An Improved Visual Tracking Approach Based on Hierarchical Convolutional Features

In recent years, visual tracking faces numerous challenges, and convolutional neural networks are used more and more frequently to extract features. The Hierarchical Convolutional Features method (HCF for short) is one of the classic applications of Convolutional Neural Network in correlation filter tracking algorithms. But it is a problem that the speed of HCF method is slow. To tackle this problem, this paper optimizes the model update strategy of the baseline (HCF). In order to reduce the model update frequency, we set an interval parameter, which not only saves time, but also avoids the problem of model drift and improves the tracking effects to a certain extent. The proposed method is compared with 10 excellent trackers on the OTB2013 data set. Experimental results indicate that our approach has satisfactory results. In addition, compared with baseline, the tracking speed of the proposed approach is also slightly faster.