CN103390273A - Multi-beam side-scan sonar image registration method based on GPS (global positioning system) positioning assistance - Google Patents

Abstract

The invention belongs to the field of acoustic image registration, and particularly relates to a multi-beam side-scan sonar image registration method based on GPS positioning assistance. The method comprises the steps as follows: multi-beam side-scan sonar data are read, and a reference image and a to-be-registered image are generated; GPS positioning assistance information of track points is recorded sequentially; feature points of the reference image and the to-be-registered image generated by the multi-beam side-scan sonar data are extracted respectively, and feature vectors are established; the feature points in the reference image and the to-be-registered image are matched; for a feature point in the reference point, a point with the smallest XOR sum value of Hamming distances of feature points in the to-be-registered image is found out and considered as a matching point; and according to the determined registration control point, a spatial alternation relationship is obtained, and registration is completed. According to the method, longitude and latitude coordinates of the feature points under a geodetic coordinate system are calculated, mistakenly matched points are effectively eliminated, and the optimized spatial alternation relationship is obtained, so that the registration is more accurate.

Description

A kind of based on the auxiliary multi-beam side-scanning sonar image method for registering in GPS location

Technical field

The invention belongs to acoustic picture registration field, be specifically related to a kind of based on the auxiliary multi-beam side-scanning sonar image method for registering in GPS location.

Background technology

Pull the important engine of the national economic development along with marine economy becomes countries in the world, ocean development becomes the key areas of weighing national development.In recent years,, due to the development of side-scan sonar technology, all increase at aspects such as the resolution of the stability of collection signal and sonar view data and sharpness, for marine charting provides more favourable hardware supported.Utilize the data of side-scan sonar collection to draw bottom relief map and can help the laying of subsea cable optical fiber and pipeline route, the petroleum well field investigation, the anchorage addressing, the seabed engineerings such as bridge spanning the sea construction obtain the terrain data in early stage, avoid the seabed disaster and easily send out district or the zone that is not suitable for carrying out engineering construction, for the safety of seabed engineering provides powerful guarantee smoothly.

The image that sonar obtains not is the geomorphologic map picture in whole seabed, but the scan image of band shape, in order to obtain complete bottom relief map, needs side-scanning sonar image is carried out spatial registration, then by the image mosaic means, obtains complete topomap.yet the multi-beam side-scanning sonar image is in registration process, because the intensity profile of sonar image is similar, after obtaining proper vector, existing method relies on merely measuring similarity, as document 2011 by P.Vanish, Vardy.A, Walker.D, Dobre.O.A. " the Side-Scan sonar image registration for AUV navigation. " that the 1-7 page is delivered in 2011IEEE symposium on and2011Workshop on Scientific Use of Submarine Cables and Related Technolgies (SSC) adopts the SIFT algorithm to extract the sonar image unique point, then determine potential matching double points by the ratio of the nearest of calculated characteristics vector and time nearly Euclidean distance, ratio is less than the match point that is considered as of certain threshold value, yet this method, the left and right matching result is understood in choosing of threshold value, cause Mismatching point more, often can't obtain best spatial alternation relation, make image to inlay, therefore, reject Mismatching point in registration process, search out best spatial alternation relation, have great importance for follow-up sonar image Processing tasks.

Summary of the invention

The object of the present invention is to provide a kind of registration more accurately based on the auxiliary multi-beam side-scanning sonar image method for registering in GPS location.

The object of the present invention is achieved like this:

(1) read multi-beam side-scan sonar data, generating reference image and image subject to registration;

(2) record successively the GPS positioning auxiliary information of track points;

(3) reference picture and the image extract minutiae subject to registration that respectively multi-beam side sonar data are generated, and set up proper vector:

1) image is carried out feature point detection, for any gradation of image value I (x) on the circumference centered by pixel p,

$N=\underset{x\∀\left(\mathrm{circle}\left(p\right)\right)}{\mathrm{\Σ}}|I\left(x\right)-I\left(p\right)|>\mathrm{\ϵ},$

N is for meeting the number of the pixel of following formula on take p as the center of circle, and I (x) is the gray scale of any point on circumference, and ε is threshold value, if N more than or equal to circumference count 3/4ths, p is defined as unique point;

2) generating feature vector: calculated characteristics point principal direction:

$m_{\mathrm{pq}}=\underset{i,j}{\mathrm{\Σ}}{x}^p{y}^{q}I(i,j),$

$\mathrm{\θ}=\arctan \left(\frac{m_{01}}{m_{10}}\right)=\arctan \left(\frac{\underset{i,j}{\mathrm{\Σ}}\mathrm{jI}(i,j)}{\underset{i,j}{\mathrm{\Σ}}\mathrm{iI}(i,j)}\right),$

Wherein, m _pqFor the p+q rank geometric moment of unique point neighborhood, i, j are the coordinates of unique point, θ is unique point principal direction, and I (i, j) is point (i, j) gray-scale value of locating, after selected principal direction, the generating feature vector, near the pixel of 11 * 11 neighborhoods selected characteristic point, this neighborhood is chosen 3 * 3 block of pixels pair clockwise successively by after principal direction rotation, to the right gray-scale value of these block of pixels and size compare, be combined into binary string, be the feature description vectors of this unique point, expression formula is:

$\mathrm{\&tau;}(g;x,y)=\left\{\begin{array}{cc}1& :g\left(x\right)<g\left(y\right)\\ 0& :g\left(x\right)>g\left(y\right)\end{array}\right.,$

$f_n\left(g\right)=\underset{1\&le;m\&le;n}{\mathrm{\&Sigma;}}{2}^{m-1}\mathrm{\&tau;}(g;x_m,y_m),$

Wherein, g (x), g (y) be in 3 * 3 block of pixels centered by an x and y gray-scale value and, τ (g; X, y) be binary string, f _n(g) be proper vector, n is the right number of block of pixels, i.e. proper vector dimension, and m represents m block of pixels pair in neighborhood;

(4) unique point in reference picture and image subject to registration is mated:, for the unique point in reference picture, find out the point of the XOR sum minimum between the Hamming distance of unique point in image subject to registration, think that this point is match point, i.e. F ₁, F ₂For the unique point in reference picture and image subject to registration vector:

F ₁=x ₀x ₁x ₂X _n, F ₂=y ₀y ₁y ₂Y _n, H (F ₁, F ₂) expression two proper vectors the Hamming distance XOR and, namely

$H(F_1,F_2)=\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i\&CirclePlus;y_i,$

Wherein, n represents the dimension of vector, and i represents i element in vector, H (F ₁, F ₂) minimum point is match point;

(5) latitude and longitude coordinates under earth coordinates according to GPS positioning auxiliary information computing reference image and each unique point of image subject to registration,

$\left\{\begin{array}{c}\mathrm{lon}\left[l\right]=\mathrm{lon}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40009000}\&times;\sin \left(\mathrm{courseangle}\right[k\left]\right)\\ \mathrm{lat}\left[l\right]=\mathrm{lat}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40075412}\&times;\cos \left(\mathrm{courseangle}\right[k\left]\right)\end{array}\right.,$

Wherein, l is any pixel, and its longitude and latitude are respectively lon[l] and lat[l], k is track points, courseangle[k] be the course angle at this some place, width is picture traverse, and range is the oblique distance of sonar, and altitude is the degree of depth of sonar,

Latitude and longitude coordinates with reference to the unique point of mating in image and image subject to registration is subtracted each other respectively,, if difference, greater than threshold value, is judged as Mismatching point, rejects;

(6) obtain the spatial alternation relation according to the registration control points of determining, complete registration.

The GPS positioning auxiliary information comprises latitude and longitude information and course angle information.

Beneficial effect of the present invention is: the present invention is by carrying out feature extraction to the multi-beam side-scanning sonar image, obtain the proper vector of the unique point of reference picture and image subject to registration, similar for the intensity profile due to sonar image, only use method for measuring similarity can cause the more problem of Mismatching point, take full advantage of the GPS positioning auxiliary information of sonar data, put latitude and longitude coordinates under earth coordinates by calculated characteristics, effectively rejected Mismatching point, obtain optimized spatial alternation relation, made registration more accurate.

Description of drawings

Fig. 1 is the basic flow sheet of the inventive method;

Fig. 2 is that existing method only relies on similarity measurement and carries out the result of registration;

Fig. 3 is according to the registration result after GPS supplementary rejecting Mismatching point.

Embodiment

For example the present invention is described in more detail below in conjunction with accompanying drawing:

Basic step of the present invention is as shown in Figure 1:

(1) read multi-beam side-scan sonar data, generating reference image and image subject to registration;

(2) record successively the GPS positioning auxiliary information of track points;

(3) reference picture and the image extract minutiae subject to registration that respectively sonar data are generated, and set up proper vector;

(4) by adopting the Hamming distance nearest neighbor algorithm that judgement is measured as similarity to mate the unique point in reference picture and image subject to registration;

(5) latitude and longitude coordinates under earth coordinates according to GPS positioning auxiliary information computing reference image and each unique point of image subject to registration, the longitude and latitude of identical unique point under earth coordinates is identical, in correct matching double points, latitude and longitude coordinates is subtracted each other respectively, difference should be very little, if difference is greater than a certain threshold value, be judged as Mismatching point, reject;

(6) obtain the spatial alternation relation according to the registration control points of determining, complete registration.

Characteristics of the present invention are:

1. in step (2), the GPS positioning auxiliary information of track points comprises latitude and longitude information and course angle information;

2. in step (3), the concrete steps of feature point extraction are:

1. at first image is carried out feature point detection, establishing p is arbitrary pixel, the gradation of image value of any point on the circumference of I (x) expression centered by p, and ε is a very little threshold value, unique point can be determined by following formula:

$N=\underset{x\&ForAll;\left(\mathrm{circle}\left(p\right)\right)}{\mathrm{\&Sigma;}}|I\left(x\right)-I\left(p\right)|>\mathrm{\&epsiv;}---\left(1\right)$

Wherein N is for meeting the number of the pixel of following formula on take p as the center of circle, be generally that circumference counts 3/4ths, even circumference has 16 points, p is defined as unique point when N＞12;

2. the generation of proper vector: have rotational invariance in order to make unique point, by following formula calculated characteristics point principal direction:

$m_{\mathrm{pq}}=\underset{i,j}{\mathrm{\&Sigma;}}{x}^p{y}^{q}I(i,j)---\left(2\right)$

$\mathrm{\&theta;}=\arctan \left(\frac{m_{01}}{m_{10}}\right)=\arctan \left(\frac{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{jI}(i,j)}{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{iI}(i,j)}\right)---\left(3\right)$

Wherein, m _pqP+q rank geometric moment for the unique point neighborhood, i, j is the coordinate of the unique point of unique point, θ is unique point principal direction, I (i, j) be point (i, j) gray-scale value of locating, after selected principal direction, but just generating feature is vectorial, its method is near the pixel of 11 * 11 neighborhoods selected characteristic point, this neighborhood is chosen 3 * 3 block of pixels pair clockwise successively by after principal direction rotation, to the right gray-scale value of these block of pixels and size compare, be combined into binary string, be the feature description vectors of this unique point, expression formula is as follows:

$\mathrm{\&tau;}(g;x,y)=\left\{\begin{array}{cc}1& :g\left(x\right)<g\left(y\right)\\ 0& :g\left(x\right)>g\left(y\right)\end{array}\right.---\left(4\right)$

$f_n\left(g\right)=\underset{1\&le;m\&le;n}{\mathrm{\&Sigma;}}{2}^{m-1}\mathrm{\&tau;}(g;x_m,y_m)---\left(5\right)$

Wherein, g (x), g (y) be in 3 * 3 block of pixels centered by an x and y gray-scale value and, τ (g; X, y) be binary string, f _n(g) be proper vector, n is the right number of block of pixels, i.e. proper vector dimension, and m represents m block of pixels pair in neighborhood.

3. nearest neighbor algorithm adopts Hamming distance as similarity judgement tolerance in step (4),, for certain unique point in reference picture, finds out in image subject to registration the point of the XOR sum minimum between its Hamming distance, thinks that this point is match point.

4. in step (5), reference picture and the latitude and longitude coordinates of each unique point of image subject to registration under earth coordinates are calculated by following formula:

$\left\{\begin{array}{c}\mathrm{lon}\left[l\right]=\mathrm{lon}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40009000}\&times;\sin \left(\mathrm{courseangle}\right[k\left]\right)\\ \mathrm{lat}\left[l\right]=\mathrm{lat}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40075412}\&times;\cos \left(\mathrm{courseangle}\right[k\left]\right)\end{array}\right.---\left(6\right)$

Wherein, l is any pixel, and its longitude and latitude are respectively lon[l] and lat[l], k is track points, courseangle[k] be the course angle at this some place, width is picture traverse, range is the oblique distance of sonar, altitude is the degree of depth of sonar, and is identical at the longitude and latitude under earth coordinates according to identical unique point, and in correct matching double points, latitude and longitude coordinates is subtracted each other respectively, difference should be very little,, if difference, greater than a certain threshold value, is judged as Mismatching point, reject.

In conjunction with Fig. 1, the concrete steps that the present invention is based on the auxiliary multi-beam side-scanning sonar image method for registering in GPS location are as follows:

(1) read multi-beam side-scan sonar data, generating reference image and image subject to registration;

(2) record successively the GPS positioning auxiliary information of track points;

(3) reference picture and the image extract minutiae subject to registration that respectively sonar data are generated, and set up proper vector, its concrete operation method is:

1. at first image is carried out feature point detection, establishing p is arbitrary pixel, the gradation of image value of any point on the circumference of I (x) expression centered by p, and ε is a very little threshold value, unique point can be determined by following formula:

$N=\underset{x\&ForAll;\left(\mathrm{circle}\left(p\right)\right)}{\mathrm{\&Sigma;}}|I\left(x\right)-I\left(p\right)|>\mathrm{\&epsiv;}---\left(1\right)$

Wherein N is for meeting the number of the pixel of following formula on take p as the center of circle, be generally that circumference counts 3/4ths, even circumference has 16 points, p is defined as unique point when N＞12;

2. the generation of proper vector: have rotational invariance in order to make unique point, by following formula calculated characteristics point principal direction:

$m_{\mathrm{pq}}=\underset{i,j}{\mathrm{\&Sigma;}}{x}^p{y}^{q}I(i,j)---\left(2\right)$

$\mathrm{\&theta;}=\arctan \left(\frac{m_{01}}{m_{10}}\right)=\arctan \left(\frac{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{jI}(i,j)}{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{iI}(i,j)}\right)---\left(3\right)$

Wherein, m _pqP+q rank geometric moment for the unique point neighborhood, i, j is the coordinate of the unique point of unique point, θ is unique point principal direction, I (i, j) be point (i, j) gray-scale value of locating, after selected principal direction, but just generating feature is vectorial, its method is near the pixel of 11 * 11 neighborhoods selected characteristic point, this neighborhood is chosen 3 * 3 block of pixels pair clockwise successively by after principal direction rotation, to the right gray-scale value of these block of pixels and size compare, be combined into binary string, be the feature description vectors of this unique point, expression formula is as follows:

$\mathrm{\&tau;}(g;x,y)=\left\{\begin{array}{cc}1& :g\left(x\right)<g\left(y\right)\\ 0& :g\left(x\right)>g\left(y\right)\end{array}\right.---\left(4\right)$

$f_n\left(g\right)=\underset{1\&le;m\&le;n}{\mathrm{\&Sigma;}}{2}^{m-1}\mathrm{\&tau;}(g;x_m,y_m)---\left(5\right)$

Wherein, g (x), g (y) be in 3 * 3 block of pixels centered by an x and y gray-scale value and, τ (g; X, y) be binary string, f _n(g) be proper vector, n is the right number of block of pixels, i.e. proper vector dimension, and m represents m block of pixels pair in neighborhood.

(4) by adopting the Hamming distance nearest neighbor algorithm that judgement is measured as similarity to mate the unique point in reference picture and image subject to registration, for certain unique point in reference picture, find out in image subject to registration the point of the XOR sum minimum between its Hamming distance, think that this point is match point.

(5) latitude and longitude coordinates under earth coordinates according to GPS positioning auxiliary information computing reference image and each unique point of image subject to registration, the longitude and latitude of identical unique point under earth coordinates is identical, in correct matching double points, latitude and longitude coordinates is subtracted each other respectively, difference should be very little, if difference is greater than a certain threshold value, be judged as Mismatching point, reject, wherein reference picture and the latitude and longitude coordinates of each unique point of image subject to registration under earth coordinates are calculated by following formula:

$\left\{\begin{array}{c}\mathrm{lon}\left[l\right]=\mathrm{lon}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40009000}\&times;\sin \left(\mathrm{courseangle}\right[k\left]\right)\\ \mathrm{lat}\left[l\right]=\mathrm{lat}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40075412}\&times;\cos \left(\mathrm{courseangle}\right[k\left]\right)\end{array}\right.---\left(6\right)$

Wherein, l is any pixel, and its longitude and latitude are respectively lon[l] and lat[l], k is track points, courseangle[k] be the course angle at this some place, width is picture traverse, and range is the oblique distance of sonar, and altitude is the degree of depth of sonar.

(6) obtain the spatial alternation relation according to the registration control points of determining, complete registration.

As shown in Figure 3, by contrast, can see, with existing method, carry out the sonar image registration, reference picture and image mismatch point subject to registration are very many, the registration poor effect, and after processing by the GPS supplementary with the inventive method, Mismatching point is all eliminated, and registration result is fine.

Claims (2)

1. locate auxiliary multi-beam side-scanning sonar image method for registering based on GPS for one kind, it is characterized in that:

(1) read multi-beam side-scan sonar data, generating reference image and image subject to registration;

(2) record successively the GPS positioning auxiliary information of track points;

(3) reference picture and the image extract minutiae subject to registration that respectively multi-beam side sonar data are generated, and set up proper vector:

1) image is carried out feature point detection, for any gradation of image value I (x) on the circumference centered by pixel p,

$N=\underset{x\&ForAll;\left(\mathrm{circle}\left(p\right)\right)}{\mathrm{\&Sigma;}}|I\left(x\right)-I\left(p\right)|>\mathrm{\&epsiv;},$

N is for meeting the number of the pixel of following formula on take p as the center of circle, and I (x) is the gray scale of any point on circumference, and ε is threshold value, if N more than or equal to circumference count 3/4ths, p is defined as unique point;

2) generating feature vector: calculated characteristics point principal direction:

$m_{\mathrm{pq}}=\underset{i,j}{\mathrm{\&Sigma;}}{x}^p{y}^{q}I(i,j),$

$\mathrm{\&theta;}=\arctan \left(\frac{m_{01}}{m_{10}}\right)=\arctan \left(\frac{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{jI}(i,j)}{\underset{i,j}{\mathrm{\&Sigma;}}\mathrm{iI}(i,j)}\right),$

Wherein, m _pqFor the p+q rank geometric moment of unique point neighborhood, i, j are the coordinates of unique point, θ is unique point principal direction, and I (i, j) is point (i, j) gray-scale value of locating, after selected principal direction, the generating feature vector, near the pixel of 11 * 11 neighborhoods selected characteristic point, this neighborhood is chosen 3 * 3 block of pixels pair clockwise successively by after principal direction rotation, to the right gray-scale value of these block of pixels and size compare, be combined into binary string, be the feature description vectors of this unique point, expression formula is:

$\mathrm{\&tau;}(g;x,y)=\left\{\begin{array}{cc}1& :g\left(x\right)<g\left(y\right)\\ 0& :g\left(x\right)>g\left(y\right)\end{array}\right.,$

$f_n\left(g\right)=\underset{1\&le;m\&le;n}{\mathrm{\&Sigma;}}{2}^{m-1}\mathrm{\&tau;}(g;x_m,y_m),$

Wherein, g (x), g (y) be in 3 * 3 block of pixels centered by an x and y gray-scale value and, τ (g; X, y) be binary string, f _n(g) be proper vector, n is the right number of block of pixels, i.e. proper vector dimension, and m represents m block of pixels pair in neighborhood;

(4) unique point in reference picture and image subject to registration is mated:, for the unique point in reference picture, find out the point of the XOR sum minimum between the Hamming distance of unique point in image subject to registration, think that this point is match point, i.e. F ₁, F ₂For the unique point in reference picture and image subject to registration vector:

F ₁=x ₀x ₁x ₂X _n, F ₂=y ₀y ₁y ₂Y _n, H (F ₁, F ₂) expression two proper vectors the Hamming distance XOR and, namely

$H(F_1,F_2)=\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i\&CirclePlus;y_i,$

Wherein, n represents the dimension of vector, and i represents i element in vector, H (F ₁, F ₂) minimum point is match point;

(5) latitude and longitude coordinates under earth coordinates according to GPS positioning auxiliary information computing reference image and each unique point of image subject to registration,

$\left\{\begin{array}{c}\mathrm{lon}\left[l\right]=\mathrm{lon}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40009000}\&times;\sin \left(\mathrm{courseangle}\right[k\left]\right)\\ \mathrm{lat}\left[l\right]=\mathrm{lat}\left[k\right]+(l-\frac{\mathrm{width}}{2})\&times;\frac{\sqrt{{\mathrm{range}}^2-{\mathrm{altitude}}^2}}{\frac{\mathrm{width}}{2}}\&times;\frac{360}{40075412}\&times;\cos \left(\mathrm{courseangle}\right[k\left]\right)\end{array}\right.,$

Wherein, l is any pixel, and its longitude and latitude are respectively lon[l] and lat[l], k is track points, courseangle[k] be the course angle at this some place, width is picture traverse, and range is the oblique distance of sonar, and altitude is the degree of depth of sonar,

Latitude and longitude coordinates with reference to the unique point of mating in image and image subject to registration is subtracted each other respectively,, if difference, greater than threshold value, is judged as Mismatching point, rejects;

(6) obtain the spatial alternation relation according to the registration control points of determining, complete registration.

2. according to claim 1 a kind of based on the auxiliary multi-beam side-scanning sonar image method for registering in GPS location, it is characterized in that: described GPS positioning auxiliary information comprises latitude and longitude information and course angle information.

Images