CN103021021B - Adopt the generalized stereopair three-dimensional rebuilding method of variance components estimate - Google Patents

Abstract

Adopt the generalized stereopair three-dimensional rebuilding method of variance components estimate, relate to a kind of three-dimensional rebuilding method, in order to solve cause due to wide-angle under the different imaging system between same image each pixel resolution difference and different image-forming condition image, differences in resolution is very large time, there is the problem with the generalized linear system of equations of serious unbalancedness in the existing generalized stereopair three-dimensional rebuilding method based on RFM.Described method is by utilizing the Pixel-level resolution relational implementation of corresponding impact point to be reconstructed in generalized stereopair variance components estimate method to be embedded in the classical three-dimensional rebuilding method crossed based on rational polynominal model RFM front, realize the weight matrix self-adaptative adjustment in the unknown three-dimensional coordinate process of iterative, thus effectively improve reconstruction accuracy, improve the feasibility that generalized stereopair carries out three-dimensional reconstruction, it is for the three-dimensional reconstruction of generalized stereopair.

Description

Adopt the generalized stereopair three-dimensional rebuilding method of variance components estimate

Technical field

The present invention relates to a kind of three-dimensional rebuilding method, particularly a kind of generalized stereopair three-dimensional rebuilding method adopting variance components estimate.

Background technology

Along with the develop rapidly of remote sensing technology, sensor technology and computer technology, the mankind achieve unprecedented great-leap-forward development in field of observing over the ground.The appearance of the observation platforms such as increasing high-definition remote sensing commercial satellite, military satellite, near space vehicle, space shuttle, scounting aeroplane, making people obtain large scale and high accuracy three-dimensional scene models becomes possibility.

Wherein, utilize remotely-sensed data to carry out three-dimensional reconstruction and also become the study hotspot of remote sensing fields.In early days, people utilize strict sensor model, by accurately obtaining the inside and outside orientation parameter of imaging sensor, the front of the multi-angle stereogram obtained by identical imaging system is crossed, least square technology is utilized to realize the three-dimensional reconstruction of interesting target point, and then, utilize interpolation technique can realize the quick generation of scene DEM on a large scale; Along with high-resolution satellite service providers such as IKONOS2, QuickBird for user provides the rational polynominal model (RationalFunctionModel, RFM) of satellite image, the three-dimensional reconstruction based on RFM obtains to be studied widely.Generally speaking, the development experience three phases based on the three-dimensional reconstruction of RFM: first stage, the stereoscopic imaging technology utilizing front to cross exactly structure least-squares estimation model, adopts given RFM parameter to realize solving of unknown three-dimensional coordinate; Subordinate phase, people come to realise original RFM model may be existed and compare serious systemic error, for this reason, can be realized the raising of reconstruction accuracy by two kinds of approach.First method is by utilizing the ground control point (GroundControlPoint, GCP) with accurate three-dimensional coordinate, adopts the correction of the realizations such as clump adjustment technology (Bundle-Adjustment) to original RFM model; Second method does not need modified R FM model, but be also utilize the GCP with accurate three-dimensional coordinate, by image area or aiming field calculating coordinate change model (revise original RFM front cross inputing or outputing in method), carry out the improvement of realize target point reconstruction accuracy.Phase III, in order to better utilize remotely-sensed data widely, people obtain stereo data under being not limited to identical imaging system, but according to demand, realize three-dimensional reconstruction by combining the remotely-sensed data utilizing different imaging system to have different resolution.But presently used method also only rests on and crosses in the light maintenance of three-dimensional rebuilding method based on RFM front to classics.

But, no matter the three-dimensional rebuilding method based on strict sensor model or the three-dimensional rebuilding method based on RFM, existing method be mostly strict with form the imaging sensor of stereo pair images have except the imaging angle of pitch strictly identical inside and outside orientation parameter, for overlapping region, there is identical scaling relation to make stereogram.This just causes and cannot be fully utilized for the remotely-sensed data under the different image-forming conditions of same observation scene, such as different platform, different angles, different phase.Therefore, just need to study correlation theory, the generalized stereopair realizing utilizing the remote sensing images with overlapping region obtained by different imaging system to form is to complete three-dimensional reconstruction task.The situation that between the same image each pixel resolution difference caused due to wide-angle under the different imaging system and different image-forming condition image, differences in resolution is often very large, classical will become a kind of generalized linear system of equations with serious unbalancedness based on the cross mathematical model of three-dimensional rebuilding method of RFM front.Therefore, in the urgent need to fundamentally solving the generalized stereopair three-dimensional reconstruction problem with serious point unbalancedness.

Summary of the invention

The object of the invention is in order to solve cause due to wide-angle under the different imaging system between same image each pixel resolution difference and different image-forming condition image, differences in resolution is very large time, there is the problem with the generalized linear system of equations of serious unbalancedness in the existing generalized stereopair three-dimensional rebuilding method based on RFM, the invention provides a kind of generalized stereopair three-dimensional rebuilding method adopting variance components estimate.

The generalized stereopair three-dimensional rebuilding method of employing variance components estimate of the present invention, it comprises the steps:

Step one: according to the parameter of imaging sensor, obtains the Pixel-level resolution of impact point to be reconstructed corresponding point in each image of generalized stereopair;

Step 2: the generalized linear system of equations of RFM general purpose transducer Construction of A Model about impact point three-dimensional reconstruction to be reconstructed utilizing each image of described generalized stereopair;

Step 3: utilize the Pixel-level resolution obtained in step one to carry out weight matrix initialization to the generalized linear system of equations constructed in step 2;

Step 4: utilize weighted least square to carry out initial estimation to the three-dimensional coordinate of impact point to be reconstructed;

Step 5: according to the initial value of the three-dimensional coordinate of current weight matrix and impact point to be reconstructed, the increment of the relatively described three-dimensional coordinate initial value of three-dimensional coordinate of impact point to be reconstructed under utilizing weighted least square to obtain current iteration, and obtain the margin of error of current estimation;

Step 6: judge whether the margin of error of described current estimation meets accuracy requirement, if meet, then exports the estimated value of the three-dimensional coordinate of current impact point to be reconstructed, completes three-dimensional reconstruction, if do not meet, then proceed to step 7;

Step 7: according to the margin of error of described current estimation, utilizes variance components estimate method to upgrade current weight matrix, and the estimated value of recycling weighted least square to the three-dimensional coordinate of impact point to be reconstructed upgrades, and proceeds to step 5.

The invention has the advantages that, the present invention is by the true resolution of impact point pixel in each image of labor generalized stereopair and solve the serious unbalanced problem of generalized stereopair three-dimensional reconstruction mathematical model based on RFM in conjunction with variance components estimate, effectively improves the precision of the classical three-dimensional reconstruction algorithm realization generalized stereopair three-dimensional reconstruction based on RFM.Make it possible to fully utilize the three-dimensional reconstruction task of generalized stereopair realization to interesting target point with different image-forming condition, improve the utilization ratio of available data.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the generalized stereopair three-dimensional rebuilding method of employing variance components estimate of the present invention.

Fig. 2 is the analysis schematic diagram of the Pixel-level resolution of the generalized stereopair three-dimensional rebuilding method of the employing variance components estimate described in the specific embodiment of the invention two, and wherein A is expressed as image-position sensor.

Fig. 3 is the analysis schematic diagram of the Pixel-level resolution of the generalized stereopair three-dimensional rebuilding method of the employing variance components estimate described in the specific embodiment of the invention three, and wherein A is expressed as image-position sensor.

Fig. 4 is that the multidate high-resolution satellite image that QuickBird satellite obtains adopts the schematic diagram at the reference mark of the generalized stereopair three-dimensional rebuilding method of employing variance components estimate of the present invention.

Fig. 5 is that the multidate high-resolution satellite image that WorldView2 satellite obtains adopts the schematic diagram at the reference mark of the generalized stereopair three-dimensional rebuilding method of employing variance components estimate of the present invention.

Embodiment

Embodiment one: composition graphs 1 illustrates present embodiment, the generalized stereopair three-dimensional rebuilding method of the employing variance components estimate described in present embodiment,

It comprises the steps:

Step one: according to the parameter of imaging sensor, obtains the Pixel-level resolution of impact point to be reconstructed corresponding point in each image of generalized stereopair;

Step 2: the generalized linear system of equations of RFM general purpose transducer Construction of A Model about impact point three-dimensional reconstruction to be reconstructed utilizing each image of described generalized stereopair;

Step 3: utilize the Pixel-level resolution obtained in step one to carry out weight matrix initialization to the generalized linear system of equations constructed in step 2;

Step 4: utilize weighted least square to carry out initial estimation to the three-dimensional coordinate of impact point to be reconstructed;

Step 5: according to the initial value of the three-dimensional coordinate of current weight matrix and impact point to be reconstructed, the increment of the relatively described three-dimensional coordinate initial value of three-dimensional coordinate of impact point to be reconstructed under utilizing weighted least square to obtain current iteration, and obtain the margin of error of current estimation;

Step 6: judge whether the margin of error of described current estimation meets accuracy requirement, if meet, then exports the estimated value of the three-dimensional coordinate of current impact point to be reconstructed, completes three-dimensional reconstruction, if do not meet, then proceed to step 7;

Step 7: according to the margin of error of described current estimation, utilizes variance components estimate method to upgrade current weight matrix, and the estimated value of recycling weighted least square to the three-dimensional coordinate of impact point to be reconstructed upgrades, and proceeds to step 5.

Weight matrix after initialization in step 3 is for being weighted required weight matrix in least-squares estimation solution procedure; Resolution relation between fundamental purpose is according to stereo pair images eliminates the imbalance problem because image resolution ratio difference causes, and provides better initial value for the optimum component of variance of iterative estimate below

Embodiment two: present embodiment is the further restriction of the generalized stereopair three-dimensional rebuilding method to the employing variance components estimate described in embodiment one,

According to the parameter of imaging sensor in step one, the method obtaining the impact point to be reconstructed Pixel-level resolution of corresponding point in each image of generalized stereopair is:

According to the height H of imaging sensor, instantaneous field of view angle θ, angle of pitch β and image size N*N, the resolution obtaining the kth pixel in the image that this sensor obtains is:

By this resolution Res _kas the Pixel-level resolution of impact point to be reconstructed corresponding point in each image of generalized stereopair, N is positive integer.

Present embodiment according to obtaining the actual conditions of data, first, if the imaging parameters of acquisition imaging sensor that can be very detailed, as shown in Figure 2.In figure, N represents sum of all pixels.The distance facing k point in substar o to Fig. 2 is:

$\begin{array}{c}L=H\·tan(k\frac{2\mathrm{\β}}{N}+\mathrm{\θ}-\mathrm{\β})\\ =H\·tan(\mathrm{\θ}+\frac{2k-N}{N}\mathrm{\β})\end{array}---\left(1\right)$

Wherein, therefore, the resolution of a kth pixel is the distance that the distance facing k point in substar o to Fig. 2 deducts k-1 point):

$\begin{array}{c}{\mathrm{Res}}_k=H\·\tan (\mathrm{\θ}+\frac{2k-N}{N}\mathrm{\β})-H\·\tan (\mathrm{\θ}+\frac{2k-N-2}{N}\mathrm{\β})\\ =H\·\left(\tan \right(\mathrm{\θ}+\frac{2k-N}{N}\mathrm{\β})-\tan (\mathrm{\θ}+\frac{2k-N-2}{N}\mathrm{\β}\left)\right)\end{array}---\left(2\right)$

Above-mentioned is carry out when θ angle is greater than 0, when θ angle is minus time, only need carry out simple modification to above-mentioned formula (2), as follows:

${\mathrm{Res}}_k=H\·\left(tan\right(\left|\mathrm{\θ}+\frac{2k-N-2}{N}\mathrm{\β}\right|)-tan(\left|\mathrm{\θ}+\frac{2k-N}{N}\mathrm{\β}\right|\left)\right)---\left(3\right)$

Aggregative formula (2) and formula (3) have:

Embodiment three: present embodiment is the further restriction of the generalized stereopair three-dimensional rebuilding method to the employing variance components estimate described in embodiment one,

According to the parameter of imaging sensor in step one, the method obtaining the impact point to be reconstructed Pixel-level resolution of corresponding point in each image of generalized stereopair is: according to the average resolution rate Res of the front elevation picture of known imaging sensor _nwith angle of pitch β, the Pixel-level resolution obtaining impact point to be reconstructed corresponding point in each image of generalized stereopair is ${\mathrm{Res}}_{\mathrm{\β}}=\frac{{\mathrm{Res}}_n}{{\cos }^2\mathrm{\β}}.$

When only knowing the situation of imaging sensor average resolution rate and the imaging angle of pitch thereof, as shown in Figure 3;

As shown in Figure 3, front elevation is Res as average resolution rate _nwhen being β with angle of image, figure average resolution rate Res _nwith Res _βbetween there is following geometric relationship:

${\mathrm{Res}}_{\mathrm{\β}}=\frac{{\mathrm{Res}}_n}{{\cos }^2\mathrm{\β}}---\left(5\right)$

Embodiment four: present embodiment is the further restriction of the generalized stereopair three-dimensional rebuilding method to the employing variance components estimate described in embodiment one,

The generalized linear system of equations of impact point three-dimensional reconstruction to be reconstructed described in step 2 is:

$\stackrel{\·\·}{Y}=\stackrel{\·\·}{A}\mathrm{\γ}+\stackrel{\·\·}{E}\⇔\left\{\begin{array}{c}{Y}_1=A_1\mathrm{\γ}+E_1\\ Y_2=A_2\mathrm{\γ}+E_2\\ .\\ .\\ .\\ Y_l=A_l\mathrm{\γ}+E_l\end{array}\right.,$ Wherein l is the number of the platform obtaining observed reading, and an above-mentioned l equation arranges from low to high according to the resolution of the corresponding platform of described equation, i.e. equation Y ₁the resolution of corresponding platform is minimum, equation Y _lthe resolution of corresponding platform is the highest; L is greater than 2, $\stackrel{\·\·}{A}=\[\begin{array}{cccc}{A}_1& A_2& ...& A_l\end{array}\]$ For design matrix;

$\stackrel{\·\·}{E}=\[\begin{array}{cccc}{E}_1& E_2& ...& E_l\end{array}\]$ For measuring error,

$E_i={\left[\begin{array}{cccc}{e}_1& e_2& ...& e_{n_i}\end{array}\right]}^T,E_i\∈R^{n_i\×1},$ Be measuring error corresponding to i-th observed reading, and

E _iwith E _juncorrelated (i ≠ j1≤i, j≤l), described E _iwith E _jfor measuring error $\stackrel{\·\·}{E}=\[\begin{array}{cccc}{E}_1& E_2& ...& E_l\end{array}\]$ Component,

be i-th platform observed reading variance, n _ibe the quantity of the observed reading that i-th platform obtains;

$\stackrel{\·\·}{Y}=\[{\begin{array}{cccc}{Y}_1& Y_2& ...& Y\end{array}}_l\]$ For observation vector, wherein $Y_i=\[\begin{array}{cccc}{y}_{i1}& y_{i2}& ...& y_{{\mathrm{in}}_i}\end{array}\]$ Be the observed reading of i-th platform, i=1,2 ..., l; The covariance matrix of residual error amount is

Described design matrix (designmatrix) is the special purpose matrix in weighted least square, for known in those skilled in the art;

In order to effectively solve the problem of the multi-source generalized stereopair three-dimensional reconstruction based on RFM sensor model, variance components estimate technology is embedded in three-dimensional reconstruction solving model, so just must from its basic mathematical model, namely weighting generalized least square model is analyzed.For a generalized linear system of equations:

Y＝Aγ+E(6)

Wherein, Y=[y ₁y ₂y _n] ^t, Y ∈ R ^{n × 1}for measured value; A ∈ R ^{n × m}for sequency spectrum design matrix; γ=[γ ₁γ ₂γ _m] ^t, γ ∈ R ^{m × 1}for unknown vector to be asked; E=[e ₁e ₂e _n] ^t, E ∈ R ^{n × 1}for measuring error.

If all measured values are all obtain from same platform, then the mathematical expectation of error vector is E{E}=Ο=[00 ... 0] ^t∈ R ^{n × 1}; The covariance matrix of error vector is Q _e=Q _y=σ ²i _n.This situation is called equally accurate (EQualprecision, EQ) pattern.Otherwise if measured value obtains from different platform, be then called unequal accuracy (UnEQualprecision, UEQ) pattern, now, formula (6) can be expressed as the form of multiple EQ pattern:

$\left\{\begin{array}{c}{Y}_1=A_1\mathrm{\γ}+E_1\\ Y_2=A_2\mathrm{\γ}+E_2\\ ...\\ Y_l=A_1\mathrm{\γ}+E_l\end{array}\right.---\left(7\right)$

Now, formula (7) can be expressed as canonical form:

$\stackrel{\·\·}{Y}=\stackrel{\·\·}{A}\mathrm{\γ}+\stackrel{\·\·}{E}---\left(8\right)$

Wherein, $\stackrel{\·\·}{Y}=\[\begin{array}{cccc}{Y}_1& Y_2& ...& Y_l\end{array}\],y=\[\begin{array}{cccc}{y}_{i1}& y_{i2}& ...& y_{{\mathrm{in}}_i}\end{array}\],$ mathematical expectation be zero, its covariance matrix is:

So far, the generalized linear least-squares estimation of formula (8) is:

$\widetilde{\mathrm{\γ}}={\left({\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}\stackrel{\·\·}{A}\right)}^{-1}{\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}\stackrel{\·\·}{Y}---\left(9\right)$

Wherein, design matrix and measured value being known, so, in order to obtain the least-squares estimation of formula (8), just needing to be completed accurately covariance matrix estimation.

Might as well make

Wherein, for known weight or initialization weight.Describe with matrix form, have:

$Q=\underset{i=1}{\overset{n}{\Σ}}{U}_i{\mathrm{\σ}}_i^2=\underset{i=1}{\overset{n}{\Σ}}{U}_i{\mathrm{\θ}}_i---\left(11\right)$

Wherein, $U_i=\left[\begin{array}{ccccc}0& ...& 0& ...& 0\\ .& & .& & .\\ .& ...& .& ...& .\\ .& & .& & .\\ 0& ...& P_i& ...& 0\\ .& & .& & .\\ .& ...& .& ...& .\\ .& & .& & .\\ 0& ...& 0& ...& 0\end{array}\right]$

So, the problem of above-mentioned accurate estimate covariance matrix is just converted to the component of variance how accurately obtained in formula (11) platform resolution is exactly the average resolution rate obtaining image platform.

Embodiment five: present embodiment is the further restriction of the generalized stereopair three-dimensional rebuilding method to the employing variance components estimate described in embodiment one, and in step 3, initialized weight matrix is:

Wherein, for the impact point to be reconstructed row resolution ratio of respective pixel in lowest resolution image and i-th higher resolution image in generalized stereopair, for the impact point to be reconstructed column split rate ratio of respective pixel in lowest resolution image and i-th higher resolution image in generalized stereopair.

Embodiment six: present embodiment is the further restriction of the generalized stereopair three-dimensional rebuilding method to the employing variance components estimate described in embodiment one,

According to the described current margin of error in step 7, variance components estimate method is utilized to the method that current weight matrix upgrades to be:

Step July 1st: according to the margin of error of current estimation $v_k=\stackrel{\·\·}{A}{\hat{X}}_k-\stackrel{\·\·}{Y}={\left[\begin{array}{cccccc}{v}_k^1& v_k^2& ...& v_k^i& ...& v_k^n\end{array}\right]}^T,$ Calculate intermediate variable $r_i=n_i-\frac{tr\left\{\stackrel{\·\·}{A}{\left({\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}{|}_k\stackrel{\·\·}{A}\right)}^{-1}{\stackrel{\·\·}{A}}^T{P}_i^{-1}\right\}}{{\widehat{\mathrm{\θ}}}_k^i};$

Step 7 two: calculate each component of variance ${\widehat{\mathrm{\θ}}}_{k+1}^i={\widehat{\mathrm{\σ}}}_{k+1}^2=\frac{{\left(v_k^i\right)}^T{P}_i^{-1}{v}_k^i}{r_i};$

Step 7 three: the weight matrix after renewal:

Wherein, represent the margin of error of kth time iteration, represent the component of variance of i-th platform observed reading in kth time iteration, P _irepresent the weighted value that in current weight matrix, i-th platform observed reading is corresponding.

In present embodiment, embed the estimation that VCE technology realizes above-mentioned component of variance, complete the accurate estimation to unknown quantity γ in formula (8).Detailed process is as follows:

From first time circulation (k=1), until meet accuracy requirement to stop iteration, order proceeds as follows:

$v_k=\stackrel{\·\·}{A}{\hat{X}}_k-\stackrel{\·\·}{Y}={\left[\begin{array}{cccccc}{v}_k^1& v_k^2& ...& v_k^i& ...& v_k^n\end{array}\right]}^T$

$T_k=v_k^T{v}_k$

$r_i=n_i-\frac{tr\left\{\stackrel{\·\·}{A}{\left({\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}{|}_k\stackrel{\·\·}{A}\right)}^{-1}{\stackrel{\·\·}{A}}^T{P}_i^{-1}\right\}}{{\widehat{\mathrm{\θ}}}_k^i}$

${\widehat{\mathrm{\θ}}}_{k+1}^i=\frac{{\left(v_k^i\right)}^T{P}_i^{-1}{v}_k^i}{r_i}$

${\hat{X}}_{k+1}={\left({\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}{|}_{k+1}\stackrel{\·\·}{A}\right)}^{-1}{Q}_{\stackrel{\·\·}{E}}^{-1}{|}_{k+1}\stackrel{\·\·}{Y}$

Wherein v _krepresent the residual of kth time iteration; T _kfor current accuracy, if be less than threshold accuracy, stop iteration, otherwise continue; r _ifor intermediate variable; with represent the estimated value of i-th component of variance in kth time and kth+1 iteration respectively; for the covariance matrix needed for kth+1 iteration.

In order to verify the validity of the inventive method, the real satellite image construction generalized stereopair with different phase obtained by utilizing different imaging system invention has been checking.Experimental result shows, the present invention is by fundamentally analyzing the pixel resolution of remote sensing images under different image-forming condition, variance components estimate technology is embedded into the solution procedure utilizing generalized stereopair to carry out three-dimensional reconstruction, effectively raise the reconstruction precision of the classical three-dimensional rebuilding method based on RFM, improve the utilization factor of different image-forming condition remotely-sensed data.

During experiment below describes, the inventive method AW-RFM represents, i.e. the abbreviation of AdaptiveWeightRFM-basedmethod.

This experiment used test image is the multidate high-resolution satellite image that QuickBird and WorldView2 satellite obtains, corresponding RFM model parameter is known, have chosen 10 ground control points with Precision Elevation (0.06m precision) and be used for algorithm reconstruction recruitment evaluation, as shown in Figure 4 and Figure 5.

In order to verify the performance of the inventive method, the initialization weight getting the inventive method Chinese style (10) is:

$P=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& {\left({\mathrm{\ρ}}_r\right)}^{\mathrm{\α}}& 0\\ 0& 0& 0& {\left({\mathrm{\ρ}}_c\right)}^{\mathrm{\α}}\end{array}\right],\mathrm{\α}\∈\[\begin{array}{cc}1.0& 2.0\end{array}\]\⋐R---\left(12\right)$

Wherein, ρ _rand ρ _cbe respectively the row, column resolution ratio of low resolution observed value and high resolving power observed value.Note: prerequisite is that in formula (7), low resolution observed value sequence number is forward, and high resolving power observed value sequence number rearward.

In experiment, parameter alpha elects 2.0,1.8,1.6,1.4,1.2 and 1.0 as respectively, can be expressed as AW-RFM20, AW-RFM18 ..., AW-RFM10.These 6 are utilized to adopt the inventive method of different parameters to compare with the classical method based on RFM (C-RFM).

It should be noted that, due to ground control point horizontal coordinate limited precision, so experiment only gives the result of height reconstruction, in order to embody contrast effect, give the correlation data of the difference of elevation at relative first reference mark, 10 reference mark simultaneously, in experiment, each algorithm reconstructed results is all high than reference mark elevation, and namely difference of elevation is equidirectional, and comparability is very strong.

Experimental result is as shown in table 1:

Table 1: rebuild elevation square error (unit: rice)

	C-RFM	AW-RFM10	AW-RFM12	AW-RFM14	AW-RFM16	AW-RFM18	AW-RFM20
								Elevation	1.0863	0.9969	1.1396	1.0987	1.0812	1.0653	2.3549
Difference of elevation	1.1595	1.0993	1.1998	1.1687	1.1557	1.1443	2.3094

The above; be only the embodiment in the present invention; but protection scope of the present invention is not limited thereto; any people being familiar with this technology is in the technical scope given by the present invention; the conversion or replacement expected can be understood, within the protection domain that all should be encompassed in claims of the present invention.

The present invention effectively cannot realize this problem of generalized stereopair three-dimensional reconstruction to the three-dimensional rebuilding method based on RFM of classics and simple deformation thereof, by the research of the classical three-dimensional rebuilding method different stages of development that utilizes remote sensing stereogram front to cross to existing and Related Mathematical Models, and the Pixel-level resolution analysis to remote sensing images under different image-forming condition, finding that key issue that the classical three-dimensional rebuilding method based on RFM effectively cannot realize generalized stereopair three-dimensional reconstruction is cannot the serious imbalance problem of three-dimensional reconstruction mathematical model that causes due to the differences in resolution cannot ignored between generalized stereopair of active balance.For this reason, the method of the invention is by the Pixel-level resolution analysis to remote sensing images under different situations, by variance components estimate (VarianceComponentEstimation, VCE) technology is embedded in the classical iterative process solved based on the three-dimensional reconstruction of RFM, effectively achieve the effective estimation to different resolution observed reading variance in iterative process, thus the mathematical model serious imbalance problem that active balance causes due to resolution difference, achieve a kind of method effectively utilizing generalized stereopair to carry out interesting target point three-dimensional reconstruction, effectively improve the precision of the classical three-dimensional rebuilding method based on RFM.

To consider in remote sensing fields that the classical three-dimensional rebuilding method based on RFM mostly uses and has identical or very close resolution and the image converging angle less forms stereogram, and often need in actual applications to utilize to there is different resolution and converge the larger image in angle and realize cannot obtaining single platform stereogram or the inadequate situation of single platform stereogram reconstruction accuracy to the three-dimensional reconstruction of interesting target point to solve to form generalized stereopair.The present invention proposes a kind of generalized stereopair three-dimensional rebuilding method adopting variance components estimate.Method of the present invention adapts to any generalized stereopair that can obtain RFM sensor model and realizes three-dimensional reconstruction.By comparing of the three-dimensional rebuilding method based on RFM with classics, method of the present invention effectively can improve the precision of generalized stereopair three-dimensional reconstruction really.Fig. 1 gives structured flowchart of the present invention.Wherein focus technology content of the present invention comprises pixel-level image resolution analysis and utilizes variance components estimate technology to realize uneven weighted least square to solve two parts.

Claims (5)

1. adopt the generalized stereopair three-dimensional rebuilding method of variance components estimate, it is characterized in that, it comprises the steps:

Step one: according to the parameter of imaging sensor, obtains the Pixel-level resolution of impact point to be reconstructed corresponding point in each image of generalized stereopair;

Step 2: the generalized linear system of equations of RFM general purpose transducer Construction of A Model about impact point three-dimensional reconstruction to be reconstructed utilizing each image of described generalized stereopair;

Step 3: utilize the Pixel-level resolution obtained in step one to carry out weight matrix initialization to the generalized linear system of equations constructed in step 2;

Step 4: utilize weighted least square to carry out initial estimation to the three-dimensional coordinate of impact point to be reconstructed;

Step 5: according to the initial value of the three-dimensional coordinate of current weight matrix and impact point to be reconstructed, the increment of the relatively described three-dimensional coordinate initial value of three-dimensional coordinate of impact point to be reconstructed under utilizing weighted least square to obtain current iteration, and obtain the margin of error of current estimation;

Step 6: judge whether the margin of error of described current estimation meets accuracy requirement, if meet, then exports the estimated value of the three-dimensional coordinate of current impact point to be reconstructed, completes three-dimensional reconstruction, if do not meet, then proceed to step 7;

Step 7: according to the margin of error of described current estimation, utilizes variance components estimate method to upgrade current weight matrix, and the estimated value of recycling weighted least square to the three-dimensional coordinate of impact point to be reconstructed upgrades, and proceeds to step 5;

In step 3, initialized weight matrix is:

Wherein, for the impact point to be reconstructed row resolution ratio of respective pixel in lowest resolution image and i-th higher resolution image in generalized stereopair, 2≤i≤l, for the impact point to be reconstructed column split rate ratio of respective pixel in lowest resolution image and i-th higher resolution image in generalized stereopair.

2. the generalized stereopair three-dimensional rebuilding method of employing variance components estimate according to claim 1, it is characterized in that, according to the parameter of imaging sensor in step one, the method obtaining the impact point to be reconstructed Pixel-level resolution of corresponding point in each image of generalized stereopair is:

According to the height H of imaging sensor, instantaneous field of view angle θ, angle of pitch β and image size N*N, the resolution obtaining the kth pixel in the image that this sensor obtains is:

By this resolution Res _kas the Pixel-level resolution of impact point to be reconstructed corresponding point in each image of generalized stereopair, N is positive integer.

3. the generalized stereopair three-dimensional rebuilding method of employing variance components estimate according to claim 1, is characterized in that,

According to the parameter of imaging sensor in step one, the method obtaining the impact point to be reconstructed Pixel-level resolution of corresponding point in each image of generalized stereopair is: according to the average resolution rate Res of the front elevation picture of known imaging sensor _nwith angle of pitch β, the Pixel-level resolution obtaining impact point to be reconstructed corresponding point in each image of generalized stereopair is ${\mathrm{Res}}_{\mathrm{\β}}=\frac{{\mathrm{Res}}_n}{{\cos }^2\mathrm{\β}}.$

4. the generalized stereopair three-dimensional rebuilding method of employing variance components estimate according to claim 1, is characterized in that, the generalized linear system of equations of impact point three-dimensional reconstruction to be reconstructed described in step 2 is:

$\stackrel{\·\·}{Y}=\stackrel{\·\·}{A}\mathrm{\γ}+\stackrel{\·\·}{E}\⇔\left\{\begin{array}{c}{Y}_1=A_1\mathrm{\γ}+E_1\\ Y_2=A_2\mathrm{\γ}+E_2\\ \·\\ \·\\ \·\\ Y_l=A_l\mathrm{\γ}+E_l\end{array}\right.,$ Wherein l is the number of the platform obtaining observed reading, and an above-mentioned l equation arranges from low to high according to the resolution of the corresponding platform of described equation, i.e. equation Y ₁the resolution of corresponding platform is minimum, equation Y _lthe resolution of corresponding platform is the highest; L is greater than 2, for design matrix, γ is unknown vector to be asked;

$\stackrel{\·\·}{E}=\left[\begin{array}{cccc}{E}_1& E_2& ...& E_l\end{array}\right]$ For measuring error,

$E_i={\left[\begin{array}{cccc}{e}_1& e_2& ...& e_{n_i}\end{array}\right]}^T,E_i\∈R^{n_i\×1},$ Be measuring error corresponding to i-th observed reading, and

E _iwith E _juncorrelated, i ≠ j, 1≤i≤l, 1≤j≤l, described E _iwith E _jfor measuring error component, be i-th platform observed reading variance, n _ibe the quantity of the observed reading that i-th platform obtains;

for observation vector, wherein be the observed reading of i-th platform, i=1,2 ..., l; The covariance matrix of measuring error is

5. the generalized stereopair three-dimensional rebuilding method of employing variance components estimate according to claim 4, is characterized in that, according to the described current margin of error in step 7, utilizes variance components estimate method to the method that current weight matrix upgrades to be:

Step July 1st: according to the margin of error of current estimation calculate intermediate variable $r_i=n_i-\frac{tr\left\{\stackrel{\·\·}{A}{\left({\stackrel{\·\·}{A}}^T{Q}_{\stackrel{\·\·}{E}}^{-1}{|}_k\stackrel{\·\·}{A}\right)}^{-1}{\stackrel{\·\·}{A}}^T{P}_i^{-1}\right\}}{{\widehat{\mathrm{\θ}}}_k^i},$ for the covariance matrix needed for kth time iteration;

Step 7 two: calculate each component of variance

Step 7 three: the weight matrix after renewal:

Wherein, represent the margin of error of kth time iteration, represent the component of variance of i-th platform observed reading in kth time iteration, P _irepresent the weighted value that in current weight matrix, i-th platform observed reading is corresponding.