Abstract: While the star images were pictured by space-based platform, there was a simultaneous relative motion between the background and the camera. The moving small object cannot be obtained through simple frame difference between adjacent frames. Thus, it was difficult to had the space object inspection. Based on the analysis of star image model, proposes an image registration method via extracting feature points and then matching the triangle. Firstly, it was the pretreatment of the images, which was to had a single-frame image segmentation from the selection of optimal thresholds, in order to remove the background noise. Then, divide the stars according to area sizes. For those eligible stars, make the feature triangles, and acquire the parameters from the matching in the adjacent frames. In order to minimize the calculation, the ignorance of the background interpolation was only applied to star coordinate matrices. Finally, detect the moving trace of the object according to multiple-frame connection. As the simulation experiments indicate, in the sequence images, the method can real time inspection in a high detection rate and a low false alarm rate.