CN104077760A - Rapid splicing system for aerial photogrammetry and implementing method thereof - Google Patents

Abstract

A rapid splicing method for aerial photogrammetry comprises the following steps of 101, acquiring image figure information photographed aerially by an unmanned aerial vehicle; 102, performing distortion correction on the image figure information, wherein the image figure information includes a width, a pixel size, a resolution ratio and camera distortion parameters of image figures; 103, performing registering splicing processing on the image figure information subjected to distortion correction through an SIFT algorithm; 104, uniformizing colors of the image figure information subjected to splicing processing to obtain a seamless spliced image figure. By means of the rapid splicing method for aerial photogrammetry, image splicing can be achieved quickly, and emergency guarantee can be provided for all kinds of applications. The rapid splicing method for aerial photogrammetry is suitable for splicing images acquired by the unmanned aerial vehicle under all kinds of conditions, and the shortcomings of long processing time spent on large data volume, slow splicing processing speed and the like can be overcome. By adding control points simply, high-accuracy spliced images can be produced, and elevation analysis is facilitated.

Description

A kind of aerophotogrammetric quick splicing system and its implementation

Technical field

The present invention relates to aerophotogrammetric technical field, especially relate to a kind of aerophotogrammetric quick splicing system and its implementation.

Background technology

Along with the development of photogrammetric measurement technology, aerial survey system has been brought into play huge advantage at aspects such as improving the surveying and mapping result trend of the times, enhancing mapping emergency service supportability, especially unmanned plane low latitude photogrammetric measurement technology.

Unmanned plane low latitude photogrammetric measurement can quick obtaining high resolution image; Yet be subject to the restriction of unmanned plane during flying height, digital camera focal length, spatial resolution, make the field range of individual image less, only rely on individual image, be difficult to whole survey region to make overall cognitive; Image joint refers to two width or several sequence images is carried out overlapping according to its public part that distributes, obtain the intersection figure that a width is new, figure after synthetic not only facilitates the global effect of our visual institute sector of observation, but also has retained the detailed information in raw video.Image joint mainly comprises Image Matching and two key links of visual fusion.In order to obtain the complete image in whole region, must realize registration and the splicing of multiple unmanned plane images, to obtain the complete image in whole region.The requirement to geography information real-time along with accident and emergency guarantee, unmanned aerial vehicle remote sensing images demand real-time and nearly real-time processing is outstanding day by day.The defects such as and to have automaticity not high due to unmanned plane image processing system in existing technology, restrictive condition is more, exists big data quantity to process consuming time longer, and splicing processing speed is slow.

Summary of the invention

The object of the invention is to design a kind of aerophotogrammetric quick splicing system and its implementation, address the above problem.

To achieve these goals, the technical solution used in the present invention is as follows:

An aerophotogrammetric quick joining method, comprises the steps:

Step 101, obtains the image graphical information that unmanned plane is taken photo by plane;

Step 102, carries out distortion correction by described image graphical information; Described image figure packets of information is drawn together fabric width, pixel size, resolution and the camera distortion parameter of image figure;

Step 103, carries out Registration and connection processing by the described image graphical information after distortion correction in conjunction with SIFT algorithm;

Step 104, the described image graphical information after splicing is processed is carried out even look, obtains seamless spliced image figure.

Preferably, in step 102, the concrete methods of realizing of described distortion correction is to utilize described camera distortion parameter, by polynomial expression, the distortion coordinate of actual graphical is proofreaied and correct and is obtained calibration coordinate; The concrete account form of described calibration coordinate is:

a wherein _ij, b _ijfor undetermined coefficient; The degree of polynomial of n for setting up according to the deformation extent of figure, x, y are actual transverse and longitudinal coordinate figure, and u, v are the transverse and longitudinal coordinate figure after proofreading and correct, and i, j are natural number.

Preferably, in step 103, described Registration and connection mainly comprises:

Step 103.1, if in described unmanned plane without POS data, adopt SIFT algorithm to extract the unique point of the same name of adjacent image graphical information, according to described unique point of the same name, set up the mapping relations between described unique point of the same name; Utilize RANSAC to reject the unique point of matching error, optimized the coupling of mapping relations between described unique point of the same name; According to the unique point described of the same name after optimizing, set up affine Transform Model, obtain transformation parameter, complete the splicing of image figure; After having spliced, carry out step 104;

Step 103.2, if there are POS data in described unmanned plane, adopts the coordinate information based on POS to splice; Coordinate information based on POS splice concrete grammar for the image latitude and longitude information that provides according to POS by described image figure information definition corresponding to described image latitude and longitude information to WGS84 coordinate system, form coordinate image figure; According to the coordinate information of described coordinate system, described coordinate image figure is spliced; After having spliced, carry out step 104;

Preferably, the specific implementation computing method of the described affine Transform Model in step 103.1 are:

$\left[\begin{array}{c}{x}_1\\ y_1\end{array}\right]=\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]\left[\begin{array}{c}{x}_2\\ y_2\end{array}\right]+\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right];$

X wherein ₁, y ₁represent the pixel coordinate before the information conversion of image figure; x ₂, y ₂pixel coordinate after the information conversion of expression image figure; $\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]$ Be expressed as the rotation of image graphical information, the synthetic conversion of flexible, shear; $\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right]$ For translation vector; m ₂represent the displacement of image graphical information in level; m ₅represent the displacement in image graphical information vertical direction; m ₀, m ₁, m ₃, m ₄, m ₂and m ₅be real number.

Preferably, after splicing is processed in step 103.2, also comprise that generation image graphical information, for preliminary splicing image graphical information, adopts SIFT merging features algorithm to become more meticulous and mate splicing described preliminary splicing image graphical information; Producing image graphical information is ultimate splicing image graphical information; After coupling splicing, carry out step 104.

Preferably, in step 103.1, the splicing concrete grammar that completes image figure is:

S1, in metric space, adopts SIFT algorithm image graphical information to be carried out to the detection of extreme point; Obtain position and the residing described metric space of key point of key point; Described metric space is for describing the feature of image figure information multi-scale; The computing method of described metric space are:

L(x,y,σ)＝G(x,y,σ)I(x,y)；

The computing method of the difference of Gaussian metric space of the changeable scale based on described metric space are:

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

Wherein L (x, y, σ) is graphical rule space, the difference of Gaussian metric space that G (x, y, σ) is changeable scale, and (x, y) is the pixel coordinate of image graphical information, σ is the metric space factor; Described image graphical information size and the smoothed degree of image graphical information are directly proportional;

S2, compares the sampled point in described difference of Gaussian metric space or 8 consecutive point 9*2 points corresponding with neighbouring yardstick, in order to take a decision as to whether extreme point;

If what 9*2 the point that described sampled point or described 8 consecutive point are corresponding with described neighbouring yardstick compared comes to the same thing, described sampled point or described 8 consecutive point are extreme point, and identical described sampled point or described 8 consecutive point are as candidate's key point;

If the result that 9*2 the point that described sampled point or described 8 consecutive point are corresponding with described neighbouring yardstick compared is not identical, continue relatively;

S3, centered by described candidate's key point, with the gradient direction of the neighborhood territory pixel of candidate's key point described in statistics with histogram, and to utilize the distribution character of described candidate's key point neighborhood territory pixel gradient direction be described candidate's key point direction initialization parameter; By adding up of the value of described direction parameter, form key point;

S4, adopt the Euclidean distance of the proper vector of key point to measure as the similarity determination in the image graphical information of splicing, if the minimum distance of the key point in the image graphical information of described splicing in key point and image pattern information to be matched, divided by the key point in key point in image graphical information and image pattern information to be matched time closely, the quotient obtaining is less than certain proportion threshold value, forms coupling key point;

S5, completes described preliminary splicing image graphical information is become more meticulous and mates splicing according to described coupling key point.

Preferably, in S3, the distribution character that utilizes described candidate's key point neighborhood territory pixel gradient direction is described candidate's key point direction initialization parameter; By adding up of the value of described direction parameter, the concrete grammar that forms key point is:

M1, centered by described candidate's key point, with the gradient direction of the neighborhood territory pixel of candidate's key point described in statistics with histogram;

M2, the distribution character that utilizes described candidate's key point neighborhood territory pixel gradient direction is described candidate's key point assigned direction parameter;

M3, take coordinate axis as benchmark, and centered by described candidate's key point, 8 * 8 the field of getting is as sample window;

M4 samples in described sample window, forms sampled point;

M5, the mode by the relative direction of described sampled point and described candidate's key point by Gauss's weighting merges to and comprises in 8 direction histograms, forms Gauss's weighted graph;

M6, is described described Gauss's weighted graph with 2 * 2 * 8 form; Described each form represents a pixel of the metric space at described candidate's key point neighborhood place; The relative direction of described candidate's key point represents the direction of pixel gradient; Direction by described pixel and described gradient forms gradient vector;

M7 builds 4 * 4 window in described sample window; In described 4 * 4 window, described gradient vector is cumulative, form key point.

Preferably, in step 104, the concrete grammar that the image graphical information after splicing is processed merges is for adopting small wave converting method splicing to process, and described small wave converting method splicing is processed and carried out even look.

Preferably, the processing mode that the image graphical information of described small wave converting method splicing after processing carried out even look comprises and merges the splicing line of described adjacent image graphical information and the heterochromia of the described adjacent image graphical information of elimination.

An aerophotogrammetric quick splicing system, comprises image pretreatment module, Image registration module and visual fusion module;

Described image pretreatment module is mainly used in that image graphical information is carried out to deformity and proofreaies and correct processing;

Described Image registration module is mainly used in overlapping according to adjacent image graphical information on the same space position

Region, is associated described adjacent image graphical information;

Described visual fusion module is mainly used in the described adjacent image graphical information of association to carry out seamless spliced.

Explanation of nouns:

(1) SIFT algorithm, be by D.G.Lowe, in 1999, to be proposed the earliest, it is a kind of algorithm of computer vision, the locality feature being used in detecting and description image, it finds extreme point in space scale, and extracts its position, yardstick, rotational invariants.The patent owner of this algorithm is UBC.-SIFT (Scale-invariant feature transform) operator, i.e. yardstick invariant features conversion.

(2) abbreviation that RANSAC is RANdomSampleConsensus, it is the sample data collection that comprises abnormal data according to a group, calculates the mathematical model parameter of data, obtains the algorithm of effective sample data.

(3) WGS84:World Geodetic System1984 is to use for GPS GPS the coordinate system of setting up.

Beneficial effect of the present invention can be summarized as follows:

By the present invention, can complete fast image joint and work, for types of applications is done emergency guarantee; Be applicable to unmanned plane image joint in various situations; Solved efficiently the defects such as big data quantity processing is consuming time longer, and splicing processing speed is slow, the present invention can be simply by the interpolation at reference mark, and the splicing image of production degree of precision is convenient to Height Analysis.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of a kind of aerophotogrammetric quick joining method of the present invention.

Fig. 2 completes the schematic flow sheet of the joining method of image figure in the present invention.

Fig. 3 is the schematic flow sheet that utilizes the distribution character formation key point of described candidate's key point neighborhood territory pixel gradient direction in the present invention.

Embodiment

In order to make technical matters solved by the invention, technical scheme and beneficial effect clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.

A kind of aerophotogrammetric quick joining method as shown in Figure 1, comprises the steps:

Step 101, obtains the image graphical information that unmanned plane is taken photo by plane;

Step 102, carries out distortion correction by image graphical information; Image figure packets of information is drawn together fabric width, pixel size, resolution and the camera distortion parameter of image figure; The concrete methods of realizing of distortion correction is to utilize camera distortion parameter, by polynomial expression, the distortion coordinate of actual graphical is proofreaied and correct and is obtained calibration coordinate; The concrete account form of calibration coordinate is: a wherein _ij, b _ijfor undetermined coefficient; The degree of polynomial of n for setting up according to the deformation extent of figure, x, y are actual transverse and longitudinal coordinate figure, and u, v are the transverse and longitudinal coordinate figure after proofreading and correct, and i, j are natural number;

Step 103, carries out Registration and connection processing by the image graphical information after distortion correction in conjunction with SIFT algorithm; Registration and connection mainly comprises:

Step 103.1, if in unmanned plane without POS data, adopt SIFT algorithm to extract the unique point of the same name of adjacent image graphical information, according to unique point of the same name, set up the mapping relations between unique point of the same name; Utilize RANSAC to reject the unique point of matching error, optimized the coupling of mapping relations between unique point of the same name; According to the unique point of the same name after optimizing, set up affine Transform Model, obtain transformation parameter, complete the splicing of image figure; After having spliced, carry out step 104; The specific implementation computing method of affine Transform Model are: $\left[\begin{array}{c}{x}_1\\ y_1\end{array}\right]=\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]\left[\begin{array}{c}{x}_2\\ y_2\end{array}\right]+\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right];$ X wherein ₁, y ₁represent the pixel coordinate before the information conversion of image figure; x ₂, y ₂pixel coordinate after the information conversion of expression image figure; $\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]$ Be expressed as the rotation of image graphical information, the synthetic conversion of flexible, shear; $\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right]$ For translation vector; m ₂represent the displacement of image graphical information in level; m ₅represent the displacement in image graphical information vertical direction; m ₀, m ₁, m ₃, m ₄, m ₂and m ₅be real number.

Step 103.2, if there are POS data in unmanned plane, adopts the coordinate information based on POS to splice; Coordinate information based on POS splice concrete grammar for the image latitude and longitude information that provides according to POS by image figure information definition corresponding to image latitude and longitude information to WGS84 coordinate system, form coordinate image figure; According to the coordinate information of coordinate system, coordinate image figure is spliced; Producing image graphical information is preliminary splicing image graphical information, adopts SIFT merging features algorithm to become more meticulous and mate splicing preliminary splicing image graphical information; Producing image graphical information is ultimate splicing image graphical information; After coupling splicing, carry out step 104;

Step 104, the image graphical information after splicing is processed adopts small wave converting method splicing to process the even look of laggard row, obtains seamless spliced image figure; The processing mode that image graphical information after small wave converting method splicing is processed is carried out even look comprises the splicing line that merges adjacent image graphical information and the heterochromia of eliminating adjacent image graphical information.

As shown in Figure 2, wherein, in step 103.1, the splicing concrete grammar that completes image figure is:

S1, in metric space, adopts SIFT algorithm image graphical information to be carried out to the detection of extreme point; Obtain position and the residing metric space of key point of key point; Metric space is for describing the feature of image figure information multi-scale; The computing method of metric space are:

L(x,y,σ)＝G(x,y,σ)I(x,y)；

The computing method of the difference of Gaussian metric space of the changeable scale based on metric space are:

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

Wherein L (x, y, σ) is graphical rule space, the difference of Gaussian metric space that G (x, y, σ) is changeable scale, and (x, y) is the pixel coordinate of image graphical information, σ is the metric space factor; Image graphical information size and the smoothed degree of image graphical information are directly proportional;

S2, compares the sampled point in difference of Gaussian metric space or 8 consecutive point 9*2 points corresponding with neighbouring yardstick, in order to take a decision as to whether extreme point;

If what 9*2 the point that sampled point or 8 consecutive point are corresponding with neighbouring yardstick compared comes to the same thing, sampled point or 8 consecutive point are extreme point, and identical sampled point or 8 consecutive point are as candidate's key point;

If the result that 9*2 the point that sampled point or 8 consecutive point are corresponding with neighbouring yardstick compared is not identical, continue relatively;

S3, centered by candidate's key point, with the gradient direction of the neighborhood territory pixel of statistics with histogram candidate key point, and to utilize the distribution character of candidate's key point neighborhood territory pixel gradient direction be candidate's key point direction initialization parameter; By adding up of the value of direction parameter, form key point;

S4, adopt the Euclidean distance of the proper vector of key point to measure as the similarity determination in the image graphical information of splicing, if the minimum distance of the key point in the image graphical information of splicing in key point and image pattern information to be matched, divided by the key point in key point in image graphical information and image pattern information to be matched time closely, the quotient obtaining is less than certain proportion threshold value, forms coupling key point;

S5, completes preliminary splicing image graphical information is become more meticulous and mates splicing according to coupling key point.

As shown in Figure 3, wherein, in S3, the distribution character that utilizes candidate's key point neighborhood territory pixel gradient direction is candidate's key point direction initialization parameter; By adding up of the value of direction parameter, the concrete grammar that forms key point is:

M1, centered by candidate's key point, uses the gradient direction of the neighborhood territory pixel of statistics with histogram candidate key point;

M2, the distribution character that utilizes candidate's key point neighborhood territory pixel gradient direction is candidate's key point assigned direction parameter;

M3, take coordinate axis as benchmark, and centered by candidate's key point, 8 * 8 the field of getting is as sample window;

M4 samples in sample window, forms sampled point;

M5, the mode by the relative direction of sampled point and candidate's key point by Gauss's weighting merges to and comprises in 8 direction histograms, forms Gauss's weighted graph;

M6, is described Gauss's weighted graph with 2 * 2 * 8 form; Each form represents a pixel of the metric space at candidate's key point neighborhood place; The relative direction of candidate's key point represents the direction of pixel gradient; Direction by pixel and gradient forms gradient vector;

M7 builds 4 * 4 window in sample window; In 4 * 4 window, gradient vector is cumulative, form key point.

An aerophotogrammetric quick splicing system, comprises image pretreatment module, Image registration module and visual fusion module;

Image pretreatment module is mainly used in that image graphical information is carried out to deformity and proofreaies and correct processing;

Image registration module is mainly used according to the overlapping region of adjacent image graphical information on the same space position,

Adjacent image graphical information is associated;

Visual fusion module is mainly used in the adjacent image graphical information of association to carry out seamless spliced.

The main thought that this system is carried out the quick splicing of aerial survey image is:

First original unmanned plane image being carried out to pre-service is that image distortion is proofreaied and correct, and then according to data cases and achievement, requires to select treatment scheme.Selection mode one without POS data in the situation that, according to the thinking based on SIFT merging features method, image sequence is processed, utilize the unique point of the same name of the adjacent image of SIFT operator extraction, and set up the mapping relations between same place, utilizing RANSAC to reject exterior point, the coupling of same place is further optimized, finally according to the same place after optimizing, set up affine Transform Model, determine corresponding transformation parameter, complete image joint, finally by Wavelet Transform, complete visual fusion, obtain a seamless spliced intersection figure.When having POS data and requiring to publish picture in real time emergent demand, the map projection splicing of employing mode two based on POS, the image latitude and longitude information providing according to POS arrives image projecting under WGS84 coordinate system, according to coordinate information, image is directly spliced to processing, finally by Wavelet Transform, complete visual fusion, obtain a seamless spliced intersection figure.When having POS data and require to make high precision achievement data, adopt mode three, first the splicing of the map projection based on POS obtains preliminary splicing effect figure, then adopt based on SIFT merging features method image is carried out to essence coupling, finally by Wavelet Transform, complete visual fusion, obtain a seamless spliced intersection figure.

Mode one is applicable to without the unmanned plane image joint under POS data cases; Mode two fast, efficiently, can obtain splicing achievement in scene in real time, but splicing precision is poor; It is high that mode three is spliced into fruit precision, slow compared with mode two speed.Total system meets the splicing work for the treatment of of various unmanned plane image data situations, and speed is fast, efficiency is high, precision is high.Below each treatment step is done to lower detailed explanation:

Unmanned plane image sequence: read unmanned plane image data and obtain the data such as image pixel size, resolution, camera distortion parameter, the subsequent treatment that can complete image by these is worked.

Image distortion is proofreaied and correct: native system adopts polynomial expression calibration model to carry out image distortion correction, the thought of this algorithm is not consider to cause the complicated factor of the concrete each side of geometric distortion, directly by a polynomial expression, the distortion coordinate figure of calibration coordinate value and actual graphical is connected, comprised the distortion factor of various complexity.Polynomial expression calibration model is as follows:

$u=\underset{i=0}{\overset{n}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-i}{\mathrm{\Σ}}}{a}_{\mathrm{ij}}{x}^i{y}^j$

$u=\underset{i=0}{\overset{n}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-i}{\mathrm{\Σ}}}{b}_{\mathrm{ij}}{x}^{i}y$

A wherein _ij, b _ijfor undetermined coefficient; N is the degree of polynomial, depends on the deformation extent of figure, and x, y are actual coordinate value, and u, v are coordinate figure after proofreading and correct, and during computing, as ideal coordinates value, participate in calculating, and the two is one-to-one relationship.

Map projection's splicing based on POS: the image latitude and longitude information providing according to POS system under WGS84 coordinate system, is directly spliced processing according to coordinate information to image by image projecting.

SIFT unique point is to extracting: first SIFT algorithm carries out extreme point detection at metric space, and position and the residing yardstick thereof of definite key point (Key points), then use the principal direction of key point neighborhood gradient as the direction character of this point, to realize the independence of operator to yardstick and direction.

Metric space is used for describing image data Analysis On Multi-scale Features, and Gaussian convolution core is the unique linear kernel that realizes change of scale, and the metric space of image is defined as:

L(x,y,σ)＝G(x,y,σ)I(x,y)

Wherein L (x, y, σ) is graphical rule space, the Gaussian function that G (x, y, σ) is changeable scale:

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

(x, y) is the pixel coordinate of image, and σ is the metric space factor, and its size and the smoothed degree of image are directly proportional.

Utilize Gaussian difference pyrene and the image convolution of different scale to generate difference of Gaussian metric space G (x, y, σ), then to each sampled point of difference of Gaussian metric space and whether compare with 8 consecutive point of yardstick and 9*2 point corresponding to neighbouring yardstick be extreme point, as key point candidate point.Then utilize three-dimensional quadratic function matching accurately to determine characteristic point position and yardstick.By the gradient direction distribution feature of key point neighborhood territory pixel, carry out again the direction parameter of designated key point.Coordinate axis is rotated to be to key point direction, centered by key point, get 8 * 8 window, then on every 4 * 4 grid, calculate the gradient orientation histogram of 8 directions and the accumulated value of each gradient direction, can form a key point.

Characteristic matching: after the SIFT proper vector of two width images generates, next step adopts the Euclidean distance of key point proper vector as the similarity determination tolerance of key point in two width images.Get certain key point in Image1, and find out nearest key point (NN) and the distance time near key point (SCN) of Euclidean in itself and adjacent image Image2, in these two key points, if minimum distance, except being closely less than certain proportion threshold value in proper order, is accepted this pair of match point.Reduce this proportion threshold value, SIFT match point number can reduce, but more stable.

Utilize the unique point centering of SIFT algorithmic match to exist the point of some mistake couplings right.RANSAC(Random Sample Consensus) be a kind of stochastic sampling algorithm that has the robustness of fault-tolerant ability, can effectively estimate the mathematical model parameter of random sample collection, for Feature Points Matching, can effectively reject exterior point.After RANSAC algorithm is purified, can reject the unique point pair of a large amount of matching errors, so the smooth execution of this step plays a very important role the splicing of image.

Computational transformation parameter: after the Feature Points Matching of the same name completing between image, just can select suitable geometric transformation model, and carry out parameter in appraising model by the mapping relations of these feature point sets of the same name.Native system adopts affine Transform Model to carry out transformation parameter and resolves.

Affine Transform Model:

$\left[\begin{array}{c}{x}_1\\ y_1\end{array}\right]=\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]\left[\begin{array}{c}{x}_2\\ y_2\end{array}\right]+\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right]$

X wherein ₁, y ₁and x ₂, y ₂represent respectively conversion front and back pixel coordinate, m ₀, m ₁, m ₃, m ₄the yardstick of representative image and rotation amount, m ₂, m ₅representative image displacement in the horizontal and vertical directions respectively.

Visual fusion: adopt Wavelet Transform to eliminate the difference that has some splicing lines and color between each single width image, the method is first by wavelet transformation, to convert image to frequency domain, according to the characteristic of frequency domain, be divided into several different frequency domain segmentations, in each frequency domain segmentation, complete respectively splicing and the fusion of image, finally the image in each frequency domain segmentation is being reassembled into a complete composograph.By the present invention, can complete fast image joint and work, for types of applications is done emergency guarantee; Be applicable to unmanned plane image joint in various situations; Solved efficiently the defects such as big data quantity processing is consuming time longer, and splicing processing speed is slow, the present invention can be simply by the interpolation at reference mark, and the splicing image of production degree of precision is convenient to Height Analysis.

More than by the detailed description of concrete and preferred embodiment the present invention; but those skilled in the art should be understood that; the present invention is not limited to the above embodiment; within the spirit and principles in the present invention all; any modification of doing, be equal to replacement etc., within protection scope of the present invention all should be included in.

Claims (10)

1. an aerophotogrammetric quick joining method, is characterized in that, comprises the steps:

Step 101, obtains the image graphical information that unmanned plane is taken photo by plane;

Step 102, carries out distortion correction by described image graphical information; Described image figure packets of information is drawn together fabric width, pixel size, resolution and the camera distortion parameter of image figure;

Step 103, carries out Registration and connection processing by the described image graphical information after distortion correction in conjunction with SIFT algorithm;

Step 104, the described image graphical information after splicing is processed is carried out even look, obtains seamless spliced image figure.

2. aerophotogrammetric quick joining method according to claim 1, it is characterized in that: in step 102, the concrete methods of realizing of described distortion correction is to utilize described camera distortion parameter, by polynomial expression, the distortion coordinate of actual graphical is proofreaied and correct and is obtained calibration coordinate; The concrete account form of described calibration coordinate is: a wherein _ij, b _ijfor undetermined coefficient; The degree of polynomial of n for setting up according to the deformation extent of figure, x, y are actual transverse and longitudinal coordinate figure, and u, v are the transverse and longitudinal coordinate figure after proofreading and correct, and i, j are natural number.

3. aerophotogrammetric quick joining method according to claim 1, is characterized in that: in step 103, described Registration and connection mainly comprises:

Step 103.1, if in described unmanned plane without POS data, adopt SIFT algorithm to extract the unique point of the same name of adjacent image graphical information, according to described unique point of the same name, set up the mapping relations between described unique point of the same name; Utilize RANSAC to reject the unique point of matching error, optimized the coupling of mapping relations between described unique point of the same name; According to the unique point described of the same name after optimizing, set up affine Transform Model, obtain transformation parameter, complete the splicing of image figure; After having spliced, carry out step 104;

Step 103.2, if there are POS data in described unmanned plane, adopts the coordinate information based on POS to splice; Coordinate information based on POS splice concrete grammar for the image latitude and longitude information that provides according to POS by described image figure information definition corresponding to described image latitude and longitude information to WGS84 coordinate system, form coordinate image figure; According to the coordinate information of described coordinate system, described coordinate image figure is spliced; After having spliced, carry out step 104;

4. aerophotogrammetric quick joining method according to claim 1, is characterized in that: the specific implementation computing method of the described affine Transform Model in step 103.1 are:

$\left[\begin{array}{c}{x}_1\\ y_1\end{array}\right]=\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]\left[\begin{array}{c}{x}_2\\ y_2\end{array}\right]+\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right];$

X wherein ₁, y ₁represent the pixel coordinate before the information conversion of image figure; x ₂, y ₂pixel coordinate after the information conversion of expression image figure; $\left[\begin{array}{cc}{m}_0& m_1\\ m_3& m_4\end{array}\right]$ Be expressed as the rotation of image graphical information, the synthetic conversion of flexible, shear; $\left[\begin{array}{c}{m}_2\\ m_5\end{array}\right]$ For translation vector; m ₂represent the displacement of image graphical information in level; m ₅represent the displacement in image graphical information vertical direction; m ₀, m ₁, m ₃, m ₄, m ₂and m ₅be real number.

5. aerophotogrammetric quick joining method according to claim 1, it is characterized in that: after in step 103.2, splicing is processed, also comprise that generation image graphical information, for preliminary splicing image graphical information, adopts SIFT merging features algorithm to become more meticulous and mate splicing described preliminary splicing image graphical information; Producing image graphical information is ultimate splicing image graphical information; After coupling splicing, carry out step 104.

6. aerophotogrammetric quick joining method according to claim 3, is characterized in that: in step 103.1, the splicing concrete grammar that completes image figure is:

S1, in metric space, adopts SIFT algorithm image graphical information to be carried out to the detection of extreme point; Obtain position and the residing described metric space of key point of key point; Described metric space is for describing the feature of image figure information multi-scale; The computing method of described metric space are:

L(x,y,σ)＝G(x,y,σ)I(x,y)；

The computing method of the difference of Gaussian metric space of the changeable scale based on described metric space are:

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

Wherein L (x, y, σ) is graphical rule space, the difference of Gaussian metric space that G (x, y, σ) is changeable scale, and (x, y) is the pixel coordinate of image graphical information, σ is the metric space factor; Described image graphical information size and the smoothed degree of image graphical information are directly proportional;

S2, compares the sampled point in described difference of Gaussian metric space or 8 consecutive point 9*2 points corresponding with neighbouring yardstick, in order to take a decision as to whether extreme point;

If what 9*2 the point that described sampled point or described 8 consecutive point are corresponding with described neighbouring yardstick compared comes to the same thing, described sampled point or described 8 consecutive point are extreme point, and identical described sampled point or described 8 consecutive point are as candidate's key point;

If the result that 9*2 the point that described sampled point or described 8 consecutive point are corresponding with described neighbouring yardstick compared is not identical, continue relatively;

S3, centered by described candidate's key point, with the gradient direction of the neighborhood territory pixel of candidate's key point described in statistics with histogram, and to utilize the distribution character of described candidate's key point neighborhood territory pixel gradient direction be described candidate's key point direction initialization parameter; By adding up of the value of described direction parameter, form key point;

S4, adopt the Euclidean distance of the proper vector of key point to measure as the similarity determination in the image graphical information of splicing, if the minimum distance of the key point in the image graphical information of described splicing in key point and image pattern information to be matched, divided by the key point in key point in image graphical information and image pattern information to be matched time closely, the quotient obtaining is less than certain proportion threshold value, forms coupling key point;

S5, completes described preliminary splicing image graphical information is become more meticulous and mates splicing according to described coupling key point.

7. aerophotogrammetric quick joining method according to claim 1, is characterized in that: in S3, the distribution character that utilizes described candidate's key point neighborhood territory pixel gradient direction is described candidate's key point direction initialization parameter; By adding up of the value of described direction parameter, the concrete grammar that forms key point is:

M1, centered by described candidate's key point, with the gradient direction of the neighborhood territory pixel of candidate's key point described in statistics with histogram;

M2, the distribution character that utilizes described candidate's key point neighborhood territory pixel gradient direction is described candidate's key point assigned direction parameter;

M3, take coordinate axis as benchmark, and centered by described candidate's key point, 8 * 8 the field of getting is as sample window;

M4 samples in described sample window, forms sampled point;

M5, the mode by the relative direction of described sampled point and described candidate's key point by Gauss's weighting merges to and comprises in 8 direction histograms, forms Gauss's weighted graph;

M6, is described described Gauss's weighted graph with 2 * 2 * 8 form; Described each form represents a pixel of the metric space at described candidate's key point neighborhood place; The relative direction of described candidate's key point represents the direction of pixel gradient; Direction by described pixel and described gradient forms gradient vector;

M7 builds 4 * 4 window in described sample window; In described 4 * 4 window, described gradient vector is cumulative, form key point.

8. aerophotogrammetric quick joining method according to claim 1, it is characterized in that: in step 104, the concrete grammar that image graphical information after splicing is processed merges is for adopting small wave converting method splicing to process, and described small wave converting method splicing is processed and carried out even look.

9. aerophotogrammetric quick joining method according to claim 7, is characterized in that: the processing mode that the image graphical information after described small wave converting method splicing is processed is carried out even look comprises the splicing line that merges described adjacent image graphical information and the heterochromia of eliminating described adjacent image graphical information.

10. an aerophotogrammetric quick splicing system, is characterized in that, comprises image pretreatment module, Image registration module and visual fusion module;

Described image pretreatment module is mainly used in that image graphical information is carried out to deformity and proofreaies and correct processing;

Described Image registration module is mainly used in overlapping according to adjacent image graphical information on the same space position

Region, is associated described adjacent image graphical information;

Described visual fusion module is mainly used in the described adjacent image graphical information of association to carry out seamless spliced.