Illumination-Adapted Long-Term Tracking Based on Multiple Correlation Framework

Object tracking is a hot field in image processing, among which the kernel correlation filtering algorithm is widely applied based on the advantages of precision and speed. However, due to information decay, it has poor performance in the long-term tracking. In this paper, we develop an illumination-adapted long-term tracking method based on correlation filtering tracker. Firstly, we introduce illumination adaptive normalization and information entropy weighted feature fusion to improve the prediction accuracy. Secondly, we estimate the object scale quickly based on the binary strategy to improve the speed of the algorithm. Then, we design re-detection based on online SVM detection for target loss. Finally, we design an appropriate updating mechanism to combine the whole tracking framework with various judgment thresholds. The experiment on the data set shows that all of our modules are effective and have great improvement as compared with other correlation filter type trackers.