WO2011151371A1 - Method of autonomous calibration of the directions of aim of the sensors of a detection strip using orthogonal snap shots - Google Patents

Abstract

According to a first aspect, the invention relates to a method of determining the directions of aim of the sensors of a detection strip (CCD) of a picture taking system carried on board an observation satellite, implementing a correlation of a reference image of a scene with a secondary image of the scene obtained by the detector strip, characterized in that the reference image (R) and the secondary image (S) consist of orthogonal snap shots of the scene captured by the detector strip.

Description

 AUTONOMOUS CALIBRATION METHOD OF THE SENSING DIRECTIONS OF THE SENSORS OF A DETECTION BAR, USING SOCKETS

 ORTHOGONALES DOMAIN OF THE INVENTION

 The field of the invention is that of observation satellites, and more particularly that of the calibration of embedded shooting systems within such satellites.

 Even more specifically, the invention relates to the calibration in flight of the sighting directions of the different sensors of a sensor array of a camera system. It will be noted at this stage that reference is also made to mapping the focal plane to designate the estimation of the aiming directions of the different sensors.

 BACKGROUND OF THE INVENTION

 A flight recipe of an observation satellite starts from the control of the good functioning of the satellite and lasts a few months. This recipe is implemented to verify the adequacy of the system to the specifications and to allow the diffusion and the marketing of images with the best possible geometry.

 The estimation of the geometrical performances is also followed throughout the life of the satellite, in order to be able to identify possible evolutions or anomalies.

 Geometric calibration is the first objective of the geometric flight recipe. It aims to provide ground treatments some geometric parameters involved in modeling and geometric treatments. These parameters were previously measured on the ground but some of them were able to evolve at the end of the launch; others could only be measured with insufficient precision.

 These parameters include the aiming directions of the sensors of a detection strip of the camera system. Thus, it is sought during a geometrical calibration in flight, to re-estimate the directions of sight of the different sensors, in order to establish a map of the focal plane.

 A known method for doing so is based on the correlation between an image acquired by the bar to be mapped and an absolute reference image, of better resolution, called geometric supersite, such as an aerial photograph.

 A description of this method can be found on pages 141 -145 of the book "Spatial Imaging, From Principles of Acquisition to the Processing of Optical Images for Earth Observation" published by Cépaduès Editions (ISBN: 978.2.85428.844.5 ).

Thus, geometric calibration is now performed using mainly geometrically perfectly known sites (planimetrically and altimetrically). These sites are of different shapes (images and geometric models, support points, 3D models ...), and different precisions. These sites are composed of two main families, one for calibrating the alignment biases and measuring the localization performance (database of points of support), the other allowing to make the fine geometry at through dynamic and static analyzes allowing a fine analysis of the attitude residues and a cartography of the different focal planes.

 The use of supersites, however, is not satisfactory in that it is accompanied by constraints related to the creation, operation and maintenance of supersites. In addition, mapping of the focal plane can only be established on specific locations.

 In general, we are looking for a technique to geometrically calibrate an earth observation system that would overcome these constraints. It is also generally desired to achieve finer maps than those obtained via supersites.

 STATEMENT OF THE INVENTION

 The aim of the invention is to meet these needs, and proposes for this purpose according to a first aspect, a method for determining the aiming directions of the sensors of a detection strip of an on-board camera system within an observation satellite, implementing a correlation of a reference image of a scene with a secondary image of the scene obtained by the detector array, characterized in that the reference image and the secondary image consist of orthogonal shots of the scene obtained by the array of detectors.

 Some preferred, but not limiting, aspects of this method are as follows:

 it furthermore comprises the averaging of a row and a column of the matrix obtained by the correlation of the reference image and the secondary image in order to estimate the offset of the aiming direction of a sensor in parallel and perpendicular to the satellite track;

 the offsets of the aiming directions of the sensors of the array are modeled as polynomials;

 - modeling in the form of polynomials implements a least squares adjustment;

it furthermore comprises the implementation of a two-dimensional least-squares adjustment of the matrix obtained by correlating the reference image and the secondary image in order to model in the form of polynomials the offsets of the sighting directions of the sensor of the array parallel and perpendicular to the trace of the satellite; the estimation of the offsets of the sighting directions of the sensors being modulated by dynamic residual effects, polynomial modeling is subtracted from the estimate to estimate the attitude residues of each of the shots of the reference image and of the secondary image;

 it comprises, prior to the correlation of the reference image and the secondary image, a step of collocating said images for projecting the secondary image into the reference image so as to make them quasi-superimposable.

 According to a second aspect, the invention relates to a computer program product comprising a set of instructions that when executed by a computer processing unit cause said unit to execute the steps of the method according to the first aspect of the invention. .

 According to a third aspect, the invention relates to a shooting system intended to be embedded in an observation satellite, comprising a detection strip formed of a plurality of detectors and a computer program according to the second aspect of the invention. which when implemented allows a calibration of the camera system.

 BRIEF DESCRIPTION OF THE DRAWINGS

 Other aspects, objects and advantages of the present invention will appear better on reading the following detailed description of preferred embodiments thereof, given by way of non-limiting example, and with reference to the appended drawings in which: :

 FIG. 1 illustrates the orthogonal shots of a reference image and a secondary image of the same scene in accordance with the invention;

 Figure 2 is a diagram illustrating the various steps of the method according to the first aspect of the invention;

 FIGS. 3a and 3b represent an estimation of the shifts of the sighting directions respectively parallel and perpendicular to the trace of the satellite obtained by a polynomial modeling during a simulation of static effects without residual dynamic effects;

FIGS. 4a and 4b represent an estimation of the shifts of the sighting directions parallel and perpendicular to the trace of the satellite obtained by polynomial modeling during a simulation of static effects with very high frequency residual dynamic effects; FIGS. 5a-5d represent an estimate of the dynamic residual effects parallel and perpendicular to the satellite track during a simulation of static effects with very high frequency residual dynamic effects.

 DETAILED DESCRIPTION OF THE INVENTION

 In a general manner, according to a first aspect, the invention proposes a method for determining the viewing directions of the sensors of a detection strip of an on-board camera system within an observation satellite which enables to overcome supersites by being based on two orthogonal acquisitions (not necessarily single-pass) of the same scene. The main advantage of this type of acquisition is to "decorrelate" the static effects of shooting dynamic effects.

 FIG. 1 shows the orthogonal images of a reference image R and a secondary image S of the same scene in accordance with the invention.

 In this FIG. 1, the reference CCD illustrates the position of the detection strip of the observation satellite's observation system. This bar, typically of the CCD type ("Charge-Coupled Device" designating a charge transfer device), comprises a plurality of photosensitive sensors arranged in line, for example of the order of 4000 to 6000 sensors per line. The reference t illustrates the movement of the satellite over time. FIG. 1 also shows an overall view of the two mergings made by the satellite for obtaining orthogonal images R, S of the same scene.

 Consider two products from "orthogonal" shots. These two products are orthogonal: if the product lines are acquired along the CCD and the product columns are acquired by time, then the product lines2 are acquired by the time and the columns of the product2 are acquired along the CCD. The use of this pair of images can be traced back to the static and dynamic characteristics of each shot.

 Figure 2 is a diagram illustrating the different steps of the method according to the first aspect of the invention.

According to a first step called "setting geometry" is chosen in the pair of orthogonal images a reference product and a secondary product. We then use the geometric models of the reference image and the secondary image, in order to inject into the images all the knowledge delivered by the system (attitudes, cartography of the existing focal plane ...). The two images are then almost superimposable. The only differences come from lack of awareness of the focal plane and residual attitude errors (not restored by the SCAO Orbit and Attitude Control System). This step of "setting geometry" thus consists in projecting the first image of the couple in the second. More specifically, a "collocation" transformation is implemented to map all the row and column positions of the first image to the row and column positions of the second image. The direct location model of the first image (using all the information contained in the auxiliary data of the product) is used to obtain information on the position of this image on the ground. This ground information is then used and the inverse localization model of the second image.

 According to a second step called "dense correlation", the secondary image and the reference image are correlated. The lines of the resulting disparity web comprise information on the mapping of the focal plane and vib2 information relating to the attitude residues of the secondary shot, while the columns of the disparity web correspond to information relating to the mapping of the focal plane and vib1 information relating to the attitude residues of the reference shot.

 Thus, in the context of the invention, the two correlated images both correspond to images taken by the satellite. In such a way, it is not necessary to resort to aerial photography of a supersite as was the case in the prior art.

 A statistical filtering of the disparity layer resulting from the correlation of the two orthogonal ones can be implemented in order to eliminate the aberrant points. As a non-limitative example, we can calculate, for each line (respectively column) of the disparity table, the mean and the standard deviation of the disparities (offsets), and consider only the points whose disparity is included between the mean minus three times the standard deviation and the average plus three times the standard deviation.

 According to a third stage called "statistics", it comes to estimate the mapping of the focal plane and possibly the attitude of each shooting.

 According to a first embodiment, an average line and an average column are calculated on the disparity layer resulting from the "dense correlation" step (and preferably after statistical filtering as presented above). We then make the assumptions according to which the mapping of the focal plane is a signal that can be modeled, by a polynomial (for example of degree less than or equal to 5) or by another function, and the residues of attitudes correspond to high frequency effects.

In this first embodiment, the method thus comprises the averaging of a row and a column of the matrix obtained by the correlation of the reference image and of the secondary image to estimate the offset of the direction of sight of a sensor parallel and perpendicular to the trace of the satellite.

 The offsets of the aiming directions of the sensors of the array can then be modeled as polynomials, for example by implementing a least-squares adjustment.

 According to a second embodiment, a two-dimensional least-squares adjustment of the matrix obtained by correlating the reference image and the secondary image to model as polynomials or other functions the shifts of the sighting directions of the sensor of the bar parallel and perpendicular to the track of the satellite. It should be noted that the solution according to the first embodiment based on the averaging of rows and columns has the advantage of simplicity.

 When the attitude residues are zero, the cartography of the focal plane is direct, without having to formulate frequency hypotheses.

 On the other hand, in the presence of attitude residuals, the estimation of the mapping of the focal plane is modulated by the dynamic residual effects. We can then correctly estimate the mapping of the focal plane by assuming a polynomial modeling for mapping the focal plane and a high frequency signal for the attitude residues. It is also possible to finely restore the attitude of each shot from the modulated signal simply by subtracting the polynomial modeling of the map of the focal plane.

 The following discussion details various simulations performed by the Applicant.

A geometric simulation chain simulates on the one hand the effects of focal planes (simulation of qx and qy polynomial models, with possible inter-array zones), on the other hand the residual dynamic effects (simulations of sinusoid combinations). parameterized by their periods, their amplitudes as well as their phasing).

 The method according to the first aspect of the invention has in particular been tested on different cases of simulations:

 - Static effects without residual dynamic effects;

 - Static effects with very high frequency residual dynamic effects;

 - Static effects with residual average dynamic effects and high frequencies; Two test images are used for the simulations, an image acquired by the satellite

QUICKBIRD and an image from the PELICAN airborne sensor. All results are similar with one or the other of the images. In what follows, the term "shooting" is used to denote shooting acquired in standard mode (composed of an image 0 and an attitude 0). Its orthogonal in the "orthogonal" shooting mode is designated by shot 90 (composed of an image 90 and an attitude 90).

 In the context of these simulations, the simulated static effects are polynomials of degree less than or equal to 5 for the directions of view in x (parallel to the trace) and in y (perpendicular to the trace), of even degree for the one and of odd degree for the other.

 According to a first simulation case, we consider static effects without residual dynamic effects.

 The mean line resulting from the disparity sheet gives information on the cartography of the focal plane and the attitude residues of the image 90. The average column resulting from the disparity layer gives information on the cartography of the plane. focal and the attitude residues of the image 0. These residues of attitudes in this case must be null.

 In all cases the average line or column is modeled by a degree 5 (least squares) polynomial. It is interesting to note that the correlation errors due to the landscape will not appear simultaneously on the row and the middle column.

 FIGS. 3a and 3b represent the estimation of the shifts of the sighting directions respectively parallel and perpendicular to the trace of the satellite obtained by the polynomial modeling resulting from the line and average column in this first simulation case (via an average of the two modelizations) .

 It will be noted here that when calculating shifts, two shifts of shifts are obtained: a grid of offsets along the lines and an offset grid along the columns. Thus, when we speak of average line (respectively average column), it is the average line (or average column) of the offsets along the lines (for information in x) and the mean line ( or average column) offsets along columns (for information in y).

 In these figures 3a and 3b, there is shown:

 in full line, the simulated static effect modeling the offsets of the sighting directions; in a "noisy" line, the measurement carried out in the context of the invention (for example via the calculation of a mean line or column on the disparity sheet resulting from the correlation of the orthogonal images);

 in dashed line, the estimation of the shifts obtained by polynomial modeling.

The performances of the conventional methods are of the order of 0.1 pixels; it is found that with the method of the invention the residual error is about 5 times smaller. According to a second case of simulation, one considers static effects with residual dynamic effects very high frequencies.

 In this second case, only high frequency disturbances are present. An additional arbitrary hypothesis is made: the amplitudes of the attitude residues of the shot 90 are larger (approximately 2 times more) than those of the shooting 0, the characteristic frequencies are also different. The simulations are representative of a local consistency specification over a horizon of one hundred lines.

 The mapping of the focal plane is modulated by the dynamic residual effects. The polynomial modeling hypothesis for the mapping of the focal plane and of the high frequency signal for the attitude residuals makes it possible to correctly estimate the mapping of the focal plane. In addition, it is possible to finely restore the attitude of each of these shots from the estimated modulated signal by simply removing the map of the focal plane (as estimated by the polynomial modeling).

 FIGS. 4a and 4b show the estimation of the shifts of the sighting directions parallel and perpendicular to the trace of the satellite obtained by the polynomial modeling resulting from the line and average column in this second modeling case.

 FIGS. 5a-5d also show the estimation of dynamic residual effects parallel and perpendicular to the satellite trace in this second modeling case. More precisely, FIGS. 5a and 5b show the dynamic residual effects restored on the shooting 90 respectively in roll and pitch from the mean line. Figures 5c and 5d represent the residual dynamic effects returned to the shooting 0 respectively roll and pitch from the middle column. In these figures 5a-5d, there is shown in full line simulated dynamic effect and dashed line dynamic effect estimated according to a possible embodiment of the method according to the first aspect of the invention.

 There is also overall performance 5 times better than those achieved with conventional methods.

 In particular, despite residual dynamic disturbances, the estimation errors of the mapping of the focal plane remain less than 2 hundredths of RMS pixels.

 It is also possible to restore residual attitudes. We note in particular that the frequencies are correctly restored while the amplitudes appear slightly underestimated.

According to a third simulation case, static effects with residual average and high frequency dynamic effects are considered. In this third case, high and medium frequency disturbances are present. This case is representative of a local consistency specification over a hundred lines horizon and a typical local consistency over a thousand line horizon.

 The performances are always 5 times better than those achieved with classical methods.

 In particular, despite residual average frequency dynamic disturbances, the estimation errors of the mapping of the focal plane remain less than 3 hundredths of an RMS pixel and 9 hundredths of a pixel at the most.

 In general, it will be noted that once the cartography of the focal plane estimated, it is possible to restore more finely the attitude. In particular, it is possible to implement the method according to the invention iteratively to improve on the one hand the estimation of the mapping of the focal plane and on the other hand the estimation of the dynamic disturbances of the satellite.

 It will be understood that the main advantage of the method according to the first aspect of the invention is that it allows the realization of an autonomous calibration that does not require aerial images of supersites. The process thus makes it possible to override the constraints of creation, operation and maintenance of supersites.

 This method also makes it possible to dispense with a specific location (related to supersites). It is then possible to map the focal planes on sites at different latitudes, which makes it possible to follow a possible thermoelastic evolution of the mapping of the focal plane, and to use in routine follow-up of the low orbits.

 This method also makes it possible to achieve mapping accuracies finer than those obtained using supersites. It has been demonstrated by simulation that by using a single shooting torque, it is possible to achieve, whatever the resolution (Pelican or Quickbird), accuracies well below one tenth of a pixel. The simulated images are small (2000x2000), and large images should logically improve the statistic.

 It will be noted that it is also possible to use a large number of these orthogonal pairs, which will allow in flight recipe to refine the mapping (increased accuracy in n, with n the number of couples used).

 Finally, it will be noted that it is also possible to restore the attitude residues.

It will be understood that the invention is not limited to the method according to its first aspect, but also extends according to a second aspect to a computer program product comprising a set of instructions which when executed by a unit of processing means cause said unit to execute the steps of the method according to the first aspect of the invention. The invention also extends to a shooting system intended to be embedded in an observation satellite, which system comprises a detection strip formed of a plurality of detectors and a computer program product according to the second aspect of the invention which when implemented allows a calibration of the camera system.

Claims

1. A method for determining the aiming directions of the sensors of a detection strip of an on-board recording system within an observation satellite, implementing a correlation of a reference image of a scene with a secondary image of the scene obtained by the array of detectors, characterized in that the reference image and the secondary image consist of orthogonal shots of the scene obtained by the array of detectors. The method of claim 1, further comprising averaging a row and a column of the matrix obtained by correlating the reference image and the secondary image to estimate the shift of the aiming direction. a sensor parallel and perpendicular to the track of the satellite. 3. The method of claim 2, wherein the offsets of the sighting directions of the sensors of the bar are modeled as polynomials. The method of claim 3, wherein the modeling in the form of polynomials implements a least squares adjustment. The method of claim 1, further comprising implementing a two-dimensional least squares adjustment of the matrix obtained by correlating the reference image and the secondary image to model under polynomial shape the offsets of the aiming directions of the sensor of the bar parallel and perpendicular to the track of the satellite. 6. Method according to one of claims 3 to 5, wherein the estimate of the offsets of the sighting directions of the sensors is modulated by dynamic residual effects and in which the polynomial modeling is deducted from said estimate to estimate the residues d attitudes of each of the shots of the reference image and the secondary image. 7. Method according to one of the preceding claims, comprising prior to the correlation of the reference image and the secondary image, a collocation step said images for projecting the secondary image into the reference image to make them quasi-superimposable. A computer program product comprising a set of instructions that when executed by a computer processing unit causes said unit to perform the steps of the method of any one of the preceding claims. 9. Shooting system intended to be embedded in an observation satellite, comprising a detection strip formed of a plurality of detectors and a computer program according to claim 8 which, when implemented, makes it possible to perform a calibration of the camera system.