Rotation invariant multi-model scene matching method based on spatial-temporal correlation

In this paper, a rotation invariant multi-model scene matching method is proposed for scene matching aided navigation system. Phase congruency transformation is introduced first to minimize the image difference between multi-model images. Then ring project transformation is processed to make the method invariant to rotation. However, multiple maximum phenomenon is likely to occur after ring project transformation. To solve this problem, a multi-frame spatial-temporal correlation matching method is proposed. Using the spatial-temporal correlation gained from the inertia system or matching of the adjacent interframes, an optimal matching position is gained by maximizing a multi-correlation surface. Afterward, surface fitting method is used to get sub-pixel accuracy matching position. This method, which is invariant to rotation, greatly increases match accuracy. Necessary simulation proves the efficiency of our method.