The space-based non-cooperative target tracking technology could play a crucial part in many aspects of the space application. Currently most well-behaved algorithms were based on video stream. Owing to different working conditions, their tracking precision, speed of operation, warning rate and false alarm rate were dissatisfied with the requirement of space-based satellite tracking systems and missions. Furthermore, the video stream tracking algorithms were too complicated for space-based conditions, where the processors were weaker than those on the ground. To solve these problem, an algorithm based on image correlation, curve fitting, Kalman filter, and SURF algorithm and combined with prediction, tracking and rectification systems was proposed. The algorithm could achieve high speed and high accuracy and be satisfied with the space-based computing environment. Proved by the image simulation experiment and semi-physics simulation, this algorithm could continuously track the target rotated in in-plane arbitrary angle, scaled from 0.4 to 2.1, and handle illumination change. The mean error of image tracking simulation experiment result was lower than 0.9 pixel and frame rate was more than 200 frames per second under most conditions. This algorithm could also deal with image blur, Gaussian noise and salt and pepper noise. Satellite model tracking experiment results showed that the algorithm also had a stably tracking performance for practical satellite model.