A real-time object tracking algorithm based on dual feature model kernel correlation filtering

Abstract

Existing target tracking algorithms are mainly divided into two categories: (1) deep learning-based approaches that have high accuracy and low speed; (2) correlation filtering-based approaches that have high speed and low accuracy. In order to balance the real-time performance and accuracy, this paper improves the KCF algorithm. The main innovation lies in the development of an adaptive dual feature model: the main feature model uses shallow texture features (Histogram of Oriented Gradients, HOG), and the auxiliary feature model uses features from convolutional neural networks (CNNs) that contain deep semantic information. These work together in an optimal manner to produce a more stable correlation filter. In order to improve the speed of the algorithm, we use principal component analysis to reduce the dimensionality of the high-dimensional CNN features. In addition, this paper also improves the accuracy of the tracking algorithm by means of scale optimization and optimization of the solution method. Our algorithm is compared with current advanced tracking algorithms with real-time speed, such as SiamFC, MEEM, SAMF, DSST, KCF, Struck, and TLD. The OPE result using the public data set OTB-2013 shows that the algorithm in this paper ranks first in the distance precision rate. Compared with the KCF algorithm, the distance precision rate and overlap success rate are improved by 25.9% and 23.2%, respectively, and the average speed of the proposed algorithm can reach 38 FPS.