In order to adapt to complex environments and changes of the fighters own stateand improve the performance of the target tracking algorithmthe original Tracking-Learning-Detection (TLD) algorithm is improved.Firstlythe improved Scale-Invariant Feature Transform (SIFT) feature matching algorithm is combined with the optical flow method in the original TLD algorithmand one of the two algorithms is selected adaptively according to different states of the fighter and the surrounding environmentso as to improve the robustness of the algorithm and its adaptability to complex environments.Secondlyparticle filter is introduced for optimization to predict the state of the fighter in real time and narrow the detection rangewhich can reduce the complexity of calculation and improve tracking speed and continuity.Finallyconfidence detection is set up to avoid misjudgment and improve tracking accuracy.The experimental results show that the improved algorithm has better tracking performance than the reference algorithm and can accomplish the task of long-time robust tracking when the fighter is in a complex scene or when its state is changing greatly.