Technology,Yantai 264005,China）
Abstract：Due to the simplicity of features and lack of background information,traditional tracking algorithms are less robust to noises. This paper proposes a tracking method which integrates visual prior and background information to tackle the problems. General visual prior is got through unsupervised learning off-line on large labeled image dataset and stored as visual prior dictionary. During on-line tracking,the learned visual prior can be used as convolutional filters to extract appearance features of the object being tracked and its local background. Correlation filter is used to predict the new location of the target. Experimental results show that the proposed tracking algorithm is both accurate and robust to deal with illumination changes and partial occlusions.
2016, Vol. 37 
