Abstract: UAVs (unmanned aerial vehicles)frequently lose sight of their targets in complex low-altitude environments due to obstacle occlusion. The primary factors are analyzed. In the perception phase, the target may be occluded, hindering the UAV's ability to accurately identify it. In the tracking phase, the visual field of one single UAV is limited and the observation angle can't be adjusted swiftly. To achieve efficient and stable target tracking, solutions in response to the challenges are proposed. For occluded target detection, a lightweight target tracking method based on Transformer networks is proposed. This method dissolves the redundant decoder structure in the traditional Transformer, and builds a pure encoder vision model, to realize the real-time tracking of the occluded target. For target tracking control, a multi-UAV cooperative tracking method is adopted. This strategy dynamically plans the trajectories of several UAVs to observe the target in multiple viewing angles. The simulation test results of occluded target detection under a fixed viewing angle show that the proposed target detection method can stably detect the target with an occlusion rate exceeding 90%. Compared with the DiMP18 and E.T.Tracker algorithms with good real-time performance in the dataset and simulation experiments, it is found that the FPS (frame per second)of the proposed algorithm is approximately twice that of the other algorithms, and the accuracy is not much different. The low-altitude flight test over outdoor dense jungle verifies that the proposed method can detect and stably track the target in real time by onboard equipments. In addition, three UAVs are controlled by the proposed multi-UAV cooperative tracking method to dynamically cover the visible area around the target in the simulation and flight experiments in a multi-obstacle environment, achieving better tracking stability than a single UAV. The proposed framework of multi-UAV cooperative target detection, positioning and tracking is applicable to low-altitude and multi-obstacle environments, and solves the problems of detection failure and tracking loss caused by target occlusion, demonstrating significant superiority in both real-time performance and tracking stability.