融合视觉显著性再检测的孪生网络无人机目标跟踪算法

针对旋翼飞行器在跟踪过程中目标尺度变化、快速运动、视角变化等问题，本文提出了一种基于MobileNetV2的孪生网络目标跟踪算法，可在无人机机载处理器上实时运行。该算法主要包含目标得分估计模块与目标尺度估计模块两个部分。结合多特征融合的策略，可准确预测出目标位置与目标框IoU，同时以目标框IoU为指导，利用梯度上升法对目标框进行迭代修正，进一步提升预测精度。针对完全遮挡而导致的目标跟丢问题，本文设计了一个基于视觉显著性的目标再检测算法，该算法可实时高效地预测出图像的显著性区域，以指导对目标的再检测，进而恢复跟踪。最后，通过标准无人机跟踪数据集测试与实际无人机跟踪实验，验证了算法的可行性。

Siamese network combined with visual saliency re-detection for UAV object tracking

Considering the problems associated with rotorcraft trackings, such as target scale variation, fast motion, and viewpoint change, this paper proposes a siamese network-based target tracking algorithm using MobileNetV2, which can run in real-time on an onboard UAV processor. The algorithm consists of target score and scale estimation modules. Combined with the multifeature fusion strategy, the target position and target box IoU were accurately predicted. At the same time, by employing the IoU, the gradient ascent method was used to iteratively modify the target bounding box to further improve the prediction accuracy. In addition, to solve the problem of target loss caused by full occlusion, a re-detection algorithm based on visual saliency detection was developed, which efficiently predicted the saliency map of the image in real-time to guide the re-detection of the target and resume tracking. Finally, the feasibility of the algorithm was verified by comparing the standard UAV tracking dataset and the actual UAV tracking experiment.