When the shooting range′s optical theodolite tracks the target in real time, the camera will randomly jitter, causing the target large-scale motion in the image. In dealing with large-scale motion, the tracking method based on the search window is easy to lose the target, and the tracking method based on the full-image search is time-consuming. Considering these problems, an improved TLD(tracking-learning-detection) framework combining KCF(kernelized correlation filter) and target position prediction was proposed. An orthogonal polynomial optimal linear filter and camera angle information were utilized to predict the position of the next frame of the target, and KCF was used for fast-tracking in this area, which can improve the success rate and save the tracking time, and can detect when the tracking fails. Simulation experiments demonstrate that the optimal linear filter can accurately predict the target position and provide KCF with a more accurate search position. Besides, the algorithm consumes only 1.1ms per frame, and the positioning accuracy is better than that of TLD and KCF, which can effectively copes with camera′s jitter. The actual task verification proves that this method can improve the automatic interpretation level of the shooting range and reduce manual intervention.