无人机保距跟踪中的视觉跟踪算法研究

针对固定翼无人机对地面目标进行保距跟踪过程中不能稳定获取目标图像的问题,设计了一种提高视觉目标跟踪稳定性的算法。该算法基于核相关滤波算法,提出了线性旋转子空间的概念,用于估计平面外旋转后的目标图像,在跟踪的过程中通过跟踪效果判断是否对线性旋转子空间进行校正。这种更新机制提高了在相对位置不断变化的情况下视觉跟踪的稳定性和准确性,有效地降低了跟踪漂移的程度。算法在无人机跟踪视角的视觉跟踪数据集中进行了测试,结果显示在跟踪的准确性和鲁棒性上明显好于当前主流跟踪算法。并使用固定翼无人机进行了实机飞行,验证了算法的可行性。

Visual Object Tracking Method for UAV Stand-off Target Tracking

To obtain the target image stably in the standoff tracking of ground target using fixed-wing UAV,a visual tracking algorithm that can reliably capture the target image is designed.Based on kernel correlation filters,linear rotate subspace is proposed,used for estimating the out-of-plane rotated image.During the tracking process,tracking performance assessment is used to determine whether to update the linear rotate subspace or not.In this way,the robustness and accuracy of visual object tracking is improved when the perspective is changing,and the tracking drift is reduced.The algorithm is tested on visual tracking dataset in drone perspective,the result shows that the tracking accuracy and robustness exceed the state-of-art algorithm.Fixed-wing drones is used for actual flight,verifying the feasibility of the algorithm.