基于重要特征的视觉目标跟踪可迁移黑盒攻击方法

Black-box attack methods for video object tracking have received increasing attention in order to evaluate the robustness of object trackers and thus improve the security of trackers. Most of the current researches are based on que⁃ ry-based black-box attacks. Although fairly good attack effects are achieved, a large number of queries still cannot be ob⁃ tained for attack in practical application. We propose a transfer based black-box attack method, which attacks the important features in the features that are highly related to the tracking target and are not affected by the source model, reduceing their importance and enhancing the unimportant features to realize the transferable attack. Specifically, the corresponding gradi⁃ ent is obtained by back propagation to reflect the importance of its features, and then the weighted feature obtained by the gradient is used to attack. In addition, this paper uses the temporal information of similarity between adjacent video frames to propose a sequential-aware feature similarity attack method to attack the object tracker by reducing the feature similarity be⁃ tween adjacent frames. This paper evaluates the proposed attack method on the current mainstream deep learning target track⁃ er. The experimental results on multiple datasets prove the effectiveness and strong mobility of this method. In OTB bench⁃ mark, the tracking success rate and accuracy of SiamRPN tracking model are reduced by 71.5% and 79.9%, respectively.