Research On SAR Image Match Algorithms For The Integrated Navigation System

In the scene matching navigation system, image matching is the core technology. With the properties of all time,all weather and strong penetration ability, SAR is able to overcome the dificiencies of the optical scene matching system, such as hard to obtain accurate topographic information continuously, and vulnerable to the impact of phenology and sky phenomena, thus the current scene matching system can be improved, and SAR has become one of the current research focus.This paper focuses on the SAR image matching algorithm in the integrated navigation system, the main research and innovations are as follows:(1)In the integrated navigation system, INS heading error will result in rotation error between the reference image and the real-time image. Hu moment invariant with its RTS invariability is selected as the similarity measure in the matching algorithm to overcome rotation error.(2)A matching algorithm for SAR images based on Canny edge detection and Hu moment invariants is proposed in this thesis. Matching method based on edge features is selected to overcome the gray level differences. When Hu moment invariants are choosen as the similarity measure, matching under arbitrary rotation angle is feasible, then the Gaussian pyramid is used to improve the performance of algorithm. Experimental results show that the algorithm can obtain correct matching results even under the condition of arbitray rotation error. But rotation errors will result in small differences between the edge extraction results, thus there will be slight errors in the matching results. In addition, when there is partical occlusion in the real-time images, no correct matching results can be achieved.(3)To improve the above algorithm with low resolution which is not feasible under the partical shading condition, a new matching algorithm for SAR image based on relative point moment invariants is proposed. Feature point coordinates of high precision can be achieved by the Harris corner point operator and the sub-pixel positioning algorithm .When Hu moment invariants with their properties of RTS invariability are combined with high precision feature points, we proposed the concept of relative point moment invariants for the first time. The relative point moment invariants also have the RTS invariability properties,therefore image matching of arbitrary rotation angle can be achieved by using the relative point moment invariants. Moreover, this matching algorithm can resist certain degree of shading. Match points witch error are excluded according to the principle of similar triangles. Finally, the optimal estimation is given by the least square method.Experimental results show that, the algorithm proposed by this thesis satisfies the high-precision, real-time and certain degree of anti-interference requirements.