UAV (unmanned aerial vehicle) and EOD (explosive ordnance disposal) robot can form an air-ground system to work together to greatly improve the working efficiency and performance of the EOD robot. For this problem, a target tracking method for UAV that can be used in the air-ground system is proposed. Firstly, a visual algorithm is used to recognize and locate the ground target, and the dynamic confidence window is used to reduce the search area. For the tracking failure problem caused by the short-term disappearance of the target because of high maneuverability, the Kalman filter algorithm is used to filter the target position and predict target motion. Also an adaptive cascade PID (proportional-integral-differential) algorithm for UAV flight control is proposed to achieve real-time tracking of the dynamic target. Finally, a drone tilt compensation algorithm is designed to reduce the landing error caused by environmental interferences, and realize accurate landing of the UAV. An air-ground system consisting of a ground robot and a UAV is built for experiments. The results show that the system can complete each part of the task reliably, which verifies the effectiveness of the proposed algorithm.