Learning binary code features for UAV target tracking

During target tracking, in order to obtain a higher tracking accuracy, the region we would like to track should have a good feature expression. Furthermore, we need to extract multilevel and complex features to deal with problems which are usually encountered during UAV tracking, such as the target deformation, scale change and occlusion. However, such features make tracker more complex which would seriously affect the real-time tracking. Considering the above problems, we take the advantage of random forest for features selection, and then transform the features to binary code, which can not only reduce redundancy but speed up the tracker. In order to further improve the accuracy of UAV tracking, we utilize structured SVM for online learning to distinguish object from background. In addition, we apply the scale pyramid to achieve the scale invariance of tracker, which help to obtain a more precise position of the object. We have verified the effectiveness and robustness of our method on the classical UAV object tracking dataset UAV123.