Robust Object Tracking Based on Complementary Feature Fusion and Channel Reliability

Abstract

In this letter, a robust object tracking method is proposed, which is divided into two parts: robust tracker based on complementary feature fusion with channel reliability, and reliable validator based on convolutional neural network. Compared with the existing tracking algorithms, our algorithm tracks the target more robustly and accurately, and suppresses template drift problems effectively. Experimental results on the standard benchmark show that our algorithm can compete with advanced algorithms and achieve a better level.