CN105005993B - A kind of quick fine matching method of dimensional topography based on isomery projection - Google Patents

Abstract

The invention discloses a fast and accurate three-dimensional terrain matching method based on heterogeneous projection. The three-dimensional terrain matching method includes acquisition and conversion of three-dimensional DEM (Digital Elevation Model) terrain data, performing orthographic projection on the three-dimensional terrain, and performing projection map matching according to the orthographic projection of the terrain. In the process of matching the projection image, the method of combining the straight line feature and the point feature is adopted. After the line detection is performed on the projection image and the straight line with the same name is matched, the virtual corner point where the two straight lines with the same name intersect is found, and the coordinates of the corner point are calculated. . Then use the improved SURF (Speed-up robust features) algorithm to match the found virtual corners. The improved SURF algorithm combines SURF algorithm, HARRIS algorithm and NCC (Normalized Cross Correlation) algorithm. Two lines are considered to match correctly if and only if the corresponding corner points match. Calculate the transformation relationship between the projection maps, and then apply the transformation parameters to the matching between the three-dimensional terrain, so as to complete the whole terrain matching process.

Description

A Fast and Accurate Matching Method for 3D Terrain Based on Heterogeneous Projection

technical field

The invention belongs to the field of three-dimensional terrain matching, in particular to a fast and accurate three-dimensional terrain matching method based on heterogeneous projections.

Background technique

Terrain matching technology is one of the key technologies for terrain-assisted navigation. It is widely used in the aviation field, and it also has broad application prospects in underwater vehicle terrain matching and positioning, robot navigation and positioning, and land vehicle navigation.

There are many existing terrain matching algorithms, and people at home and abroad have been studying them, and the algorithms are constantly being updated and improved. Among the existing methods, there is a method of using the points with the largest and smallest Gaussian curvatures in the Gaussian curvature image as feature points for terrain matching. A matching method based on terrain contours is used to describe and match contours using normalized wavelet descriptors, but this method is not suitable for situations where the contours of the terrain are not obvious. There is a method using visual dictionaries to prove that the direct 2D to 3D matching method can improve the matching performance, but if the amount of data in this method is too large, it will affect the memory and take a long time, and the wrong descriptors are not eliminated during the matching process. Later, a 3D terrain matching algorithm based on surface features appeared. The disadvantage of this method is that if the terrain data is too large or the terrain map is complex, the matching time will be very long and the matching speed will be too slow. For this reason, the present invention proposes a new fast and accurate matching method for 3D terrain based on terrain orthographic projection, which makes full use of the surface features of the terrain. In many cases, the projection images are not homogeneous but heterogeneous images. Heterogeneous images lack grayscale information, and traditional image matching methods are no longer applicable. The present invention uses the method of combining the straight line feature and the point feature to match the projection image. Firstly, the straight line matching algorithm is used to match the straight lines with the same name in the image, and the intersection point of two straight lines with the same name is used as a virtual corner point, and then the point feature is used for the second matching, and the wrongly matched straight lines are eliminated. When matching point features, the improved SURF (Speed-up robust features) algorithm with higher matching accuracy is used, which combines SURF algorithm, HARRIS algorithm and NCC (Normalized Cross Correlation) algorithm.

Contents of the invention

The purpose of the present invention is to provide a fast and accurate matching method for three-dimensional terrain based on heterogeneous projections, so as to improve the matching efficiency and accuracy of three-dimensional terrain.

The purpose of the present invention is achieved in this way, a method for fast and accurate matching of three-dimensional terrain based on heterogeneous projections, characterized in that it at least includes the following steps:

Step 1, convert the original digital elevation model terrain data format USGS-DEM into digital elevation model terrain data format CNSTDF-DEM;

Step 2, performing orthographic projection on the reference 3D terrain of the terrain format described in step 1 and the 3D terrain to be matched;

Step 3, according to the orthographic projection described in step 2, the projection map is matched to obtain the projection map transformation parameters, and the matching process adopts the method of combining line features and point features;

Step 4, according to the transformation parameters described in step 3, match the reference 3D terrain and the 3D terrain to be matched and obtain a matching result.

Said step 1 includes the following steps:

Step 11, open a terrain file;

Step 12, judging whether the opened file is in the USGS-DEM terrain data format;

Step 13, if not then go to step 11, if it is then start to read the data header;

Step 14, extracting relevant terrain file header information;

Step 15, storing file header information;

Step 16, open up data storage space;

Step 17, read the data body;

Step 18, filtering the first 144 bytes of each line of the data body;

Step 19, storing data-related information;

Step 110, save as CNSTDF-DEM terrain format file in conjunction with stored file header and data body;

Step 111, terrain conversion is completed.

Said step 2 comprises the following steps:

Step 21, open the obtained three-dimensional CNSTDF-DEM terrain data;

Step 22, obtaining terrain surface texture features;

Step 23, obtaining a three-dimensional terrain texture;

In step 24, the terrain orthographic projection map is obtained according to the texture.

Said step 3 includes the following steps:

Step 31, performing line detection and matching on the projection image to be matched and the reference projection image, after matching the straight lines with the same name, using the intersection point where two straight lines intersect in the projection image to be matched as the virtual corner point of the projection image to be matched;

Step 32, using the intersection point where two straight lines intersect in the reference projection diagram as the virtual corner point of the reference projection diagram;

Step 33, performing rough matching of SURF algorithm on virtual corner points;

Step 34, performing SURF algorithm fine matching on the virtual corner points;

Step 35, through the first matching, the accurate feature point pair set φ _AB of the reference image and the image to be registered is obtained, and the perspective transformation matrix H between the reference image and the image to be registered can be obtained through φ _AB ;

Step 36, transforming the projection image to be matched according to the obtained transformation matrix H;

Step 37, interpolating the projection image to be matched;

Step 38, obtain the intermediate image after transformation and interpolation;

Step 39, obtaining the reference projection image and the projection image to be matched;

Step 310, find out the overlapping parts between the reference projection image and the projection image to be matched as respective regions of interest, and divide the reference image into multiple sub-regions according to the size of the overlapping regions. When the overlapping area is larger, the size of the sub-area is 64×64, and when the overlapping area is smaller, the sub-area becomes smaller accordingly;

Step 311, in the sub-region of the reference projection image, take the center of the region as the center, and perform HARRIS feature point extraction in a neighborhood of 32×32, that is, 0.5 times the size of the reference image sub-region, and take all the HARRIS feature points in this region The point with the largest R value, that is, the most distinguishable point from the surrounding points is used as the feature point of the reference projection map; if there is no HARRIS feature point in the 32×32 neighborhood, the center of the sub-region is treated as a feature point;

Step 312, after all the reference projection map feature points are extracted, perform NCC algorithm matching;

Step 313, in the intermediate image, search within the neighborhood of 96×96 centered on the coordinates of the feature points of the reference projection image, that is, 1.5 times the size of the sub-area of the reference projection image, record the correlation coefficient and the coordinates of the feature points, and obtain rough matching points ;

Step 314, after the 96×96 area search is completed, compare the correlation coefficients of all recorded rough matching points, select the largest correlation coefficient, and limit the threshold T _NCC ;

Step 315, if it is greater than the given threshold T _NCC , then the corresponding coordinate point is used as the fine matching point between the feature point of the intermediate image and the feature point of the reference projection image;

Step 316, fitting with the least squares method according to the fine matching points;

In step 317, the transformation matrix between the intermediate image and the reference projection image is obtained, and the final transformation parameters for projection matching are obtained to complete the projection image matching.

Said step 4 comprises the following steps:

Step 41, obtaining the transformation parameters obtained after the projection matching is completed;

Step 42, return to the 3D terrain, and convert the 3D terrain;

Step 43, complete the three-dimensional terrain matching.

The method includes acquisition and conversion of 3D DEM (Digital Elevation Model) terrain data, orthographic projection of the 3D terrain, and matching of projection maps according to the orthographic projection of the terrain. In the process of matching the projection image, the method of combining the straight line feature and the point feature is adopted. After the line detection is performed on the projection image and the straight line with the same name is matched, the virtual corner point where the two straight lines with the same name intersect is found, and the coordinates of the corner point are calculated. . Then use the improved SURF algorithm to match the found virtual corners. If and only when the corresponding corners match, the two straight lines are considered to match correctly. Calculate the transformation relationship between the projection maps, and then apply the transformation parameters to the matching between the three-dimensional terrain, so as to complete the whole terrain matching process. The invention uses three-dimensional terrain projection for matching, and is a new method for three-dimensional terrain matching. The present invention can be applied to unmanned aerial vehicle visual navigation, under the condition of degraded environment, there is occlusion, and is aimed at the matching of heterogeneous projection.

The beneficial effect of the invention is that the surface features of the terrain are fully utilized to match the orthographic projection of the three-dimensional terrain, and a new three-dimensional terrain matching algorithm is provided. The algorithm also works for heterogeneous terrain projections. The improved SURF point feature algorithm improves the accuracy of projection matching.

Description of drawings

Fig. 1 flow chart of the present invention;

Fig. 2 flow chart of terrain data format conversion;

Fig. 3 acquisition of terrain projection map;

Fig. 4 improves the SURF algorithm flow chart;

Figure 5 Terrain matching process;

Figure 6 The results of SURF and improved SURF algorithm for heterogeneous projection graphs.

Detailed ways

As shown in Figure 1, the characteristics of the flowchart steps of 3D terrain matching are:

Step 1, convert the original digital elevation model terrain data format USGS-DEM into digital elevation model terrain data format CNSTDF-DEM;

Step 2, performing orthographic projection on the reference 3D terrain of the terrain format described in step 1 and the 3D terrain to be matched;

Step 3, according to the orthographic projection described in step 2, the projection map is matched to obtain the projection map transformation parameters, and the matching process adopts the method of combining line features and point features;

Step 4, according to the transformation parameters described in step 3, match the reference 3D terrain and the 3D terrain to be matched and obtain a matching result.

As shown in Figure 2, said step 1 includes the following steps, which are characterized in that:

Step 11, open a terrain file;

Step 12, judging whether the opened file is in the USGS-DEM terrain data format;

Step 13, if not then go to step 11, if it is then start to read the data header;

Step 14, extracting relevant terrain file header information;

Step 15, storing file header information;

Step 16, open up data storage space;

Step 17, read the data body;

Step 18, filtering the first 144 bytes of each line of the data body;

Step 19, storing data-related information;

Step 110, save as CNSTDF-DEM terrain format file in conjunction with stored file header and data body;

Step 111, terrain conversion is completed.

As shown in Figure 3, said step 2 includes the following steps, which are characterized in that:

Step 21, open the obtained three-dimensional CNSTDF-DEM terrain data;

Step 22, obtaining terrain surface texture features;

Step 23, obtaining a three-dimensional terrain texture;

In step 24, the terrain orthographic projection map is obtained according to the texture.

As shown in Figure 4, said step 3 includes the following steps, characterized in that:

Step 31, performing line detection and matching on the projection image to be matched and the reference projection image, after matching the straight lines with the same name, using the intersection point where two straight lines intersect in the projection image to be matched as the virtual corner point of the projection image to be matched;

Step 32, taking the intersection point where two lines intersect in the reference projected map as the virtual corner point of the reference projected map.

Step 33, performing rough matching of SURF algorithm on virtual corner points;

Step 34, performing SURF algorithm fine matching on the virtual corner points;

Step 35, through the first matching, the accurate feature point pair set φ _AB of the reference image and the image to be registered is obtained, and the perspective transformation matrix H between the reference image and the image to be registered can be obtained through φ _AB ;

Step 36, transforming the projection image to be matched according to the obtained transformation matrix H;

Step 37, interpolating the projection image to be matched;

Step 38, obtain the intermediate image after transformation and interpolation;

Step 39, obtaining the reference projection image and the projection image to be matched;

Step 310, find out the overlapping parts between the reference projection image and the projection image to be matched as respective regions of interest, and divide the reference image into multiple sub-regions according to the size of the overlapping regions. When the overlapping area is larger, the size of the sub-area is 64×64, and when the overlapping area is smaller, the sub-area becomes smaller accordingly;

Step 311, in the sub-region of the reference projection image, take the center of the region as the center, and perform HARRIS feature point extraction in a neighborhood of 32×32, that is, 0.5 times the size of the reference image sub-region, and take all the HARRIS feature points in this region The point with the largest R value, that is, the most distinguishable point from the surrounding points is used as the feature point of the reference projection map; if there is no HARRIS feature point in the 32×32 neighborhood, the center of the sub-region is treated as a feature point;

Step 312, after all the reference projection map feature points are extracted, perform NCC algorithm matching;

Step 313, in the intermediate image, search in the 96×96 (1.5 times the size of the sub-area of the reference projection image) area centered on the coordinates of the feature points of the reference projection image, record the correlation coefficient and the coordinates of the feature points, and obtain rough matching points ;

Step 314, after the 96×96 area search is completed, compare the correlation coefficients of all recorded rough matching points, select the largest correlation coefficient, and limit the threshold T _NCC ;

Step 315, if it is greater than a given threshold T _NCC , the corresponding coordinate point is used as a precise matching point between the feature point of the intermediate image and the feature point of the reference projection image.

Step 316, fitting with the least squares method according to the fine matching points;

In step 317, the transformation matrix between the intermediate image and the reference projection image is obtained, and the final transformation parameters for projection matching are obtained to complete the projection image matching.

As shown in Figure 5, said step 4 includes the following steps, which are characterized in that:

Step 41, obtaining the transformation parameters obtained after the projection matching is completed;

Step 42, return to the 3D terrain, and convert the 3D terrain;

Step 43, complete the three-dimensional terrain matching.

As shown in Figure 6, Figure 6(a) is the visible light projection map of the terrain, Figure 6(b) is the infrared projection map of the terrain, Figure 6(c) is the matching result using the SURF algorithm, and the matching accuracy is 0.2108 pixels, Figure 6(d) is the matching result using the improved SURF algorithm in this paper, and the matching accuracy is 0.0338 pixels. The results show that the improved SURF algorithm improves the accuracy of the matching process.

The parts not described in detail in the steps belong to common means and algorithms known in the art, and will not be described here one by one.

Claims (4)

1. a kind of quick fine matching method of dimensional topography based on isomery projection, it is characterised in that including at least following steps：

Step 1, original figure elevation model terrain data form USGS-DEM is converted into digital elevation model terrain data lattice Formula CNSTDF-DEM；

Step 2, to digital elevation model terrain data form CNSTDF-DEM reference dimensional topography described in step 1 and to be matched Dimensional topography carries out orthogonal projection；

Step 3, the orthogonal projection according to step 2, is matched to perspective view, is obtained perspective view transformation parameter, was matched The method that Cheng Caiyong linear features and point feature combine；

Step 4, the transformation parameter according to step 3, to being matched and being obtained with reference to dimensional topography and dimensional topography to be matched To matching result；

The step 3, comprises the following steps：

Step 31, treat matching pursuit figure and the detection and matching of straight line are carried out with reference to perspective view, after having matched homonymous line, Using the intersection point of straight line intersection two-by-two as the virtual angle point of perspective view to be matched in perspective view to be matched；

Step 32, with reference in perspective view using the intersection point of straight line intersection two-by-two as referring to the virtual angle point of perspective view；

Step 33, SURF algorithm is carried out to virtual angle point slightly to match；

Step 34, SURF algorithm essence matching is carried out to virtual angle point；

Step 35, matching that the two is combined by the thick matching of step 33 and the essence matching of step 34, obtained reference picture and The accurate feature points of image subject to registration are to set φ_AB, pass through φ_ABCan be saturating between reference picture and image subject to registration to ask for Depending on transformation matrix H；

Step 36, treat matching pursuit figure and line translation is entered according to the transformation matrix H obtained；

Step 37, treat matching pursuit figure and enter row interpolation；

Step 38, convert interpolation and obtain intermediate image afterwards；

Step 39, obtain and refer to perspective view and perspective view to be matched；

Step 310, find out with reference to the lap between perspective view and perspective view to be matched as respective area-of-interest, and And it is polylith subregion to divide reference picture according to overlapping region size；When overlapping region is larger, subregion size be 64 × 64, when overlapping region is smaller, subregion also diminishes accordingly；

Step 311, in the subregion with reference to perspective view, centered on regional center, 32 × 32 around, i.e., reference picture is sub HARRIS feature point extractions are carried out in 0.5 times of neighborhood of area size, take in the region R values maximum in all HARRIS characteristic points , i.e., most there is the point of discrimination with surrounding point as the characteristic point with reference to perspective view；If do not have in 32 × 32 neighborhoods HARRIS characteristic points, then sub-areas center handled as a characteristic point；

Step 312, after all reference perspective view feature point extractions are complete, NCC algorithmic match is carried out；

Step 313, in intermediate image, 96 × 96 centered on reference perspective view feature point coordinates, i.e., with reference to perspective view Scanned in 1.5 times of neighborhoods of area size, record coefficient correlation and its feature point coordinates, obtain thick match point；

Step 314, after the completion of 96 × 96 range searchings, the coefficient correlation of the thick match point of more all records, maximum is selected Coefficient correlation, carry out threshold value T_NCCLimit；

Step 315, if greater than given threshold value T_NCC, then corresponding coordinate point is as the characteristic point of intermediate image and with reference to perspective view The smart match point of characteristic point；

Step 316, it is fitted according to smart matching double points with least square method；

Step 317, the transformation matrix between intermediate image and reference perspective view is obtained, obtains the last transformation parameter of projection matching, Complete perspective view matching.

2. a kind of quick fine matching method of dimensional topography based on isomery projection according to claim 1, its feature exist In：The step 1, comprises the following steps：

Step 11, a terrain file is opened；

Step 12, whether the file for judging to open is USGS-DEM terrain data forms；

Step 13, if not step 11 is then gone to, if it is start to read data head；

Step 14, related terrain file header is extracted；

Step 15, storage file header；

Step 16, data space is opened up；

Step 17, data volume is read；

Step 18, preceding 144 bytes of the often row of data volume are filtered；

Step 19, data storage relevant information；

Step 110, the file header and data volume with reference to storage save as CNSTDF-DEM topography format files；

Step 111, terrain transition is completed.

3. a kind of quick fine matching method of dimensional topography based on isomery projection according to claim 1, its feature exist In：The step 2, comprises the following steps：

Step 21, obtained three-dimensional CNSTDF-DEM terrain datas are opened；

Step 22, topographical surface textural characteristics are obtained；

Step 23, dimensional topography texture is obtained；

Step 24, landform orthogonal projection figure is obtained according to texture.

4. a kind of quick fine matching method of dimensional topography based on isomery projection according to claim 1, its feature exist In：The step 4, comprises the following steps：

Step 41, the transformation parameter obtained after the completion of projection matching is obtained；

Step 42, return in dimensional topography, dimensional topography is changed；

Step 43, three_dimensional topograph model is completed.