CN103630120B - Martian surface linear array image core line method for resampling based on tight geometric model - Google Patents

Abstract

The present invention relates to martian surface linear array image core line method for resampling based on tight geometric model, first obtain martian surface line-scan digital camera camera site, attitude relevant parameter, build rigorous geometric model；Calculate picpointed coordinate core line on raw video, obtain a series of discrete point on core line；Linear array stereoscopic image, as perspective plane, is carried out differential and corrects the horizontal image of acquisition by setting Mean height plane；Discrete point on raw video core line is changed to horizontal image, calculates core line geometrical relationship on horizontal image and determine core line direction；Horizontal image carries out along core line direction core line resampling, obtains approximate epipolar image；The present invention is to utilize tight geometric model to combine projected footprint method core line geometry to analyze and the linear array image core line method for resampling of the differential correction horizontal image of generation, core line resampling is carried out to eliminate the impact of geometric distortion, raising matching precision and reliability further on the basis of differential corrects the horizontal image of generation.

Description

Martian surface linear array image core line method for resampling based on tight geometric model

Technical field

The present invention relates to martian surface HRSC camera image core line method for resampling.

Background technology

Mars is the most possible planet finding life vestige in the solar system, and scientist even dreams of that the mankind have one day can Migrate Mars.By orbit detectors such as Mars whole world cruiser MGS, Mars Express MEX, Mars reconnaisance orbit device MRO and The successful execution of the Mars landing detector tasks such as courage number, Opportunity Rover, phoenix number, curiosity number, Mars has been had by the mankind Brand-new understanding.Mars mapping is the basis carrying out Mars research, and obtaining high-resolution Mars terrain data is also to land The premise of device safe landing.Utilize photogrammetric survey method that martian surface stereoscopic image data carries out process and remain current acquisition The main method of Mars terrain data.Mars Express (MarsExpress, the MEX) detector that European Space Agency launches for 2003 carries It is exclusively used in the high-resolution solid line-scan digital camera HRSC of Mars mapping, by HRSC image is carried out Photogrammetric Processing Mars whole world high-resolution terrain data can be obtained.One of key problem of HRSC image photographic Measurement and Data Processing is linear array The high accuracy of image, quick, dense Stereo Matching, utilizing the constraint of core line geometry is the effective ways solving this problem.

Due to the particularity of line-scan digital camera row central projection, the core line geometry relation of linear array image is more multiple than frame width formula image Miscellaneous.Kim et al. has derived linear array image core line equation based on the tight geometric model of collinearity equation, but this formula form is complicated, And be non-linear form, it is difficult to actual application.It is satellite linear array image core during constant that Habib et al. have studied speed with attitude Line geometry relation.Morgan et al. utilizes parallel projection model to be studied satellite image core line method for resampling.Open ancestral Merit utilizes fitting process to have studied the approximate epipolar arrangement of SPOT film sequence.Gong Danchao have studied line based on rational function model Battle array CCD image horizontal correction method.Hu Fen uses linear model to simplify core curve of the same name, it is proposed that the satellite shadow of a kind of practicality As approximate epipolar resampling fast algorithm, and propose the method generating approximate epipolar image based on thing side's projection reference surface.? Army proposes a kind of linear array satellite image approximate epipolar method for resampling based on rational polynominal coefficient Yu thing side's longitude and latitude forever. When analyzing linear array image core line geometry relation, generally sensor model simplified or carry out certain approximate processing, such as Kim Et al. position and attitude are expressed as second order polynomial form, the core line geometry analysis of Habib et al. is set up in speed and attitude On the premise of keeping being basically unchanged, it is clear that the suitability of these core line geometry models is limited.Visible, line array loudspeakers Core lineation opinion immaturity, do not form unified, general core line geometry formula or core line theoretical model yet, therefore high accuracy, Generate linear array core line image efficiently and be still worth further investigation.

Jiang Wanshou proposes a kind of satellite stereo image pairing approximation core line based on elevation datum and generates method, but The method sets up the coordinate transformation relation of core line image and raw video based on perspective transform, and the present invention is based on collinear condition The tight geometric transformation model of equation, theory is the tightest, and precision is higher.

Summary of the invention

It is an object of the invention to provide a kind of martian surface linear array image core line resampling side based on tight geometric model Method, with high accuracy, efficiently production linear array core line image.

For achieving the above object, the martian surface linear array image core line resampling side based on tight geometric model of the present invention Method step is as follows:

(1) obtain martian surface line-scan digital camera camera site, attitude relevant parameter, build image based on collinearity condition equation Rigorous geometric model；

(2) based on tight geometric model, projected footprint method is utilized to calculate the picpointed coordinate on left image at original right image On core line, core line on left image corresponding to picture point on right image；

(3) setting Mean height plane is as perspective plane, utilizes rigorous geometric model that linear array stereoscopic image is carried out differential and entangles Just obtaining horizontal image；

(4) discrete point on raw video core line is changed on horizontal image, calculate several on horizontal image of core line What relation also determines core line direction；

(5) on horizontal image, carry out core line resampling along core line direction, obtain approximate epipolar image.

Further, described step (1) neutral body camera camera site, attitude relevant parameter include camera focus, pixel Size, pixel calibration coordinate, camera position, attitude data.

Further, described step (1) neutral body camera image rigorous geometric model is

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{c}{X}_S\\ Y_S\\ Z_S\end{array}\right]+{\mathrm{\λR}}_{J2000}^{\mathrm{MARS}}{R}_{\mathrm{star}}^{J2000}{R}_{\mathrm{body}}^{\mathrm{star}}[\left[\begin{array}{c}{d}_x\\ d_y\\ d_z\end{array}\right]+R_{\mathrm{camera}}^{\mathrm{body}}\left[\begin{array}{c}x\\ y\\ -f\end{array}\right]]---\left(1\right)$

Wherein, (x, y) is the picpointed coordinate in linear array image, [X, Y, Z]^TFor Mars topocentric coordinates, [X_S, Y_S, Z_S]T For the Mars probes position after precise orbit determination,It is tied to Mars body-fixed coordinate system spin matrix for J2000 coordinate, The detector attitude measured for star sensor,For the spin matrix of satellite body to star sensor,For camera with defend Installation matrix between star body, [dx, dy, dz]^TFor image center eccentricity component under satellite body.

Further, in described step (3), tight geometric model based on collinearity condition equation sets up horizontal image with vertical Coordinate transformation relation between body image, recycling ground point Inverse Projection sets up the core line geometry relation of corresponding image points.

Further, the optimum scanning line search process that in described ground point Inverse Projection, ground point is corresponding is as follows: (11) Obtain each base line time needed for linear array image tight geometry location model, elements of exterior orientation, camera focus, linear array CCD Visit unit's location parameter；(12) the perspective plane equation of each base line of linear array image is calculated；(13) according to camera projection centre and each Distance D between perspective plane Equation for Calculating each base line perspective plane of base line_i, D_iFor the closest scanning of ground point P is advanced Row Fast estimation；(14) distance d of ground point and Article 1 base line perspective plane is first judged_i, i=1, if d_iMore than two base lines Distance D between perspective plane_i, i=1, then utilize n=d_i/n_iBase line increment n is estimated by formula；(15) iteration judge ground point with Article i-th, the distance on base line perspective plane, until d_i<D_iAnd go to (16th) step；(16) utilize the base line closest with P accurate Interpolation goes out optimum scanning row

The martian surface linear array image core line method for resampling based on tight geometric model of the present invention is to combine projection rail Mark method core line geometry is analyzed and is corrected the linear array image core line method for resampling generating horizontal image with differential, divides in projected footprint method On the basis of analysis linear array image core line geometry relation, first original linear array image is carried out differential and corrects the horizontal image of generation, analyze The core line that projected footprint method draws geometrical relationship on horizontal image also determines approximate epipolar direction, then on horizontal image Carry out core line to reset and generate core line image, this correct at differential that to carry out core line resampling on the basis of the horizontal image of generation permissible Eliminate the impact of geometric distortion further, improve matching precision and reliability.And in projected footprint method and geometric correction The linear array image ground point back projection problem related to, it is proposed that optimum scanning row based on perspective plane, thing side geometrical constraint is quickly searched Rope algorithm, can realize high efficiency, high-precision optimum scanning row quickly positions.Many group HRSC image datas are utilized to test, Result shows, the core line image same place vertical parallax that the method for the present invention generates is better than 1 pixel, on relative level image Carry out core line rearrangement, can effectively change geometric distortion, beneficially Image Matching.

Accompanying drawing explanation

Fig. 1 is frame width formula image core line geometry graph of a relation；

Fig. 2 is projected footprint method linear array image core line geometry graph of a relation；

Fig. 3 is that HRSC camera linear array arranges schematic diagram；

Fig. 4 is to utilize SPICE storehouse to carry out time and coordinate system conversion；

Fig. 5 is that linear CCD image re-projection generates relative level image；

Fig. 6 is horizontal image coker line direction and object coordinates system X-axis angle；

Fig. 7 is the schematic diagram of HRSC linear array image core line method for resampling of the present invention；

Fig. 8 is optimum scanning row fast search process schematic diagram；

Fig. 9 is projected footprint method core line analysis overall contrast figure；

Figure 10 is projected footprint method core line analysis local 1:1 display comparison figure；

Figure 11 is the core line in embodiment corresponding to S2 passage picture point q；

Figure 12 is HRSC approximate epipolar image effect figure；

Figure 13 is HRSC approximate epipolar image corresponding image points vertical parallax schematic diagram.

Detailed description of the invention

1 core line ultimate principle

1.1 frame width formula image core line geometries

Frame width formula image core line geometry relation as it is shown in figure 1, p, q are left and right image corresponding image points, S₁、S₂For taking the photograph station coordinates, S₁With S₂Line constitutes photographic base B, ground point P and S₁、S₂Constitute core face S₁S₂P, S₁S₂P intersects composition core line with left and right image I_pWith I_q, I_pWith I_qFor corresponding epipolar line, corresponding image points p, q must be positioned at corresponding epipolar line I_pWith I_qOn.The core line shadow of frame width formula image Picture can be by realizing initial dip image re-projection to horizontal image (honest image), p_nWith q_nCorrespond respectively to p, q in level Picpointed coordinate on image, I_pnWith I_qnCorresponding to I_pWith I_qProjection on horizontal image, I_pnWith I_qnOn horizontal image mutually Parallel.Picture point on horizontal image is as follows with picture point corresponding relation on initial dip image

$u_n=u\left(\frac{-f}{w}\right)=-f\frac{a_{1}x+a_{2}y-a_{3}f}{c_{1}x+c_{2}y-c_{3}f}$

$v_n=u\left(\frac{-f}{w}\right)=-f\frac{b_{1}x+b_{2}y-b_{3}f}{c_{1}x+c_{2}y-c_{3}f}---\left(3\right)$

Wherein u_n, v_nFor the picpointed coordinate on horizontal image, (x y) is corresponding picpointed coordinate, a on initial dip image₁, a₂..., c₃Included angle cosine for each axis.The core line of frame width formula image has a following characteristic:

(1) core line I_pWith I_qIt is straight line；

(2) the core line I on left image_pOn the most all projections to the corresponding epipolar line I of right image_qOn；

(3) two core line one_to_one corresponding that corresponding image points is corresponding, the institute's the most also one_to_one corresponding on two core lines.

Initial dip image can be carried out core line resampling according to above characteristic and generate core line image, core line image column direction Vertical parallax is zero, by two-dimensional search, coupling can be converted to linear search accordingly.

1.2 linear array image projected footprint method core line geometries

Particularity due to line array loudspeakers row central projection, it is impossible to strict core set up like that by frame width formula image Line geometry relation, generally uses projected footprint method to analyze the core line geometry characteristic of linear array image.Projected footprint method is based on conllinear bar Part equation, theory is the tightest, and its ultimate principle is as in figure 2 it is shown, ground point P(X, Y, Z) image in respectively on the image of left and right P, q, picture point p and ground point P and the projection centre S (X of this base line_si, Y_si, Z_si) may make up a projection ray, this projection Light and object space horizontal plane h_iIntersect and obtain a series of intersection point P_i, by P_iBack projection is to available a series of on right image Point q_i, this series of some q_iConstitute the core curve lq of picture point p, it is clear that the corresponding image points q mono-of p point is positioned on core curve lq. In like manner, by picture point q on right image, projection centre S ' and ground point P constitute projection ray carry out phase with thing side's horizontal plane Handing over, will also can get a series of point in intersection point back projection to left image, the corresponding image points p of the core curve lp, q that constitute q also must So it is positioned on lp.

The concrete formula of projected footprint method core line geometry, if left picture picture point p (x_l, y_l) and this base line projection centre S (X_si, Y_si, Z_si) line constitute projection ray qS, from picture point p point through the light of projection centre S thing side sit Can be to be expressed as form under mark system

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{c}{X}_{\mathrm{si}}\\ Y_{\mathrm{si}}\\ Z_{\mathrm{si}}\end{array}\right]+\mathrm{\&lambda;}\left[\begin{array}{ccc}{r}_{11}& r_{12}& r_{13}\\ r_{21}& r_{22}& r_{23}\\ r_{31}& r_{32}& r_{33}\end{array}\right]\left[\begin{array}{c}{x}_l\\ y_l\\ -f\end{array}\right]---\left(4\right)$

Wherein, r_mn(m, n=1,2,3) is the spin matrix coefficient that this base line is corresponding, and λ is scale factor, by above formula is Can determine that a series of topocentric three-dimensional coordinates on projection ray qS.By in the ground point back projection on qS light to right image, Computing formula is as follows

$x_r=-f\frac{r_{11}^{\&prime;}(X-X_{\mathrm{sj}}^{\&prime;})+r_{21}^{\&prime;}(Y-Y_{\mathrm{sj}}^{\&prime;})+r_{31}^{\&prime;}(Z-Z_{\mathrm{sj}}^{\&prime;})}{r_{13}^{\&prime;}(X-X_{\mathrm{sj}}^{\&prime;})+r_{23}^{\&prime;}(Y-Y_{\mathrm{sj}}^{\&prime;})+r_{33}^{\&prime;}(Z-Z_{\mathrm{sj}}^{\&prime;})}$

$y_r=-f\frac{r_{12}^{\&prime;}(X-X_{\mathrm{sj}}^{\&prime;})+r_{22}^{\&prime;}(Y-Y_{\mathrm{sj}}^{\&prime;})+r_{32}^{\&prime;}(Z-Z_{\mathrm{sj}}^{\&prime;})}{r_{13}^{\&prime;}(X-X_{\mathrm{sj}}^{\&prime;})+r_{23}^{\&prime;}(Y-Y_{\mathrm{sj}}^{\&prime;})+r_{33}^{\&prime;}(Z-Z_{\mathrm{sj}}^{\&prime;})}---\left(5\right)$

Wherein, r'_mn(m, n=1,2,3) is the spin matrix coefficient that right image jth base line is corresponding, (X'_sj,Y'_sj,Z'_sj) It it is right image jth base line projection centre coordinate.Formula (4), the merging of (5) simultaneous can be drawn corresponding to picture point p on left image (x_l, y_l) core line equation

$x_r=-f\frac{r_{11}^{\&prime;}A+r_{21}^{\&prime;}B+r_{31}^{\&prime;}C}{r_{13}^{\&prime;}A+r_{23}^{\&prime;}B+r_{33}^{\&prime;}C}$

(6)

$y_r=-f\frac{r_{12}^{\&prime;}A+r_{22}^{\&prime;}B+r_{32}^{\&prime;}C}{r_{13}^{\&prime;}A+r_{23}^{\&prime;}B+r_{33}^{\&prime;}C}$

Wherein

A=X_si-X'_sj+λ(r₁₁·x_l+r₁₂·y_l-r₁₃f)

B=Y_si-Y′_sj+λ(r₂₁·x_l+r₂₂·y_l-r₂₃f)

C=Z_si-Z'_sj+λ(r₃₁·x_l+r₃₂·y_l-r₂₃F) (7)

When ground point elevation on given projection ray qS (given Z value), proportionality factors lambda can be calculated, utilize public Formula (6) i.e. can determine that with (7) left image is as core curvilinear path lq corresponding to p point.Under practical situation, the x coordinate of linear array image is not One is set to zero, provides the strict difinition (x of projected footprint method core line equation herein_r, y_r).Due in the row of line array loudspeakers Heart projection property, spin matrix r_mnWith r'_mnChange with base line, the linear array image core line equation drawn based on projected footprint method Form is complicated, and is non-linear form, and formula (6) and (7) are only suitable for being analyzed core line, it is impossible to be directly used in core line and heavily adopt Sample.Based on forefathers' research to linear array image core line geometry and the core line geometry relation of projected footprint method, following knot can be drawn Opinion:

(1) there is not the core line on the strict difinition of similar frame width formula image in linear array image, and the core line of linear array image is the most not It is straight line, but similar hyperbola；

(2) in subrange linear array image core line close to straight line；

(3) there is not the corresponding epipolar line pair of strict difinition in linear array image, but there is core one to one in subrange Line, on this basis, can be converted to linear search by the coupling of linear array image by two-dimensional search.

2 martian surface linear array image core line resamplings

Below as a example by Mars Express HRSC image, introduce martian surface linear array image core line based on tight geometric model Method for resampling.

2.1 Mars Express HRSC rigorous geometric model

Mars Express is the first mars exploration task of European Space Agency, launches in June, 2003, and in December, 2003 flies to fire Star.The HRSC camera carried on Mars Express is exclusively used in Mars mapping, and the high resolution image of acquisition covers fire the most substantially The star whole world.When orbit altitude is 250Km, HRSC panchromatic wave-band image resolution can obtain 9 up to 10m, HRSC camera simultaneously Wave band linear array image, i.e. 5 panchromatic wave-band and 4 multi light spectrum hands (red, green, blue and near-infrared).HRSC camera geometric parameter It is shown in Table 1.

Table 1HRSC camera geometric parameter

Camera parameter	Index
		Wave band number	9
Panchromatic wave-band	0°、±18.9°、±12.8°
		Multi light spectrum hands	Near-infrared, red, green, blue
Camera focus/mm	175
		Pixel size/um	7
Instantaneous field of view angle/arcsec	8.2
		Linear array pixel count	5184
Quantizing bit number/bit	12
		Orbit altitude/km	250
Image resolution/m	10
		Base-height ratio	0.68/0.45

As it is shown on figure 3,9 line array CCD arranged parallel of HRSC camera are on focal plane, every line array CCD has 5184 pictures Element, coordinate system Y-axis is identical with detector heading, and X-axis is along line array CCD orientation and is perpendicular to Y-axis, and Z axis is perpendicular to Jiao Plane and sensing photography direction, XYZ tri-axle constitutes right-handed system.

The initial elements of exterior orientation obtaining HRSC relates to time and coordinate system conversion, the most first determines the space-time datum of Mars. Convert the time into the J2000 ephemeris time, the unified conversion of photography measurement object space coordinate system to Mars body-fixed coordinate system (MarsBody- FixedCoordinateSystem).Mars reference ellipsoid is positive spheroid, and radius is 3396.0km, and the first meridian is defined on Airy-0 crater central area, longitude and latitude definition mode is identical with the earth.Time provides with coordinate system conversion NAIF SPICE storehouse is carried out, as shown in Figure 4.SPICE storehouse provides the position of satellite, attitude, satellite clock information, instrument parameter and row The data such as star coordinate system definition.

Mars Express HRSC sensor rigorous geometric model formula based on collinearity condition equation is as follows

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{c}{X}_S\\ Y_S\\ Z_S\end{array}\right]+{\mathrm{\&lambda;R}}_{J2000}^{\mathrm{MARS}}{R}_{\mathrm{star}}^{J2000}{R}_{\mathrm{body}}^{\mathrm{star}}[\left[\begin{array}{c}{d}_x\\ d_y\\ d_z\end{array}\right]+R_{\mathrm{camera}}^{\mathrm{body}}\left[\begin{array}{c}x\\ y\\ -f\end{array}\right]]---\left(6\right)$

Wherein, (x, y) is the picpointed coordinate on HRSC image, [X, Y, Z]^TFor Mars topocentric coordinates, [X_S, Y_S, Z_S]^TFor Mars probes position after precise orbit determination,It is tied to Mars body-fixed coordinate system spin matrix for J2000 coordinate,For The detector attitude that star sensor measures,For the spin matrix of satellite body to star sensor,For camera and satellite Installation matrix between body, [d_x, d_y, d_z]^TFor image center eccentricity component under satellite body.

2.2 linear array image differential are corrected and are generated relative level image

The core line resampling of frame width formula image can be by realizing the method for initial dip image rectification to horizontal image.Line Battle array image can not provide clear and definite core line geometry definition by frame width formula image like that, and core line based on the analysis of projected footprint method is several Although what formula is tight in theory, but is not easy to practical operation.For linear array image core line resampling problem, frame can be used for reference Width formula image differential corrects the method generating horizontal image, carries out core line resampling on horizontal image.To this end, herein based on throwing Shadow method of loci first analyzes corresponding image points core line geometry relation on raw video, and then, given Mean height plane is as projection Face, carries out differential and corrects the horizontal image of acquisition linear array image, owing to horizontal image and raw video can set up strict turning Change relation, therefore the core line in original linear array image can be changed to horizontal image, analyze several on horizontal image of core line What relation also determines core line direction, then carries out core line resampling on horizontal image.

As it is shown in figure 5, corresponding image points p, q on raw video correspond to corresponding image points p ', the q ' on horizontal image, based on Core curve lp Yu lq that projected footprint method obtains forms Lp ' and Lq ', owing to eliminating the shadow of heeling error on horizontal image Ringing, Lp ' and Lq ' is approximately straight line.Experimental result from behind is it can be seen that core line direction is close to detector flight side To, and detector heading and perspective plane X-direction have certain angle, can be by horizontal image along core after calculating this angle Line direction carries out resampling.It addition, carry out core line resampling on the basis of differential corrects the horizontal image of generation can also enter one Step eliminates the impact of geometric distortion, improves matching precision and reliability.

2.3 horizontal image coker line geometry relation analyses

The coordinate that tight geometric model based on collinearity condition equation can be set up between horizontal image and raw video turns Changing relation, recycling ground point backprojection algorithm can set up the core line geometry relation of corresponding image points.For ease of describing, the most not List specific formula for calculation again, and use Function Mapping mode to represent transformation process

Wherein formula (7) represents left image picture point and the mutual conversion of picture point on horizontal image, and positive transition represents by original Image picpointed coordinate (x_l, y_l) picpointed coordinate (x on calculated level image_ln, y_ln), inverse transformation represents by the picture on horizontal image Point coordinates (x_ln, y_ln) calculate raw video picpointed coordinate (x_l, y_l).Formula (8) represents on right image picture point image horizontal with it The mutual conversion of picture point.Formula (9), (10) represent that utilizing ground point back projection (projected footprint method) to set up left and right image picture point sits Mark and the transformational relation of horizontal image picpointed coordinate, as a example by left image, can be by the picpointed coordinate (x on left image_l, y_l) calculate Go out the picpointed coordinate (x on horizontal image_ln, y_ln), draw topocentric coordinates (X, Y, Z), carry out ground point back projection and obtain right shadow As upper picpointed coordinate (x_r, y_r).Above-mentioned formula is to carry out same place measurement and stereotactic basis on horizontal image, I.e. on left and right horizontal image, measure corresponding image points (x_ln, y_ln) and (x_rn, y_rn), then can be converted on original image Picpointed coordinate, and then utilize collinearity condition equation to carry out forward intersection calculating.

Utilize some conversion on the core curve that projected footprint method can obtain by formula (7)～(10) to horizontal image, Having corrected, owing to correcting through differential, the projection error that image inclination brings, the core curve Lp ' on horizontal image is approximately with Lq ' Straight line, by calculating the slope of Lp ' and Lq ', it can be deduced that core line direction (see figure 6), heavily adopts along core line direction Sample, can obtain approximate epipolar image.

2.4 correct the core line method for resampling of the horizontal image of generation based on differential

Based on the core line geometry relation analysis on original image and horizontal image of the above-mentioned projected footprint method, herein for Mars Express HRSC linear array image proposes a kind of core line method for resampling correcting the horizontal image of generation based on differential, and algorithm has Body flow process is as it is shown in fig. 7, Mars Express HRSC image core line resampling steps is as follows:

(1) SPICE storehouse is utilized to obtain the data such as Mars Express detector position, attitude and be converted to elements of exterior orientation, base Mars Express HRSC image rigorous geometric model is built in collinearity condition equation；

(2) utilize projected footprint method to calculate picpointed coordinate core line on raw video, obtain on core line a series of from Scatterplot；

(3) setting Mean height plane is as perspective plane, utilizes rigorous geometric model that linear array stereoscopic image is carried out differential and entangles Just obtaining horizontal image；

(4) discrete point on raw video core line is changed to horizontal image, and calculate core line on horizontal image Geometrical relationship also determines core line direction；

(5) on horizontal image, carry out core line resampling along core line direction, obtain approximate epipolar image.

3 ground point back projection fast methods

Projected footprint method core line analysis and horizontal image rectification all refer to ground point back projection, i.e. give topocentric coordinates and search The optimum scanning row that this ground point of rope is corresponding, and calculate the picpointed coordinate of correspondence.Traditional linear array image ground point back projection Algorithm relates to the successive ignition computing of collinearity equation, inefficient.Make full use of linear array image each base line perspective plane herein Geometrical-restriction relation quickly positions optimum scanning row, and ultimate principle is as follows: as shown in Figure 8, the every one scan line of HRSC linear array image May make up a perspective plane on object space with projection centre, ab is that i-th scan line array CCD first, last visits unit, S_iFor The projection centre of this base line, image space ab and projection centre S_iAnd ground point AB may make up the perspective plane of i-th base line.With Sample can calculate image space cd and projection centre S_i+1And the i+1 bar base line perspective plane that ground point CD is constituted.Ground point P is positioned at Article i-th, between base line perspective plane and i+1 bar base line perspective plane, utilize the geometry between each base line perspective plane, thing side to close System, by judging the distance of ground point P and each base line perspective plane, can quickly position the base line i closest with ground point P and I+1, recycling ground point P go out optimum scanning row with closest scanning distance accurate interpolation in the ranks.

Ground point backprojection algorithm based on perspective plane, thing side geometrical constraint is also required to iteration and carries out, but iterative computation is Simple interspace analytic geometry computing, compares collinearity condition equation computation amount, and can use each during iterative computation Distance between base line perspective plane carries out Fast estimation, compares tradition image space collinearity equation iterative computation repeatedly, and this method is only Needing interspace analytic geometry computing the most several times, therefore optimum scanning line search efficiency is improved significantly.

The step of linear array image optimum scanning line search method based on perspective plane, thing side geometrical constraint is as follows

Step 1: obtain each base line time needed for linear array image tight geometry location model, elements of exterior orientation, camera Focal length, linear array CCD visit the parameters such as unit position.

Step 2: calculating the perspective plane equation of each base line of linear array image, computing formula is

$\left|\begin{array}{ccc}x-x_1& y-y_1& z-z_1\\ x_2-x_1& y_2-y_1& z_2-z_1\\ x_3-x_1& y_3-y_1& z_3-z_1\end{array}\right|=0---\left(12\right)$

Above formula is converted to the general type of space plane

Ax+By+Cz+D=0 (13)

Wherein A, B, C are the direction number of normal n of space plane, and are not simultaneously equal to zero.1, space M (x₀,y₀,z₀) Distance to space plane is

$d=\frac{|{\mathrm{Ax}}_0+{\mathrm{By}}_0+{\mathrm{Cz}}_0+D|}{\sqrt{A^2+B^2+C^2}}---\left(14\right)$

This range formula is main amount of calculation during optimum scanning row fast locating algorithm iteration.

Step 3: according between perspective plane Equation for Calculating each base line perspective plane of camera projection centre and each base line Distance D_i, D₁Represent the distance between Article 1 base line perspective plane and Article 2 base line perspective plane, by that analogy.D_iFor right The closest base line of ground point P carries out Fast estimation.

Step 4: first judge distance d of ground point P and Article 1 base line perspective plane_i(i=1), if d_iMore than two scannings Distance D between row perspective plane_i(i=1), then utilize following formula that base line increment n is estimated, n=d_i/n_i。

Step 5: iteration judges P and the distance on i-th base line perspective plane, until d_i<D_iAnd go to the 6th step.

Step 6: utilize the base line accurate interpolation closest with P to go out optimum scanning row

$L=i+\frac{D_i}{D_i+D_{i+1}}---\left(15\right)$

4 experiments and analysis

Choosing HRSC image near your crater of curiosity touch-down zone lid to test, the image capturing time is 2008 2 The moon 9, image resolution about 15 meters, choose two channel image of S1 with S2 that in HRSC9 passage, base-height ratio is maximum and try Test.

Ground point back projection is the base that projected footprint method core line geometry analysis herein and differential correct the horizontal image of generation Plinth, the precision of the precision of ground point backprojection algorithm and the direct image of efficiency HRSC image core line method for resampling herein and effect Rate.Test randomly chooses 1,000,000 points on S1 Yu S2 passage HRSC, and given random elevation obtains topocentric coordinates, utilizes Backprojection algorithm is calculated corresponding picpointed coordinate by topocentric coordinates and contrasts with picpointed coordinate, thus analyzes ground Point backprojection algorithm precision and efficiency.Table 2 is to utilize tradition based on image space collinearity condition equation iterative algorithm and algorithm herein Comparing result.

Table 2 ground point backprojection algorithm precision and efficiency comparative

In analytical table, data understand, and cake backprojection algorithm is based on image space collinearity condition equation iteration with tradition textually Method precision is consistent, all can reach the high accuracy optimum scanning row location being better than 0.001 pixel, and algorithm calculates speed more herein Hurry up, practical level is higher.Utilize this ground point back projection fast method can significantly promote projected footprint method core line analysis and The computational efficiency of differential correction procedure.

Choose corresponding image points p (1836.25,5124.25) and q (1951.50,5430.50) on S1, S2 image, utilize and throw Shadow method of loci obtains p, q core curve on corresponding image, and when projected footprint method analyzes core line, elevation change step is taken as image / 3rd of resolution, i.e. 5 meters.The core line calculated is superimposed upon on original HRSC image, core curve entirety and local Display result is shown in Fig. 9 and Figure 10 respectively, and in Fig. 9, red solid line is core curve.For analyzing the geometric position of same place and core line Relation, in Figure 11, curve is picture point q (1951.50,5430.50) core curve in S1 image capturing range, and round dot is picture point p.

As shown in Figure 11, the core line that projected footprint method obtains on original HRSC image along detector heading, overall On remain close to straight line, but as seen from the figure, the core line on original HRSC image can not use straight line to approximate, i.e. The core line geometry relation that projected footprint method is analyzed can not be utilized, on original HRSC, directly carry out core line resampling.Figure 10 Central Plains The core line that beginning scale shows, close to straight line, shows can approximate in subrange at HRSC original image coker line to regard as Straight line.In Figure 11 on S2 image the corresponding image points p (1836.25,5124.25) of picture point q (1951.50,5430.50) generally within On core curve, show that the core curve drawn along projected footprint method can search for corresponding image points.

Calculate HRSC image core line direction with the core line image generating method of the present invention, carry out resampling along core line direction, S1 Yu S2 is constituted stereo-picture, and result is as shown in figure 12.On stereo-picture, measurement 20 is to corresponding image points, analyzes it and regards up and down Difference, the results are shown in Table 3 and Figure 13.

Table 3HRSC approximate epipolar image corresponding image points vertical parallax

Period	xl	yl	xr	yr	dy
						1	235.42	1337.45	247.38	1337.49	0.04
2	1082.75	1408.77	1093.42	1408.32	-0.45
						3	1472.39	1492.32	1402.55	1492.11	-0.21
4	2032.78	2308.49	2017.59	2308.9	0.41
						5	2389.45	2692.88	2389.45	2693.03	0.15
6	2548.32	3208.45	2526.19	3208.66	0.21
						7	3039.42	3348.82	3012.76	3349.31	0.49
8	3108.46	3921.32	3108.46	3921.04	-0.28
						9	3286.39	4408.62	3243.52	4408.32	-0.3
10	3498.8	4921.77	3460.28	4921.38	-0.39
						11	3591.76	5820.57	3543.36	5820.39	-0.18
12	3683.92	6020.32	3639.78	6020.49	0.17
						13	3781.64	6508.41	3757.34	6508.76	0.35
14	3904.47	7782.57	3894.57	7782.06	-0.51
						15	3988.64	8090.31	3908.35	8090.34	0.03
16	4011.56	8842.47	3922.47	8842.28	-0.19
						17	4321.66	8917.59	4201.32	8917.33	-0.26
18	4420.68	9028.32	4410.52	9027.97	-0.35
						19	4790.92	9421.49	4756.8	9421.68	0.19
20	4890.32	9632.83	4863.45	9632.68	-0.15

Analyzing the stereoscopic image being made up of core line image and same place vertical parallax understands, the method for the present invention generates Core line view stereoscopic measures effective, and same place vertical parallax is respectively less than 1 pixel, demonstrate the feasibility of the inventive method with High efficiency.Can carry out the dense Stereo Matching of HRSC image on the basis of this method core line image, promote the precision of dense Stereo Matching with Reliability.

Claims (4)

1. martian surface linear array image core line method for resampling, it is characterised in that the step of the method is as follows:

(1) obtain martian surface line-scan digital camera camera site, attitude relevant parameter, build image based on collinearity condition equation tight Geometric model；The tight geometric model of image is

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=\left[\begin{array}{c}{X}_S\\ Y_S\\ Z_S\end{array}\right]+{\mathrm{\&lambda;R}}_{J2000}^{MARS}{R}_{star}^{J2000}{R}_{body}^{star}\&lsqb;\left[\begin{array}{c}{d}_x\\ d_y\\ d_z\end{array}\right]+R_{camera}^{body}\left[\begin{array}{c}x\\ y\\ -f\end{array}\right]\&rsqb;---\left(1\right)$

Wherein, (x, y) is the picpointed coordinate in linear array image, [X, Y, Z]^TFor Mars topocentric coordinates, [X_S, Y_S, Z_S]^TFor precision Mars probes position after orbit determination,It is tied to Mars body-fixed coordinate system spin matrix for J2000 coordinate,For star The detector attitude that sensor measures,For the spin matrix of satellite body to star sensor,For camera and satellite Installation matrix between body, [dx, dy, dz]^TFor image center eccentricity component under satellite body, λ is scale factor, and f is The master of camera away from；

(2) based on tight geometric model, projected footprint method is utilized to calculate the picpointed coordinate on left image on original right image The core line on left image that picture point on core line, right image is corresponding；

(3) setting Mean height plane is as perspective plane, utilizes tight geometric model that linear array stereoscopic image carries out differential correction and obtains Take horizontal image；

(4) discrete point on raw video core line is changed to horizontal image, calculate core line geometry on horizontal image and close It is and determines core line direction；

(5) on horizontal image, carry out core line resampling along core line direction, obtain approximate epipolar image.

Martian surface linear array image core line method for resampling the most according to claim 1, it is characterised in that described step (1) in, camera camera site, attitude relevant parameter include camera focus, Pixel size, pixel calibration coordinate, camera position, appearance State data, Mars coordinate system uses the definition of positive spheroid, and radius is 3396Km.

3. according to the martian surface linear array image core line method for resampling according to any one of claim 1-2, it is characterised in that The coordinate that in described step (3), tight geometric model based on collinearity condition equation is set up between horizontal image and stereoscopic image turns Changing relation, recycling ground point Inverse Projection sets up the core line geometry relation of corresponding image points.

Martian surface linear array image core line method for resampling the most according to claim 3, it is characterised in that described ground point The optimum scanning line search process that in Inverse Projection, ground point is corresponding is as follows: (11) obtain linear array image tight geometry location model Required each base line time, elements of exterior orientation, camera focus, linear array CCD visit unit's location parameter；(12) linear array image is calculated The perspective plane equation of each base line；(13) according to each base line of perspective plane Equation for Calculating of camera projection centre and each base line Distance D between perspective plane_i, D_iFor the closest base line of ground point P is carried out Fast estimation；(14) first judge ground point with Article 1, distance d on base line perspective plane_i, i=1, if d_iMore than distance D between two base line perspective planes_i, i=1, then utilize N=d_i/n_iBase line increment n is estimated by formula；(15) iteration judges ground point and the distance on i-th base line perspective plane, directly To d_i<D_iAnd go to the 16th step；(16) the base line accurate interpolation closest with P is utilized to go out optimum scanning row $L=i+\frac{D_i}{D_i+D_{i+1}}.$