CN105160665A - Double-circle sub-template underwater terrain matching method - Google Patents

Abstract

The present invention provides a double-circle sub-template underwater terrain matching method. The method comprises: acquiring a strip by a multi-beam sounding system, forming a two-dimensional elevation array by means of data processing, transforming the two-dimensional elevation array into a gray-scale value, forming a to-be-matched template, calling an original database terrain elevation map, transforming the original database terrain elevation map into a gray-scale map, and forming a search mother map; selecting a relatively large template, selecting two small sub-templates as inscribed circles, determining a positional relationship between the double circles, and performing double-circle sub-map search matching on the mother map; and extracting an image corner point feature amount, matching double-circle sub-maps searched by the mother map with the sub-templates, and acquiring the position of an underwater vehicle. Compared with a conventional single template matching method, the double-circle sub-template underwater terrain matching method provided by the present invention solves the problem of excessively long traversal search matching time; and by using a circular template, the calculation complexity of a corner point feature detection algorithm can be lowered, the influence of problems such as rotation, affine transformation and the like is overcome, the matching time is shortened, the matching precision is improved, and very good generalization and applicability are achieved.

Description

A kind of two dumpling made of glutinous rice flour template underwater terrain matching method

Technical field

The present invention relates to a kind of two dumpling made of glutinous rice flour template underwater terrain matching method, belong to underwater navigation technical field.

Background technology

Landform correlation matching algorithm is applied as far back as aviation aspect, becomes to have carried out Successful tests with cruise missile in opportunity of combat, as the Models in Terrain Aided Navigation of company of British Aerospace office development, referred to as topographic profile matching system (TERCOM); The Sang Diya terrain auxiliary navigation method (SITAN) of U.S. Sang Diya laboratory development, has has domesticly also researched and developed Models in Terrain Aided Navigation, Successful tests on weaponize.

The maturation application of domestic and international terrain match technology mainly concentrates on aviation, ground, since 21 century, ocean resources are more and more important in the strategic position of various countries, and accurately location navigation becomes the emphasis of various countries' research under water, common are inertial navigation, acoustic navigation, berth reckoning navigation etc.Underwater terrain matching technology is developed rapidly in the nearly more than ten years as a kind of assisting navigation technology under water, and the particularly use of multibeam echosounding equipment and the research of high-quality matching algorithm, underwater terrain matching precision increases substantially.At present the external development of terrain match airmanship is under water very fast, as the autonomous underwater vehicle of HUGIN series developed by FFI; AUV62F and Sapphires of KTH of Sweden development.

Underwater terrain matching algorithm is the research emphasis of Chinese scholars, has the ICCP algorithm based on single beam echosounding; Based on the matching algorithm of sea-bottom contour; Based on the matching algorithm of canon of probability; Based on the method for Kalman filtering; Based on the matching process of cross-correlation.But above-mentioned algorithm or be coupling based on one-dimensional sequence, quantity of information is less, or there is the problem such as anti-rotation, translation, and mostly adopts square single mode plate to mate, and there is the problems such as match time is long.Two dumpling made of glutinous rice flour template matching method overcomes that single mode plate quantity of information is in the past few, match time long problem, and circular shuttering can overcome the problems affect such as rotation, for underwater terrain matching provides a kind of new matching process.

Summary of the invention

Technology of the present invention is dealt with problems and is: for the deficiencies in the prior art, proposes a kind of two dumpling made of glutinous rice flour template underwater terrain matching method.

Technical solution of the present invention is: the two dumpling made of glutinous rice flour template underwater terrain matching method of the one that the present invention proposes, realizes through the following steps:

(1) band obtained by multibeam sounding system, forms two-dimentional elevation array by data processing, and is converted into gray-scale value, form template to be matched, transfers raw data base landform altitude figure and is converted into gray-scale map, forms search supergraph;

(2) choose a larger template, more therefrom choose two little subtemplates, make incircle, determine the position relationship between two circle, form two dumpling made of glutinous rice flour template model, on supergraph, then carry out two dumpling made of glutinous rice flour graph search coupling;

(3) extract image angle point patterns amount, the two dumpling made of glutinous rice flour figure searched by supergraph mate with subtemplate, obtain the position of underwater hiding-machine.

The present invention also comprises following characteristics:

1. step (1) is specially: the two-dimentional elevation array obtain multibeam echosounding interpolation and the elevation array of raw data base carry out gray processing, and be converted into gray level image, transform mode is as follows:

$p_{i,j}=round(\frac{\left|h_{i,j}\right|-\underset{i,j}{\min }\left\{\right|h_{i,j}\left|\right\}}{\underset{i,j}{\max }\left\{\right|h_{i,j}\left|\right\}-\underset{i,j}{\min }\left\{\right|h_{i,j}\left|\right\}}\×255)$

Wherein, h_ (i, j) represents the water depth value of corresponding two-dimensional array point, and round (*) expression rounds nearby, and p_ (i, j) represents the gray-scale value of this some correspondence

2. step (2) is specially:

(1) choose in the template set up in real time there are relative position relation, discrepant two subtemplates of landform tool;

(2) make the incircle of subtemplate, the grid that home position is formed by institute's interpolation is determined;

(3) by determining the distance in two centers of circle and determining in conjunction with course to be designated as L (ρ, θ) by the position relationship of two circle.

3. in step (3), adopt SIFT Corner Feature operator extraction characteristic quantity, specific as follows:

Piece image metric space function is defined as:

L(x,y,σ)＝G(x,y,σ)*I(x,y)

_(x ²+y ²)/2σ ²

$G(x,y,\mathrm{\σ})=\frac{1}{2{\mathrm{\π\σ}}^2}e$

Wherein, G (x, y, σ) is changeable scale Gaussian function; σ is the metric space factor; L (x, y, σ) is metric space.

First, set up Gaussian difference scale space, and choose metric space extreme point as unique point; Gaussian function convolution under image and different scale is obtained gaussian pyramid, and adjacent metric space subtracts each other and obtains Gaussian difference scale space, is defined as:

D(x,y,σ)＝(G(x,y,kσ)-G(x,y,σ))*I(x,y)＝L(x,y,kσ)-L(x,y,σ)

Pixel between Gauss's yardstick difference empty and this layer are compared with two-layer up and down, if maximal value or minimum value, then records, alternatively unique point;

For ensureing that unique point has rotational invariance, according to the unique point gradient magnitude of each pixel and statistical conditions in direction in done circle shaped neighborhood region, be unique point specific characteristic direction, formula is as follows:

$m(x,y)=\sqrt{{\left(L\right(x+1,y)-L(x-1,y\left)\right)}^2+{\left(L\right(x,y+1)-L(x,y-1\left)\right)}^2}$

$\mathrm{\θ}(x,y)={\tan }^{-1}\left(\frac{L(x,y+1)-L(x,y-1)}{L(x+1,y)-L(x-1,y)}\right)$

M (x, y) and θ (x, y) is (x, y) place modulus value and gradient direction;

The unique point obtained being generated SIFT feature vector, carries out dimension-reduction treatment to traditional SIFT algorithm, take incircle as statistics neighborhood, generates SIFT feature vector, for reducing illumination effect and being normalized, forms final matching characteristic vector; Then adopt Euclidean distance to carry out the tolerance of proper vector, determine matched position.

The present invention's advantage is compared with prior art:

(1) the present invention utilizes multibeam echosounding to obtain two-dimensional array, one-dimensional sequence was compared in the past, obtain terrain information amount larger, improve and have matched precision, add the adaptability of matching area, and array switching for two-dimentional elevation be gray level image, utilize images match correlation matching algorithm to carry out underwater terrain matching, widened the method for underwater terrain matching.

(2) the present invention adopts two dumpling made of glutinous rice flour template to replace single mode plate in the past to mate, and shortens match time, simultaneously because the subtemplate landform chosen has some difference, ensure that the precision of coupling, reduce error hiding region.

(3) the present invention adopts and mates by extracting Corner Feature, simultaneously owing to adopting incircle, simplifies the complexity adopting SIFT Corner Feature algorithm, shortens match time, solve anti-rotation problem, improve matching precision.

Accompanying drawing explanation

Fig. 1 is artificially generated terrain three-dimensional plot;

Fig. 2 be elevation array switching after gray-scale map

Fig. 3 is two dumpling made of glutinous rice flour template gray figure

Fig. 4 is two dumpling made of glutinous rice flour template matches schematic diagram

Embodiment

The two dumpling made of glutinous rice flour template underwater terrain matching method of the one that the present invention proposes, following concrete mode is adopted to realize: the band that (1) is obtained by multibeam sounding system, two-dimentional elevation array is formed by data processing, and be converted into gray-scale value, form template to be matched, transfer raw data base landform altitude figure and be converted into gray-scale map, form search supergraph; Transfer a certain data of national marine scientific library, by process formation rule grid array, landform three-dimensional plot is as Fig. 1.

The two-dimentional elevation array obtain multibeam echosounding interpolation and the elevation array of raw data base carry out gray processing, and be converted into gray level image, transform mode is as follows:

$p_{i,j}=round(\frac{\left|h_{i,j}\right|-\underset{i,j}{\min }\left\{\right|h_{i,j}\left|\right\}}{\underset{i,j}{\max }\left\{\right|h_{i,j}\left|\right\}-\underset{i,j}{\min }\left\{\right|h_{i,j}\left|\right\}}\×255)$

Wherein, h_ (i, j) represents the water depth value of corresponding two-dimensional array point, and round (*) expression rounds nearby, and p_ (i, j) represents the gray-scale value of this some correspondence, and the gray-scale map after conversion is as Fig. 2.

(2) choose in the template set up in real time there are relative position relation, discrepant two subtemplates of landform tool, make the incircle of subtemplate, the grid that home position is formed by institute's interpolation is determined, by determining the distance in two centers of circle and determining the position relationship of two circle in conjunction with course, be designated as L (ρ, θ), incircle exterior pixel point zero setting.The gray scale degree image that two dumpling made of glutinous rice flour template is formed is as Fig. 3.

(3) extract SIFT Corner Feature amount, mate, coupling figure is as Fig. 4, and concrete mode is as follows:

Piece image metric space function is defined as:

L(x,y,σ)＝G(x,y,σ)*I(x,y)

-(x ²+y ²)/2σ ²

$G(x,y,\mathrm{\σ})=\frac{1}{2{\mathrm{\π\σ}}^2}e$

Wherein, G (x, y, σ) is changeable scale Gaussian function; σ is the metric space factor; L (x, y, σ) is metric space.

First, set up Gaussian difference scale space, and choose metric space extreme point as unique point; Gaussian function convolution under image and different scale is obtained gaussian pyramid, and adjacent metric space subtracts each other and obtains Gaussian difference scale space, is defined as:

D(x,y,σ)＝(G(x,y,kσ)-G(x,y,σ))*I(x,y)＝L(x,y,kσ)-L(x,y,σ)

Pixel between Gauss's yardstick difference empty and this layer are compared with two-layer up and down, if maximal value or minimum value, then records, alternatively unique point;

For ensureing that unique point has rotational invariance, according to the unique point gradient magnitude of each pixel and statistical conditions in direction in done circle shaped neighborhood region, be unique point specific characteristic direction, formula is as follows:

$m(x,y)=\sqrt{{\left(L\right(x+1,y)-L(x-1,y\left)\right)}^2+{\left(L\right(x,y+1)-L(x,y-1\left)\right)}^2}$

$\mathrm{\θ}(x,y)={\tan }^{-1}\left(\frac{L(x,y+1)-L(x,y-1)}{L(x+1,y)-L(x-1,y)}\right)$

M (x, y) and θ (x, y) is (x, y) place modulus value and gradient direction;

The unique point obtained being generated SIFT feature vector, carries out dimension-reduction treatment to traditional SIFT algorithm, take incircle as statistics neighborhood, generates SIFT feature vector, for reducing illumination effect and being normalized, forms final matching characteristic vector; Then adopt Euclidean distance to carry out the tolerance of proper vector, determine matched position.

Claims (4)

1. a two dumpling made of glutinous rice flour template underwater terrain matching method, is characterized in that, comprise the following steps:

(1) band obtained by multibeam sounding system, forms two-dimentional elevation array by data processing, and is converted into gray-scale value, form template to be matched, transfers raw data base landform altitude figure and is converted into gray-scale map, forms search supergraph;

(2) choose a larger template, more therefrom choose two little subtemplates, make incircle, determine the position relationship between two circle, form two dumpling made of glutinous rice flour template model, on supergraph, then carry out two dumpling made of glutinous rice flour graph search coupling;

(3) extract image angle point patterns amount, the two dumpling made of glutinous rice flour figure searched by supergraph mate with subtemplate, obtain the position of underwater hiding-machine.

2. the two dumpling made of glutinous rice flour template underwater terrain matching method of one according to claim 1, it is characterized in that, two dumpling made of glutinous rice flour template model is set up and is realized by following concrete steps:

(1) choose in the template set up in real time there are relative position relation, discrepant two subtemplates of landform tool;

(2) make the incircle of subtemplate, the grid that home position is formed by institute's interpolation is determined;

(3) by determining the distance in two centers of circle and determining in conjunction with course to be designated as L (ρ, θ) by the position relationship of two circle.

3. the two dumpling made of glutinous rice flour template underwater terrain matching method of one according to claim 1, it is characterized in that, image angle point patterns is SUSAN operator, Harris operator and SIFT operator.

4. the two dumpling made of glutinous rice flour template underwater terrain matching method of the one according to any one of claim 1-3, it is characterized in that, image angle point patterns adopts two dumpling made of glutinous rice flour templates of SIFT operator, and specific implementation step is as follows:

(1) two-dimentional elevation array multibeam echosounding interpolation obtained and the elevation array of raw data base carry out gray processing, and be converted into gray level image, transform mode is as follows:

$p_{i,j}=round(\frac{\left|h_{i,j}\right|-{\min }_{i,j}\left\{\right|h_{i,j}\left|\right\}}{{\max }_{i,j}\left\{\right|h_{i,j}\left|\right\}-{\min }_{i,j}\left\{\right|h_{i,j}\left|\right\}}\×255)$

Wherein, h _i,jrepresent the water depth value of corresponding two-dimensional array point, round (*) expression rounds nearby, p _i,jrepresent the gray-scale value of this some correspondence;

(2 pairs of multi-beams are swept side in real time and are carried out process to the data gathered and form template, choose two dumpling made of glutinous rice flour templates with position relationship L (ρ, θ) and terrain differences, start to search for two dumpling made of glutinous rice flour figure successively mate original supergraph;

(3) SIFT operator extraction characteristic quantity is adopted, specific as follows:

Piece image metric space function is defined as:

$\begin{array}{c}L(x,y,\mathrm{\σ})=G(x,y,\mathrm{\σ})*I(x,y)\\ G(x,y,\mathrm{\σ})=\frac{1}{2{\mathrm{\π\σ}}^2}{e}^{-(x^2+y^2)/2{\mathrm{\σ}}^2}\end{array}$

Wherein, G (x, y, σ) is changeable scale Gaussian function; σ is the metric space factor; L (x, y, σ) is metric space;

First, set up Gaussian difference scale space, and choose metric space extreme point as unique point; Gaussian function convolution under image and different scale is obtained gaussian pyramid, and adjacent metric space subtracts each other and obtains Gaussian difference scale space, is defined as:

D(x,y,σ)＝(G(x,y,kσ)-G(x,y,σ))*I(x,y)＝L(x,y,kσ)-L(x,y,σ)

Pixel between Gauss's yardstick difference empty and this layer are compared with two-layer up and down, if maximal value or minimum value, then records, alternatively unique point;

For ensureing that unique point has rotational invariance, according to the unique point gradient magnitude of each pixel and statistical conditions in direction in done circle shaped neighborhood region, be unique point specific characteristic direction, formula is as follows:

$m(x,y)=\sqrt{{\left(L\right(x+1,y)-L(x-1,y\left)\right)}^2+{\left(L\right(x,y+1)-L(x,y-1\left)\right)}^2}$

$\mathrm{\θ}(x,y)={\tan }^{-1}\left(\frac{L(x,y+1)-L(x,y-1)}{L(x+1,y)-L(x-1,y)}\right)$

M (x, y) and θ (x, y) is (x, y) place modulus value and gradient direction;

The unique point obtained being generated SIFT feature vector, carries out dimension-reduction treatment to traditional SIFT algorithm, take incircle as statistics neighborhood, generates SIFT feature vector, for reducing illumination effect and being normalized, forms final matching characteristic vector; Then adopt Euclidean distance to carry out the tolerance of proper vector, determine matched position.