Support Vector Correlation Filter with Long-Term Tracking

Zhongpei Wang,Hao Wang,Baofu Fang,Chengjun Xie

DOI: https://doi.org/10.1007/s11760-018-1310-0

2018-01-01

Abstract:Boosted by the promising advancement of the correlation filter-based tracker, we propose an algorithm called the SLT (support vector correlation filter with long-term tracking) that is based on the new SCF (support vector correlation filter) framework to handle long-term tracking. To perform long-term tracking, we propose using a detector to refine the position that includes occlusion and deformation and is out-of-view. We used a new judgment criterion called the max response to the average response rate (MAR) to activate the re-detection procedure and then exploit the linear support vector machine (SVM) classifier to obtain a positive refinement. Moreover, we do not update the SVM classifier every frame to reduce the number of computations and obtain better samples to improve the accuracy of the classifier. We use the online passive–aggressive learning algorithm for online learning and use the same MAR criterion to active it. Extensive experimental results on the OTB50 benchmark dataset show its superior performance in terms of accuracy and robustness.