CN103129752B - Dynamic compensation method for attitude angle errors of optical remote sensing satellite based on ground navigation - Google Patents

Abstract

The invention discloses a dynamic compensation method for attitude angle errors of an optical remote sensing satellite based on ground navigation. The dynamic compensation method includes the following steps: (1) performing effectiveness processing on ephemeris data and attitude data downloaded from the satellite, namely eliminating outliers points in the data and re-determining the values of the points; (2) determining an extracting range of control points on a first level image which is an image corrected by radiation, determining an extracting range of control points on a reference image according to the extracting range on the first level image and data after processing in the step (1), intercepting the extracting ranges on the first level image and the reference image respectively, and generating one sub-image of the first level image which is an image subjected to radiometric correction and one sub-image of the reference image which is an orthoimage; (3) performing automatic pairing of the control points by adopting intercepted the sub-images of the first level image and the reference image, and eliminating mispairing points; (4) selecting at least 3 pairs of homonymy control points with different sequence orders and smallest Euclidean distances apart from a scenario center from the sub-images of the first level image and the reference image, and calculating attitude angle errors; and (5) performing compensation on an attitude angle of the optical remote sensing satellite by adopting determined attitude angle errors.

Description

A kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation

Technical field

The invention belongs to Optical remote satellite data processing field, be applicable to optical remote sensing camera.

Background technology

Along with the raising of domestic remote sensing satellite spatial resolution, the geometric positioning accuracy of remote sensing picture becomes an important inspection target.The positioning precision of remote sensing picture mainly affects by timing tracking accuracy on orbital data precision, attitude data precision and star.Adopt current orbit determination technology, as double frequency differential GPS orbit determination accuracy, can to reach meter level even higher; On the star of current satellite, timing tracking accuracy can reach Microsecond grade, and if timing tracking accuracy on star is 1ms, satellite velocities is 7.6km/s, and the satellite position deviation caused by time irreversibility is 0.44m; But it is larger that satellite surveys appearance difficulty, systematic error in attitude angle, long-period error and random error etc. are larger on the impact of geometric positioning accuracy, automatically, in real time, calibration accurately goes out the error of attitude data and compensates, and image geometry positioning precision can be made to have a more substantial increase.

Publish an article and open source information from current, the research in the calibration and compensation of attitude data error, mainly contains following content:

1, " the satellite remote-sensing image systematic error compensation based on bias matrix " (opening, Liaoning Project Technology University's journal, in August, 2007)

Solve three independent parameters in excursion matrix respectively, it can be used as systematic error to compensate.

First utilize camera facing the image under condition, carry out geometric correction according to rigorous geometry model, add up along rail direction error Δ N ' P and vertical rail direction error calculating Δ ρ and Δ Ψ according to formula (1) and formula (2) is exactly pitch angle in excursion matrix and roll angle.Then utilize the image of camera under maximum side-looking condition and substitute into and walk the excursion matrix obtained, carry out correcting without the image geometry at controlling point according to rigorous geometry model, add up the error along track alignment solving Δ κ according to formula (3), is exactly the yaw angle in excursion matrix.

$\mathrm{\Δ}\tilde{N}P(\frac{(1+\frac{H}{R})\cos \mathrm{\ψ}}{\sqrt{1-{(1+\frac{H}{R})}^2{\sin }^2\mathrm{\ψ}}}-1)\mathrm{R\Δ\ψ}---\left(1\right)$

$\mathrm{\Δ}{N}^{\′}P=\{H+R[1-\cos \left(\arcsin (1+\frac{H}{R})\sin \left(\mathrm{\ψ}\right)\right)-\mathrm{\ψ})\left]\right\}$

${\sec }^2\left(\mathrm{\ρ}\right)\mathrm{\Δ\ρ}=\{H+R[1-\cos (\arcsin (1+\frac{H}{R})\sin \left(\mathrm{\ψ}\right))-\mathrm{\ψ})\left]\right\}\mathrm{\Δ\ρ}---\left(2\right)$

$\mathrm{\Δ}\tilde{P}Q=\left(R(\arcsin \left((1+\frac{H}{R})\sin \left(\mathrm{\ψ}\right)\right)-\mathrm{\ψ})\right)\cos \mathrm{\κ\Δ\κ}---\left(3\right)$

H is orbit altitude, and R is earth radius

2, " the constant angular error calibration of high resolution ratio satellite remote-sensing image " (Yuan Xiuxiao etc., mapping journal, in February, 2008)

1) math modeling of attitude angle calibration

Image coordinate (the x of known control point _c, y _c) and coordinate (X under WGS84 system of axes _c, Y _c, Z _c), utilize image row-coordinate y _c, carry out according to the elements of exterior orientation discrete observation value that secondary file provides the principal point coordinate that polynomial interpolation can obtain respective scanned row and attitude angle initial value when controlling point and line element survey precision enough high time, μ in formula (4) _cthe exact value of picture point standardised space auxiliary coordinate can be thought, in this, as the foundation of attitude angle calibration.

${\mathrm{\μ}}_c=\frac{{{\mathrm{\μ}}^{\′}}_c}{\left|\right|{{\mathrm{\μ}}^{\′}}_c\left|\right|}$

In formula, ${{\mathrm{\μ}}^{\′}}_c=\left[\begin{array}{c}{X}_c-X_{P_i}\\ Y_c-Y_{P_i}\\ Z_c-Z_{P_i}\end{array}\right].$

By three components of image space auxiliary coordinate, following error equation group can be listed to single controlling point.

μ in formula _sfor normalisation image space auxiliary coordinate, $L=\left[\begin{array}{c}{\left({\mathrm{\μ}}_c\right)}_X{-\left({\mathrm{\μ}}_S\right)}_X\\ {\left({\mathrm{\μ}}_c\right)}_Y-{\left({\mathrm{\μ}}_S\right)}_Y\\ {\left({\mathrm{\μ}}_c\right)}_Z-{\left({\mathrm{\μ}}_S\right)}_Z\end{array}\right]$ For μ _cwith μ _sdifference between each component.

With as attitude angle initial value for people's formula (5), its augmentation Δ ω can be obtained through 3-4 iterative _i, Δ κ _iboth be added just obtain this spot scan row attitude angle or value, i.e. calibration value.

2) a small amount of controlling point is utilized to calculate constant angular error

When the attitude angle observed value that Image-aided file provides exists constant, it is estimated can to utilize a small amount of dominating pair of vertices.Because the attitude angle of each baseline in image all exists the constant communicated, when single controlling point can only be found on a scanning line image, then visual 1) attitude angle augmentation (the Δ ω calculated in _i, Δ κ _i) be constant value valuation; When n controlling point can be found on the n bar scanning line image of image, then can respectively to each controlling point according to 1) in method calculate the attitude angle augmentation of each baseline, finally get valuation (the Δ ω of attitude angle augmentation center line average values as the constant angular error of image of each baseline Δ κ) after, the attitude angle value being added to each baseline that formula (6) calculates just can be eliminated the impact of its constant.For avoiding constant to leave data noise, resection method can be adopted to do further adjustment processing, now generally needing more than 6 controlling points.

In formula, a ₀, b ₀, c ₀, d ₀, e ₀, f ₀, centered by the elements of exterior orientation of baseline; a ₀..., f _kfor multinomial coefficient, the elements of exterior orientation discrete observation value matching that can provide according to auxiliary data file obtains; T is the time of exposure that the i-th baseline is capable relative to centre scan.

3, " the constant angular error calibration of ALOS PRISM image " (Liu Chubin etc., Surveying and mapping technology journal, the 28th volume the 4th phase in 2011)

Constant angular error calibration model:

Known control point coordinate (X _c, Y _c, Z _c) and principal point coordinate (X _s, Y _s, Z _s) time, then the picture point standardised space auxiliary coordinate calculated by controlling point coordinate and principal point coordinate is

u＝[X _c-X _s，T _c-Y _s，Z _c-Z _c] (7)

Unit vector after its normalisation is

$u_c=\frac{u}{\left|\right|u\left|\right|}$

When controlling point coordinate and principal point co-ordinate measurement accuracy enough high time, then the u that formula (8) can be calculated _cbe used as the true value of picture point standardised space auxiliary coordinate.

By the computing value u of picture point standardised space auxiliary coordinate _sthe observational equation of class is

u _c＝u _s＝R _XYR _GASTR _PNR(pitch)R(roll)R(yaw)u ₁(9)

R _xYfor Ghandler motion matrix, R _pNfor precession of the equinoxes nutating matrix, R _gASTfor rotation on Sunday matrix.

Take attitude error into account, treated as constant and process.By 3 components of image space auxiliary coordinate, following error equation group can be arranged to single controlling point:

$\left[\begin{array}{c}{V}_X\\ V_Y\\ V_Z\end{array}\right]={\left[\begin{array}{c}\frac{\∂\left(u_s\right)}{\∂\left(\mathrm{yaw}\right)}\\ \frac{\∂\left(u_s\right)}{\∂\left(\mathrm{roll}\right)}\\ \frac{\∂\left(u_s\right)}{\∂\left(\mathrm{pitch}\right)}\end{array}\right]}^T\left[\begin{array}{c}\mathrm{\Δyaw}\\ \mathrm{\Δroll}\\ \mathrm{\Δpitch}\end{array}\right]-L---\left(10\right)$

Δ pitch, Δ yaw and Δ roll are attitude angle correction; L is u _cwith u _sdifference between each component, V is observed value correction column matrix.Iterative pitch, yaw and roll.

4, " the satellite attitude angle systematic error compensation based on rigorous geometry model " (Lei Yufei etc., spacecraft engineering, in February, 2012)

Under body series, error vector E is

E(Δφ _P，Δφ _r，Δφ _y，ξ)＝V _boby-U _body(11)

φ _p, φ _r, φ _yfor pitch angle, roll angle and yaw angle, Δ φ _p, Δ φ _r, Δ φ _yfor φ _p, φ _r, φ _ycompensation.V _body, U _bodyunder being respectively body coordinate system, actual observation is vectorial; ξ is other uncertain factors causing position error.Utilize a small amount of controlling point (>=2), by attitude angle φ _p, φ _r, φ _ydirect adjustment, like actual observation vector V _bodywith desirable measurement vector U _bodybetween angle minimum, thus realize the compensation of attitude angle systematic error.

Analyze above-mentioned data and find to there is following problem:

The first, research contents mainly concentrates on and extracts controlling point manually, the systematic error of calibration attitude angle.Installation situation and the analysis of principle of work aspect of appearance device (as star sensor, gyro, infrared horizon instrument etc.) is surveyed from star, its attitude data measured contains systematic error, long-period error and random error etc., only calibration and compensation are carried out to systematic error, limitation is improved to the geometric positioning accuracy of image, is difficult to the application requirement meeting high-resolution remote sensing image geometric positioning accuracy.The impact of the factor such as long-period error, random error on geometric positioning accuracy also can not be ignored.

The second, in the process of attitude error calibration, the selection range at controlling point is excessive.Due to the defect of optics design and processing, always there is certain geometric distortion in optical camera, its value increases from primary optical axis gradually to field of view edge, particularly wide visual field camera, can reach tens even tens pixels near the distortion of field of view edge position.If the controlling point chosen is away from primary optical axis, its deviations is jointly caused by the factor of attitude error and lens distortion two aspect, and both decoupling zeros are very difficult.The attitude error that this kind of situation solves also is inaccurate.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, and provide a kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation, the method can increase substantially the geometric positioning accuracy of remote sensing picture.

Technical solution of the present invention is: a kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation, and step is as follows:

(1) almanac data passed up and down star and attitude data carry out validity process, namely reject the outlier in data and redefine the value of this point;

(2) the extraction scope at controlling point is determined from first order image, and according to the Control point extraction scope of the data determination reference picture after process in this extraction scope and step (1), Control point extraction scope on above-mentioned two images is intercepted out respectively, generates first order image subgraph and reference picture subgraph; Described first order image is the image after radiant correction, and reference picture is orthograph picture;

(3) utilize the first order image subgraph intercepting out and reference picture subgraph to carry out controlling point Auto-matching, and carry out mispairing process;

(4) from selected distance scape center Euclidean distance first order image subgraph and reference picture subgraph minimum at least 3 to row a number different controlling point of the same name carry out the calculating of attitude error value;

(5) the attitude error value determined is utilized to compensate Optical remote satellite attitude angle.

In described step (2), on first order image, the extraction scope at controlling point is determined to meet following three principles simultaneously:

A Control point extraction scope should with the scape line of centers along rail direction for rotational symmetry, and the rail direction yardstick that hangs down is as far as possible little; The out to out in rail direction of hanging down is camera optics parts radial distortion absolute value apart from the difference of primary optical axis pixel row farthest number and scape center point range number within a pixel;

B in a scape image as far as possible little along rail direction yardstick;

The pixel number that c Control point extraction scope comprises should meet the needs of controlling point Auto-matching.

In described step (2), the Control point extraction scope determining step of reference picture is as follows:

(2.1) ranks number of four angle points of the Control point extraction scope that first order image is determined are extracted respectively, utilize rigorous geometry model to carry out forward projection, calculate the longitude and latitude of four angle points;

(2.2) scope delimited with the longitude of the latitude of upper left angle point and bottom right angle point, lower-left angle point and upper right angle point;

(2.3) scope that step (2.2) delimited is extended out at least 50 meters along longitude and latitude direction, obtain the extraction scope at reference picture controlling point.

The implementation procedure of described step (5) is as follows:

In whole rail, each attitude data measures the attitude angle compensation value in moment to utilize the attitude error value of the scape image calculated in step (4) to determine, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.

The implementation procedure of described step (5) is as follows:

(5.1) from whole rail image, choose many scapes image, every scape image calculates attitude error value according to the step of step (1) ~ (4);

(5.2) validity process is carried out to the attitude error value that step (5.1) calculates, utilize the attitude error value in moment before and after it to carry out interpolation undesirable attitude error value;

(5.3) the attitude error value after step (5.2) being processed carries out interpolation, obtain each attitude data in whole rail image and measure the attitude angle compensation value in moment, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.

The present invention compared with prior art beneficial effect is:

(1) the present invention can automatically, in real time, calibration accurately goes out the error (wherein containing systematic error, long-period error and random error etc.) of attitude data, and the modeling of comprehensive whole rail error condition, realizes the optimization of whole rail attitude data; Thus increase substantially the geometric positioning accuracy of remote sensing picture.

(2) the invention provides the scheme that controlling point selection range is determined, attitude error and the decoupling zero of distortion two factors can be realized, make the controlling point information extracted in this region can characterize the positioning precision deviation caused by the error of attitude angle exactly.

Accompanying drawing explanation

Fig. 1 is diagram of circuit of the present invention;

Fig. 2 is that schematic diagram is determined in Control point extraction region of the present invention;

Fig. 3 is that attitude error value of the present invention resolves diagram of circuit;

Fig. 4 be A rail data of the present invention when taking different appearance rail prioritization scheme level image in the deviations distribution situation of X and Y-direction;

Fig. 5 be B rail data of the present invention when taking different appearance rail prioritization scheme level image in the deviations distribution situation of X and Y-direction.

Detailed description of the invention

The principal element affecting high-resolution optical remote sensing satellite positioning precision comprises timing tracking accuracy on attitude data precision, orbital data precision and star.Adopt current orbit determination technology, as double frequency differential GPS orbit determination accuracy, can to reach meter level even higher; On the star of current satellite, timing tracking accuracy can reach Microsecond grade, and if timing tracking accuracy on star is 1ms, satellite velocities is 7.6km/s, and the satellite position deviation caused by time irreversibility is 0.44m; But it is larger that satellite surveys appearance difficulty, systematic error in attitude angle, long-period error and random error etc. are larger on the impact of geometric positioning accuracy, automatically, in real time, calibration accurately goes out the error of attitude data and compensates, and image geometry positioning precision can be made to have a more substantial increase.

One, method introduction

The present invention is directed to the technology Problems existing mentioned in above-mentioned data, and in conjunction with the level that current satellite is developed, propose a kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation, as shown in Figure 1, concrete steps are as follows:

1 almanac data and the process of attitude data validity

Almanac data and attitude data extract from the auxiliary data that star passes up and down.Due at a time or survey the mode of operation at that time that appearance surveys rail instrument in certain a period of time and may occur exception, and number passes the reasons such as error codes, and almanac data and attitude data may occur outlier.Validity process is exactly outlier is rejected and recalculates the value of this point.

1.1 almanac data validity process

The process of almanac data validity utilizes the data of the almanac data around the moment to this moment to verify, concrete grammar is as follows:

P _iand v _ithe position vector of position satellite in the i-th moment and the observed reading of speed vector respectively.Use p _iand v _i, the position vector of prediction the i-th+1 moment satellite and speed vector, obtain predictor with when satellite is at the position vector observed reading p in the i-th+1 moment _i+1and predictor the mould of difference value vector be less than threshold delta _p, speed vector observed reading v _i+1and predictor the mould of difference value vector be less than threshold delta _vtime, then think p _i+1and v _i+1effectively.Otherwise think invalid.When there is invalid value (outlier), the method for interpolation is adopted to utilize the coordinate of 8 points around to calculate the coordinate figure of this point.

1.2 attitude data validity process

The process of attitude data validity utilizes the data of the attitude data around the moment to this moment to verify, concrete verification method is as follows:

|ξ _i+1-(ξ _i+ξ _i′(t _i+1-t _i))|＜δ (12)

Wherein ξ _iand ξ _i' be attitude data ξ _i+1the attitude data of previous moment and attitude rate, δ is threshold value, can set according to actual conditions.When bad point (outlier) of condition above occurring not meeting, just obtain the value of current point with the point interpolation of surrounding.

2 Control point extraction regions are determined

The method that Control point extraction region is determined is mainly based on following 3 principles:

1) decoupling zero of attitude error and lens distortion, the controlling point information chosen can characterize the positioning precision deviation caused by the error of attitude angle exactly;

2), during the Auto-matching of controlling point, this area division can meet feature point extraction and the calculating needs mated;

3) Control point extraction scope first order image determined is consistent as far as possible with the scope that reference picture is determined.

According to above-mentioned 3 principles, the defining method in Control point extraction region is specific as follows:

First, first scope should be determined from first order image.1) there is distortion in camera lens, increase gradually from primary optical axis to lens edge, the controlling point chosen is the closer to primary optical axis, and the deviations that distortion causes is less, therefore Control point extraction scope should with the scape line of centers along rail direction for rotational symmetry, and the rail direction yardstick that hangs down is as far as possible little.Can according to the mirror image distortion model of camera optics parts, calculate distortion distance primary optical axis pixel row number farthest within a pixel, these row number are arranged with the pixel at scape center and number subtracts each other and take absolute value, in this, as the out to out in vertical rail direction, achieve the decoupling zero of attitude error and lens distortion.2) if the imaging time span of each identical point is large, then solve pose compensation angle by its ranks on first order image number and longitude and latitude on a reference, solving precision can be caused not high, therefore should as far as possible little along rail direction yardstick when delimiting Control point extraction scope along rail direction.3) Auto-matching algorithm in controlling point is when extracting and matching feature points, the point of some is needed to participate in calculating, if the region chosen is too small, there is no enough counting, then can cause feature point extraction or it fails to match, therefore, when Control point extraction scope is determined, can not pursue at the yardstick along rail and vertical rail direction too small simply, first should ensure the needs of the enough feature point extraction of pixel number that this scope comprises and matching primitives.Determine Control point extraction scope corresponding on reference picture for convenience, the extraction scope on first order image is generally rectangle (or square), determines that extraction scope makes it to meet above-mentioned three principles simultaneously by the mode trying to gather.

Then, according to the Control point extraction scope that first order image is determined, determine the Control point extraction scope of reference picture.The ranks number of four angle points of the Control point extraction scope that first order image is determined are extracted respectively, rigorous geometry model is utilized to carry out forward projection, calculate four angle points longitude and latitude (rigorous geometry model be by the multicenter projection approximation of spaceborne line array CCD be single centre projection, the corresponding relation between ground point and imaging point is set up by a series of coordinate transformation in the basis of collinearity condition equation, complete successively by image space observation vector and be tied to sensor coordinate system from focal plane coordinate, satellite body system of axes is tied to from sensor coordinates, from satellite body to orbital coordinate system, from orbit coordinate with the being tied to solid a series of coordinate transformation being, obtain the observation vector under ground is admittedly, be added with the position vector of satellite again, thus obtain ground point location).A scope delimited with the longitude of the latitude of upper left angle point and bottom right, lower-left angle point and upper right angle point.Considering that appearance rail data exist error, for ensureing to try one's best consistent (or at utmost comprising) with the region on corresponding first order image in the region delimited, at least 50 meters need be extended out with reference to this scope that image delimited along longitude and latitude direction.

Determine Control point extraction scope, intercepted out, spanning subgraph picture.

In the present invention, take the method intercepting subgraph from first order image scape immediate vicinity, as shown in Fig. 2 (note: this figure is the schematic diagram that Control point extraction scope is determined, and not true to scale).

3 controlling point Auto-matchings

1) SIFT algorithm carries out controlling point Auto-matching

Controlling point Auto-matching is done with the first order image subgraph intercepted out and reference picture subgraph.Because the atural object on reference picture is relative to first order image has the rotation of certain angle (being greater than 10 °), the present invention when Auto-matching, that selected industry maturation, that matching capacity is stronger SIFT algorithm.This algorithm can process the matching problem occurred between two width images in translation, rotation, affined transformation situation preferably.After the SIFT feature vector of two width images generates, the Euclidean distance of key point proper vector is adopted to measure as the similarity determination of key point in two width images.

SIFT feature coupling have employed simple effective method, and the minimum euclidean distance namely between 128 dimension description vectors of two width image key points is less than certain threshold xi with the ratio of time little Euclidean distance _mctime, think that feature is to being coupling.ξ _mcless, mismatch rate lower.Owing to participating in control of the same name that attitude error angle calculates, to count be 6 (as described below), even if there is a mispairing point, on attitude error angle calculation accuracy impact be also very large, therefore establish ξ in the present invention _mc=0.4.

2) mispairing process is carried out with RANSAC algorithm

Although SIFT algorithmic match ability is comparatively strong, still can not avoid occurring mispairing (when there is the same or similar scenery of large area in image, easily occurring mispairing), so need mispairing process.RANSAC algorithm is adopted to carry out mispairing process in the present invention.RANSAC algorithm has been widely used in eliminating mispairing at present, and it can improve the characteristic matching effect of SIFT algorithm to a certain extent.

RANSAC algorithm is by carrying out multi-times random sampling to matching point, each random taking-up is the least possible but abundant multiple matching point carrys out Confirming model parameter, according to fixed model, all matching points are divided again, part matching point is considered as interior point within the scope of the certain error of this model, and a part of data are considered as exterior point outside error limit.Because exterior point is rambling abnormal data, its determined model, the matching point dropped in error limit occupies the minority, and most of matching point all drops on outside error limit.And to by the determined Model approximation of interior point in real model, most of matching point will drop in error limit.After multi-times random sampling test, RANSAC algorithm finds out the set dropping on matching points maximum in error limit, then does optimization, the parameter of final Confirming model with this set.

3) choose a fixed point to resolve participation attitude error value

After going mispairing, minimum some to (setting its value as n of selected distance scape center Euclidean distance _gCP, in this example, n _gCP=6) a row number different controlling point of the same name participates in the calculating of attitude error value; If the controlling point logarithm of successful match is less than n _gCP, then the attitude error value being considered as this scape calculates unsuccessfully, and the attitude error value of carrying out the next attitude data measurement moment is resolved.

Extract the ranks number (m, n) of controlling point on one-level subgraph, at the longitude and latitude (lon, lat) with reference to subgraph.

4 attitude error values are resolved

Attitude error value is resolved and is adopted forward model method, and calculation process as shown in Figure 3.

Flow process is mainly divided into 2 tunnels, one tunnel is by the ranks information (m at the controlling point gone out by 1 grade of image zooming-out, n), through a series of coordinate transformation, obtain the nominal observation vector u of imaging vector in controlling point in satellite body system of axes under the system of axes of camera focal plane _sAT; Another road is by by the controlling point latitude and longitude information (lon of the same name extracted in reference picture, and altitude figures information (h) extracted from altitude figures file lat), through a series of ordinate transform and Vector operation, obtain the actual observation vector of controlling point of the same name under satellite body system of axes concrete steps are as follows:

1) image space observation vector u is calculated according to the row n at controlling point on first order image, pixel dimension, pixel number and focal distance f _fP

u _FP＝[x/f，y/f，1] ^T。

Wherein x=0 $y=y_{\mathrm{pixel}}\·(n-\frac{N_{\mathrm{pixel}}+1}{2}),$

Y _pixelfor pixel dimension, N _pixelfor pixel number.

2) by image space observation vector u _fPbe converted to the vectorial u under sensor coordinate system _sEN

u _SEN＝T _SEN/FP×u _FP。

Wherein, T _sEN/FPfor focal plane coordinate is tied to the transition matrix of sensor coordinate system.

3) by the vectorial u under sensor coordinate system _sENbe transformed into the vectorial u under satellite body system of axes _sAT

u _SAT＝T _SAT/SEN×u _SI。

Wherein, T _sAT/SENfor sensor coordinates is tied to the transition matrix of satellite body system of axes, i.e. the installation matrix of sensor.

4) be coordinate (X, Y, Z) admittedly with being converted to reference to the elevation information (h) in the latitude and longitude information (lon, lat) at controlling point of the same name on image and altitude figures file _eCR

X＝(N+h)·cos(Lat/180·π)·cos(Lon/180·π)

Y＝(N+h)·cos(Lat/180·π)·sin(Lon/180·π)

Z＝(N+h)·e ²·sin(Lat/180·π)

Wherein, a, b are respectively the long and short axle radius of the earth.e ²＝1-b ²/a ²，

$N=\frac{a}{{(1-e^2\·{\sin }^2(\mathrm{Lat}/180\·\mathrm{\π}))}^{1/2}}.$

5) be coordinate (X, Y, Z) admittedly according to the ground at controlling point of the same name _eCRwith the almanac data (P of this imaging moment _x, P _y, P _z), generate the measurement vector under the solid system in ground

${\hat{u}}_{\mathrm{ECR}}={[X-P_X,Y-P_Y,Z-P_Z]}^T$

6) measurement vector under by ground being admittedly converts the measurement vector under track system to

$\hat{u}\mathrm{ORB}=T_{\mathrm{ORB}/\mathrm{ECR}}\×\hat{u}\mathrm{ECR}$

Wherein, T _oRB/ECRfor ground is tied to the transition matrix of track system admittedly,

$T_{\mathrm{ORB}/\mathrm{ECR}}={\left[\begin{array}{ccc}{\hat{b}}_1& {\hat{b}}_2& {\hat{b}}_3\end{array}\right]}^{-1}$

${\hat{b}}_3=-P/\left|\right|P\left|\right|$

${\hat{b}}_2=({\hat{b}}_3\×V)/\left|\right|{\hat{b}}_3\×V\left|\right|$

${\hat{b}}_1={\hat{b}}_2\×{\hat{b}}_3$

Satellite position vector (the P that P provides for almanac data _x, P _y, P _z), the satellite velocity vector (V that V provides for almanac data _x, V _y, V _z), || || be the compute sign of vector field homoemorphism.

7) by the measurement vector under track system convert the observation vector under satellite body system of axes to

$\hat{u}\mathrm{SAT}=T_{\mathrm{SAT}/\mathrm{ORB}}\×\hat{u}\mathrm{ORB}$

Wherein T _sAT/ORBfor track is tied to the transition matrix of satellite body system,

$T_{\mathrm{ORB}/\mathrm{SAT}}=\left[\begin{array}{ccc}\cos {\mathrm{\ξ}}_Y& \sin {\mathrm{\ξ}}_Y& 0\\ -\sin {\mathrm{\ξ}}_Y& \cos {\mathrm{\ξ}}_Y& 0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos {\mathrm{\ξ}}_R& \sin {\mathrm{\ξ}}_R\\ 0& -\sin {\mathrm{\ξ}}_R& \cos {\mathrm{\ξ}}_R\end{array}\right]\left[\begin{array}{ccc}\cos {\mathrm{\ξ}}_P& 0& -\sin {\mathrm{\ξ}}_P\\ 0& 1& 0\\ \sin {\mathrm{\ξ}}_P& 0& \cos {\mathrm{\ξ}}_P\end{array}\right]$

8) attitude error value (Δ ξ is solved _y, Δ ξ _r, Δ ξ _p)

$\hat{u}\mathrm{SAT}=R(\mathrm{\Δ}{\mathrm{\ξ}}_Y,\mathrm{\Δ}{\mathrm{\ξ}}_R,\mathrm{\Δ}{\mathrm{\ξ}}_P)u_{\mathrm{SAT}}$

Wherein R is by attitude error R (Δ ξ _y, Δ ξ _r, Δ ξ _p) function composition matrix,

$R=\left[\begin{array}{ccc}\cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_Y\right)& -\sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_Y\right)& 0\\ \sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_Y\right)& \cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_Y\right)& 0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_R\right)& -\sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_R\right)\\ 0& \sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_R\right)& \cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_R\right)\end{array}\right]\left[\begin{array}{ccc}\cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_P\right)& 0& \sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_P\right)\\ 0& 1& 0\\ -\sin \left(\mathrm{\Δ}{\mathrm{\ξ}}_P\right)& 0& \cos \left(\mathrm{\Δ}{\mathrm{\ξ}}_P\right)\end{array}\right]$

The compensation of 5 Optical remote satellite attitude angle

Satellite attitude angle compensation in this step is when accuracy requirement is not high, the attitude error value of the scape image calculated in above-mentioned four steps can be adopted to compensate, now need to determine that in whole rail, each attitude data measures the attitude angle compensation value in moment according to the attitude error value calculated, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.In whole rail, the attitude angle compensation value in each attitude data measurement moment can be worth identical with the attitude error calculated.

In order to submit precision to, the compensation process of attitude angle can also adopt following manner to carry out:

(5.1) from whole rail image, choose many scapes image, every scape image calculates attitude error value according to the step of step (1) ~ (4);

(5.2) validity process is carried out to the attitude error value that step (5.1) calculates, utilize the attitude error value in moment before and after it to carry out interpolation undesirable attitude error value;

Although carried out going mispairing process, but still 100% elimination mispairing point can not be ensured, need to carry out validity process to the attitude error value of whole rail.Image position accuracy situation before compensating according to attitude error, in conjunction with the survey appearance of this satellite, orbit measuring precision and timing tracking accuracy, the threshold value of setting attitude error (is set to ζ, such as ζ=0.1 °), when attitude error is less than this value, thinks and calculate correct, otherwise think and calculate unsuccessfully, then abandon this value during failure, need the value of the attitude error in several moment before and after utilizing to carry out interpolation to it.During interpolation, first consider the good interpolation method of smooth degree, adopt linear interpolation when the failure of the method interpolation.

(5.3) the attitude error value after step (5.2) being processed carries out interpolation, obtain each attitude data in whole rail image and measure the attitude angle compensation value in moment, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.

Two, method validation

2.1 verification msg explanations

The present invention verifies that the first order image, almanac data and the attitude data that the adopt digital simulation subsystem by XX high-resolution optical remote sensing satellite (grinding) data processing prototype system is according to the track, sensor parameters etc. of satellite, utilizes geometry imaging model, radiant image model simulation obtains; In addition, also need timing tracking accuracy present situation on the almanac data precision according to current satellite, attitude data precision and star, error is added to almanac data, attitude data.

When verifying the present invention, first order image have chosen the simulate data of the satellite of two rails (calling A rail, B rail in the following text) different technologies state: the spatial resolution of A rail data is 2 meters, almanac data position error is 10 meters (1 σ), and attitude measurement accuracy is 0.003 ° (1 σ); The spatial resolution of B rail data is 1 meter, and almanac data position error is 1 meter (1 σ), and attitude measurement accuracy is 0.003 ° (1 σ) °

Reference picture derives from certain high precision orthography storehouse, and positioning precision is better than 5 meters; Altitude figures is STRM90, and vertical precision is about 10 meters.

2.2 verification method explanations

In a rail image, in order to improve computational efficiency, be not the calculating that each scape all participates in attitude error value, but calculate once every some scapes.In order to verify the optimization ability of the present invention to appearance rail data, select appearance rail prioritization scheme in 4 respectively: 1. every rail only calculates once, 2. calculates once every 20 scapes, 3. calculates once every 10 scapes, 4. calculates once every 5 scapes.Compare the situation that these four kinds of schemes are improved level image positioning precision.Positioning precision method of measurement is: 5 scape level image chosen arbitrarily by every rail, choose 6 controlling points in every scape, contrasts 6 controlling points identical point deviations in the x-direction and the z-direction and circular error on a reference.

2.3 the result

Fig. 4 and Fig. 5 with the form of scatter diagram respectively show A rail and B rail data take different appearance rail prioritization scheme time, level image is in the deviations distribution situation of X and Y-direction; Table 1 illustrates A rail and B rail data when taking different appearance rail prioritization scheme, and level image is in the deviations (root of mean square) of X and Y-direction and circular error (root of mean square).

The deviations of level image when table 1A rail and B rail data take different appearance rail prioritization scheme

2.4 the result analysis and conclusions

1) the result analysis

When not doing appearance rail and optimizing, the setting circle error (root of mean square) of A rail and B rail data level image is respectively 483.36 meters and 398.57 meters (as shown in table 1); At the deviations of X-direction and Y-direction, except there is random error, also there is larger systematic error (as shown in Fig. 4 (a) and Fig. 5 (a)).

After adopting the present invention to carry out the optimization of appearance rail, the deviations of A rail and B rail data level image obviously reduces.Along with the increase of attitude error value calculated rate, the systematic error of location and random error all have further reduction.When frequency increases to " calculating an attitude error value every 10 scapes ", the systematic error of location is eliminated substantially, and residual random error is also less, and error amount is more stable, without the point that deviations is larger.

2) conclusion is verified

The present invention can carry out the optimization of appearance rail to satellite data effectively, and the positioning precision of level image is significantly improved;

For satellite in orbit, can require to determine the frequency that attitude error value calculates to the requirement of positioning precision and computing time according to its real appearance rail data precision and user.

The unspecified part of the present invention belongs to general knowledge as well known to those skilled in the art.

Claims (3)

1., based on an Optical remote satellite attitude error dynamic compensation method for ground navigation, it is characterized in that step is as follows:

(1) almanac data passed up and down star and attitude data carry out validity process, namely reject the outlier in data and redefine the value of this point;

(2) the extraction scope at controlling point is determined from first order image, and according to the Control point extraction scope of the data determination reference picture after process in this extraction scope and step (1), Control point extraction scope on above-mentioned two images is intercepted out respectively, generates first order image subgraph and reference picture subgraph; Described first order image is the image after radiant correction, and reference picture is orthograph picture; On described first order image, the extraction scope at controlling point is determined to meet following three principles simultaneously:

A Control point extraction scope should with the scape line of centers along rail direction for rotational symmetry, and the rail direction yardstick that hangs down is as far as possible little; The out to out in rail direction of hanging down is camera optics parts radial distortion absolute value apart from the difference of primary optical axis pixel row farthest number and scape center point range number within a pixel;

B in a scape image as far as possible little along rail direction yardstick;

The pixel number that c Control point extraction scope comprises should meet the needs of controlling point Auto-matching;

The Control point extraction scope determining step of described reference picture is as follows:

(2.1) ranks number of four angle points of the Control point extraction scope that first order image is determined are extracted respectively, utilize rigorous geometry model to carry out forward projection, calculate the longitude and latitude of four angle points;

(2.2) scope delimited with the longitude of the latitude of upper left angle point and bottom right angle point, lower-left angle point and upper right angle point;

(2.3) scope that step (2.2) delimited is extended out at least 50 meters along longitude and latitude direction, obtain the extraction scope at reference picture controlling point;

(3) utilize the first order image subgraph intercepting out and reference picture subgraph to carry out controlling point Auto-matching, and carry out mispairing process;

(4) from selected distance scape center Euclidean distance first order image subgraph and reference picture subgraph minimum at least 3 to row a number different controlling point of the same name carry out the calculating of attitude error value;

(5) the attitude error value determined is utilized to compensate Optical remote satellite attitude angle.

2. a kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation according to claim 1, is characterized in that: the implementation procedure of described step (5) is as follows:

In whole rail, each attitude data measures the attitude angle compensation value in moment to utilize the attitude error value of the scape image calculated in step (4) to determine, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.

3. a kind of Optical remote satellite attitude error dynamic compensation method based on ground navigation according to claim 1, is characterized in that: the implementation procedure of described step (5) is as follows:

(5.1) from whole rail image, choose many scapes image, every scape image calculates attitude error value according to the step of step (1) ~ (4);

(5.2) validity process is carried out to the attitude error value that step (5.1) calculates, utilize the attitude error value in moment before and after it to carry out interpolation undesirable attitude error value;

(5.3) the attitude error value after step (5.2) being processed carries out interpolation, obtain each attitude data in whole rail image and measure the attitude angle compensation value in moment, the attitude angle compensation value in each measurement moment is added in the attitude angle in corresponding moment, completes the compensation of Optical remote satellite attitude angle.