As the Harris detector can produce false and unstable corners, and obtained matching points have different accuracies in aerial video registration, an aerial video registration algorithm using optimal gradient filters and projective invariant was proposed. Firstly, a Harris detector based on optimal gradient filters was presented to determine the location of corners, and the locally most stable points were selected to be matching points. Then, the Delaunay triangulation was used to perform the initial matching. Finally, the most "useful" matching points that best satisfied the cross-ratio invariant were presented to estimate the geometry transformation and finish the image registration . Experiment results show that the proposed method by the optimal gradient filters and cross-ratio invariant can realize the registration between the frames, and the average geometric fidelity error is 0869 for 8 sets of the unmanned video sequences with a resolution of 320 pixel×240 pixel. The method can capture moving objects effectively.