With the increasing amounts of UAVs usage, the supervision of unmanned aerial vehicles (UAV) has become particularly important, and the demand for detecting and following UAVs has grown rapidly. Compared with ground targets, UAVs are more difficult to track because of the high speed of the target and the interference caused by the shadow of either a target or a tracker. In addition, the problem of how to research the target when the target leaves the camera’s field of view has not received sufficient attention. In this paper, a shadow recognition algorithm and the detection network of a target based on deep learning are combined to eliminate the interference caused by shadows. Fuzzy control is applied in the process of following and the dynamic characteristics of UAV are considered in obstacle avoidance, which ensures the stability of the UAV for tracking. Finally, a spatial probability distribution algorithm based on Bayesian prediction is proposed for re-searching a lost target, which can rediscover a target after that target is lost. For this work, a UAV experimental platform has been built and the algorithm feasibility is verified through both simulation and a physical experiment.

随着无人机使用量的不断增加，对无人机的监管变得尤为重要，对无人机检测和跟踪的需求快速增长。与地面目标相比，无人机由于目标速度快，且受到目标或跟踪器阴影的干扰，跟踪难度更大。此外，当目标离开相机视场时如何研究目标的问题还没有得到足够的重视。本文将阴影识别算法与基于深度学习的目标检测网络相结合，消除阴影带来的干扰。跟随过程中采用模糊控制，避障时考虑了无人机的动态特性，保证了无人机跟踪的稳定性。最后，提出了一种基于贝叶斯预测的空间概率分布算法，用于重新搜索丢失的目标，可以在目标丢失后重新发现目标。本工作搭建了无人机实验平台，并通过仿真和物理实验验证了算法的可行性。