The Researchof 3D Terrain Matching And Scene Matching

The concept of 3D terrain matching is proposed in the 70 s in the last century. Then it got the attention of the scholars who conduct the thorough research. So a rapid development of terrain matching has got. Terrain matching is widely used not only in the initial theory but also in application gradually, especially in national defense. Terrain matching is widely used. Especially in aided navigation system, terrain matching is essential in modern combat system.This paper summarizes the terrain matching research status at home and abroad. In view of the existing terrain matching difficult problem, the paper puts forward a new kind of terrain matching algorithm. Firstly, we got the terrain orthographic projection image by the study of the orthographic projection of the terrain. Secondly obtaining projection transformation parameters from projection matching between the projection and the projection, the parameters correspond to 3D terrain. Finally we completed terrain matching. Experimental results have shown that the algorithm is very effective, and strong adaptability.In reality, we often get images from different sensors whose structure and imaging principle is different which lead to large difference in image grayscale contrast.Homologous of traditional image matching algorithm is no longer optimal matching between the different from different source images. Although there have been many scholars who have put forward many multi-source image matching algorithms, but these algorithms can only suitable for special occasions. If it is a good way to solve the problem of image matching, especially the problem of multi-sensor images matching, we can provide a new and reliable theory and matching method for remote sensing, navigation,guidance, medical diagnosis, computer vision, and other fields. Then we can make good use of the technique.A novel multi-source image matching strategy based on edge feature extraction algorithm was proposed against to the shortcoming of point feature matching algorithm. Firstly,detecting the image edge points and connect into curves, multi-scale strategy is utilized to extract the edge line segment. Then, based on spatial proximity and spatial proximity, edge line segment would be divided into lots of groups which is regarded as a feature of line called line labels. Finally, the similarity between the lines can be computed according to the line geometry. Consequently, it can verdict whether the two line labels is similar and then fast matching algorithm can be used for matching. Similar to local feature, line feature is robust and adapted to the angle of view changes. Experimental results indicate that our algorithm is scale invariant, rotate invariant and illumination invariant. The algorithm can solve the situation that point feature matching algorithm is not available and is very effective to a certain extent.Based on the situation need real-time positioning, this paper proposes a fast matching strategy. Firstly, we extract the feature of the two scene images. Using the method of image retrieval, the most similar small area between the big scene image and target image can be first roughly determined. Thus we can find out the transform relationship between the big scene image and the target image, so as to complete the accurate positioning of the target scene. Experiments show that the method has more obvious advantages than SURF algorithm whose matching time is far less than SURF algorithm. The algorithm is very effective in some scene images.