An anti-occlusion tracking system for UAV imagery based on Discriminative Scale Space Tracker and Optical Flow

Abstract

In this paper, an anti-occlusion tracking system is proposed for Unmanned Aerial Vehicle (UAV) images. Innovatively combining the efficient Discriminative Scale Space Tracker (DSST) and optical flow, the system can realize accurate and successive tracking in long-range UAV images, providing a foundation for unmanned autonomous system. The DSST contains two correlation filters, i.e., the scale filter and the translation filter to achieve satisfying tracking performance in terms of speed and accuracy when compared with the recently introduced tracking-by-detection strategy which relays on deep learning model of the convolutional neural network. Optical flow including the motion information is used to extract regions of interests (ROIs) which will be matched with a template to complete the anti-occlusion tracking. Field experiments in different backgrounds and various occlusions are conducted to verify that the proposed system integrating DSST and optical flow can track target robustly, especially for the occluded moving target.