Modification of Tracking Algorithm Struck by the Application of the Scale Estimation Method

Abstract

The article considers work of visual object tracking algorithm “Struck”, scales estimation algorithm based on the correlation filter and their cooperative work. The results of the experiments showed that the combination of these algorithms allows achievement of gain in precision of visual object tracking in different variations of sizes of tracked object, compared to the original version of “Struck”. The average gain is equal from 30% and depends on the number of scales in tested video sequences. The features of these sequences: increasing and decreasing of object sizes in 3 - 30 times. The localization accuracy of the object degrades the accuracy of the estimation of the object size. Performance is decreased by 5 - 8 frames per second on averages in comparison to the original algorithm of the “Struck”. However, our proposed method provides real-time working.