UAV plays an important role in continuous target tracking because of its high mobility. Due to the limited resolution of the on-board camera of the UAV, the UAV needs to track the ground target in the 3D low-altitude environment. Because the target is easy to evade by using the blind area of the UAV, the UAV fails to track the evasive target. In this paper, a hybrid target tracking method combining deep reinforcement learning (DRL) and target intention inference is proposed to solve the above problem. In order to infer the intention of the target in the short and long term, a convolutional long short term memory model (CNN-LSTM) is first established. Therefore, the prediction of the target trajectory is guided by the inferred target intention, which can also effectively guide the target search process. A deep reinforcement learning (DRL) framework is used to create a target search algorithm, which is trained by interacting with the task environment to achieve optimization and iteration of the strategy. Simulation data show that the precision and stability of target tracking can be significantly improved by target intention inference. The continuous tracking algorithm has high adaptability to the 3D low altitude environment.