Method for improving the geolocation of satellite images

Abstract

Method (100) for improving the geolocation of images acquired by scrolling an image sensor (1) along a scrolling path of the sensor (1), said sensor being configured to acquire at least one image segment (2, 3) during a time period, the at least one image segment (2, 3) comprising a sequence of images of the Earth's surface acquired from successive positions of the sensor along the scrolling path at successive times during the time period, the method comprising the following step: - determining (102) at least one new image segment (5) acquired during a new time period which respectively begin and end within the time period of the at least one image segment (2, 3), the at least one new image segment (5) comprising a new sequence of images acquired at successive times during the new time period,The new sequence of images covering a geographical area observed by the sensor on the Earth's surface during the new time period, said observed geographical area being defined by at least one clipping parameter. Figure 1,

Description

Title of the invention: Method for improving the geolocation of satellite images

[0001] The present invention relates to the field of geolocation of Earth observation satellite images, also called spatiotriangulation.

[0002] It is known to use ground support points, whose coordinates are known, identified on reference images on the one hand, and on the images to be geolocated on the other hand, to determine the parameters of a line-of-sight correction model for the images to be geolocated.

[0003] Applying this known geolocation method to images acquired by a device moving along a path, such as a pushbroom, aimed at the nadir, leads to successive corrections of the different temporal acquisition segments, on which the control points are measured. The quality and distribution of the control points along the segment determine the quality of the correction. However, it is not possible to measure control points on water surfaces (oceans, seas, lakes) and moving objects such as clouds. The distribution of measurements is therefore dependent on the land / water ratio, which is predictable, and on cloud cover, which is statistically predictable. The distribution within the segment is also determined by the segmentation of the processed image segments along the path of the acquisition device.

[0004] The invention therefore aims to provide a solution to all or part of these problems.

[0005] To this end, the present invention relates to a method for improving the geolocation of images acquired by scrolling an image sensor along a scrolling trajectory of the sensor, said sensor being configured to acquire at least one image segment during a period of time, the at least one image segment comprising a series of images of the surface of the earth acquired, from successive positions of the sensor along the scrolling trajectory, at successive times of the period of time, the method comprising the following step: - determining at least one new segment of images acquired during a new period of time which respectively begins and ends within the time period of the at least one segment of images, the at least one new segment of images comprising a new sequence of images acquired at successive instants of the new period of time, the new sequence of images covering a geographical area observed by the sensor on the surface of the earth during the new period of time, said observed geographical area being defined by at least one cutting parameter.

[0006] According to one embodiment, the invention comprises one or more of the following features, alone or in a technically acceptable combination.

[0007] According to one embodiment, the sensor is carried on a satellite, or on an aircraft.

[0008] According to one embodiment, the at least one image segment comprises at least a first image segment acquired during a first period of time and at least a second image segment acquired during a second period of time, the method further comprising the following steps, implemented prior to the step of determining at least one new image segment: - assembly of the first image segment and the second image segment, to form a merged segment of images acquired during a cumulative period of time corresponding to the union of the first period of time concatenated with the second period of time; - determination that, for the step of determining at least one new image segment, the at least one segment is the merged segment and that the time period is the cumulative time period.

[0009] According to one embodiment, the at least one first image segment and the at least one second image segment are generated as a function of the sensor passing into visibility of a receiving station.

[0010] According to one embodiment, the first time period ends at a time when the second time period begins.

[0011] According to these provisions, the choice of an appropriate cutting parameter will make it possible to find in the images of the new image segment support points well distributed over the geographical area, so that good quality spatio-triangulation of the images of this area will be possible.

[0012] It will also be possible to adapt the spatio-triangulation processing according to the geographical areas covered by the image segments, for example by adapting the complexity of an error model of a spatio-triangulation processing to a geographical area. Thus, in an area where statistically few support points are found, for example an island or a snowy area, a less complex error model will preferably be used, for example a polynomial error model of lower order.

[0013] According to one embodiment, at least one clipping parameter includes for example: a useful registration area, a distribution of the useful area, an extension of the geographical area covered by the new image segment along a direction of movement of the sensor.

[0014] According to these provisions, the distribution of support points, identified on the surface useful, will be optimized to avoid the propagation effects of recalibration errors on areas where support points are absent, fewer in number, or incorrect.

[0015] According to one embodiment, the useful registration area is an area of ​​emerged land, the area of ​​emerged land being included within the geographical area, and being associated with a value of a visibility parameter, the value of the visibility parameter being greater than or less than a predetermined threshold.

[0016] According to one embodiment, the surface of emerged land is determined from the images of the new image segment, for example acquired at the nominal spatial resolution.

[0017] According to one embodiment, the value of the visibility parameter associated with the surface of emerged land is a function of an average cloud cover of the surface.

[0018] According to one embodiment, the average cloud cover of the surface is determined, for example for a set of days in a period of one year, from a history of surface images acquired during a period of one year, prior, of one or more years, to said period of the year.

[0019] According to one embodiment, the surface area of ​​emerged land is determined after excluding the surface area of ​​specific zones.

[0020] According to one embodiment, the specific areas include at least one of the following: an ice pack, a glacier, a snowy area, an untextured area at nominal spatial resolution, a water surface, a surface including moving objects.

[0021] According to one embodiment, the specific areas include an area with a temporal change indicator greater than a predetermined threshold.

[0022] According to one embodiment, the change indicator is a function of a difference between a radiometry at one time of at least one point of an image of the area and another radiometry of at least one point at another time, referred to a time elapsed between the time and the other time.

[0023] According to one embodiment, the step of determining the at least one new image segment includes a selection of the at least one new image segment from a set of potential new image segments, the at least one new segment being selected on the basis of a function calculated for each potential new image segment, such that the function is optimized by the at least one new segment selected.

[0024] According to one embodiment, the desired optimum of the function is a minimum, or a maximum.

[0025] According to one embodiment, the function is defined as a linear combination of a first term and a second term, the first term being a the first function of the useful area of ​​the new image segment and the distribution of the useful area over the new image segment, the second term being a second function of the extent of the geographical area covered by the new image segment along a direction of sensor scrolling.

[0026] According to one embodiment, the step of determining at least one new segment by optimizing the function includes implementing a nonlinear optimization method such as simulated annealing, graph-cutting, or dynamic programming.

[0027] According to one embodiment, the geographical extent of the geographical area covered by at least one new image segment is bounded by a minimum value and / or by a maximum value.

[0028] According to these provisions, an excessive geographical extent is avoided which would lead to complex spatio-triangulation registration processing, potentially incompatible with available computing resources; according to these provisions, a minimum geographical extent is ensured which makes it possible to rationalize the distribution of available resources for the processing of all new image segments; according to these provisions, by limiting the geographical extent covered by the new image segment, the registration processing is based on an error model whose complexity is controlled so as to allow a more robust inversion.

[0029] For the sake of clarity, an embodiment and / or implementation of the invention is described with reference to the accompanying drawings, which represent, by way of non-limiting example, an embodiment or implementation of a device and / or method according to the invention. The same reference numerals in the drawings designate similar elements or elements with similar functions.

[0030] [Fig.1] is an illustration of the process of creating at least one new image segment from a first image segment and a second image segment generated as a function of the satellite passing into the visibility of a ground receiving station.

[0031] [Fig.2] is a representation of a succession of covered geographical areas respectively by a series of image segments generated according to a logic other than that related to the optimization of the registration, for example according to a logic related to the passage of the satellite into visibility of the ground receiving stations.

[0032] [Fig.3] is a representation of another succession of geographical areas obtained from the succession of geographical areas of [Fig.2], after optimization of the segmentation according to an implementation method of the process according to the invention.

[0033] [Fig.4] is a schematic representation of the sequence of the process steps according to the invention.

[0034] The invention aims to improve the geolocation of images acquired by scrolling an image sensor 1 along a scrolling trajectory of the sensor 1. The images which are acquired during a period of time form an image segment 2, 3 comprising a series of images of the surface of the earth acquired, from successive positions of the sensor along the scrolling trajectory, at successive times of the period of time. According to this definition of an image segment, each image of the series of images forming a segment can in particular correspond to a line of points acquired by a linear array of elementary detectors, the array scanning the acquisition field in pushbroom mode.

[0035] For example, as illustrated in [Fig. 1], when the sensor is on board a satellite orbiting the earth, the image segment comprises a series of images acquired before the satellite comes into view of a receiving station 6 located on the ground, to which the satellite transmits one or more image segments 2, 3 acquired between two successive passes into view of a receiving station 6.

[0036] The logic for forming the image segments 2, 3 acquired by the sensor 1 is therefore generally determined by the sequence of passes of the sensor 1 over, or near, the ground receiving stations 6. This logic for forming the image segments 2, 3, linked primarily to the positioning of the ground receiving stations 6, leads to the formation of image segments 2, 3 covering geographical areas where it may be difficult to determine reference points for geolocation by spatio-triangulation of the images of these segments.For example, if the geographic area covered by the image segment thus formed mainly comprises specific areas such as sea ice, a glacier, a snowy area, an untextured area at nominal spatial resolution, a water surface, or a surface containing moving objects, it will be difficult to determine stable reference points on all or part of this image segment that can be correlated with corresponding reference points on reference images 7, the reference images 7 being themselves geolocated. The same will be true if the geographic area covered by the image segment mainly comprises specific areas characterized by rapid changes in radiometry in successive images of the area, i.e., by strong temporal variability in the radiometry of points in the area in successive images of the area.

[0037] Thus, as illustrated in [Fig. 4], to improve the geolocation of the acquired images, the method 100 comprises a step 102 of determining a new segment 5 of images acquired during a new period of time which respectively begins and ends within the time period of the segment of images 2, 3, the new segment 5 of images comprising a sequence of images acquired at successive instants of the new period of time, so that the new image sequence covers a geographic area defined by at least one of the following clipping parameters: a useful registration area, a distribution of the useful area, an extension of the geographic area along a direction of sensor movement.

[0038] According to a particular embodiment, the method 100 further comprises the following steps, carried out prior to the determination step 102 of the new image segment 5: - assembly 101 of a first segment of images 2 acquired during a first period of time and of a second segment of images 3 acquired during a second period of time, to form a merged segment 4 of images acquired during a cumulative period of time corresponding to the union of the first period of time concatenated with the second period of time; - determination Olbis that, for the determination step 102 of at least one new segment 5 of images, at least one segment 2, 3 is the merged segment 4 and that the time period is the cumulative time period.

[0039] According to these provisions, the first image segment 2 and the second image segment 3 are for example generated on the basis of the logic of the transition of the sensor 1 into the visibility of the receiving stations 6, and the determination step 102 of the process 100 is implemented on an aggregated segment which is not subject to the logic of the transition of the sensor 1 into the visibility of the receiving stations 6.

[0040] In particular, the image segments used for assembly 101 can be in continuation of one another, the first time period ending for example at a time when the second time period begins.

[0041] The determination 102 of the new sequence of images of the new segment of images 5 is carried out so that the new sequence of images covers a geographical area defined by at least one of the cutting parameters indicated above, namely: a useful registration surface, a distribution of the useful surface, an extension of the geographical area along a direction of movement of the sensor.

[0042] By convention, the first clipping parameter considered, i.e., the useful registration area, is defined as the surface of visible land, i.e., not obscured by clouds, at a nominal spatial resolution of the image sequence of the new segment 5. Excluded from the useful area are zones for which it is difficult (unlikely) to find reference points using state-of-the-art image matching techniques, in particular areas of significant change such as sea ice, glaciers, and snow-covered surfaces at high latitudes, and areas without texture at the image resolution, or areas where significant changes in radiometric content can occur in a short time. In particular, areas for which a change indicator is greater than a pre-defined threshold will be excluded. determined, the change indicator being for example a function of a difference between a radiometry at one instant of at least one point of an image of the area and another radiometry of the at least one point at another instant, related to a time elapsed between the instant and the other instant.

[0043] Thus, for example, if the new image segment 5 determined includes a useful registration area greater than or equal to a predetermined surface area threshold, expressed for example as a percentage of the total surface area of ​​the geographical zone covered by the new image segment, then the new image segment will allow the identification of a sufficient number of control points, of sufficient quality, to ensure satisfactory geolocation of the images of the new segment by spatio-triangulation. For example, the predetermined useful registration area threshold may advantageously take a value between 15% and 100%, preferably between 70% and 100%, and preferably also equal to 100%.

[0044] In particular, the useful registration area can be associated with a value of a visibility parameter, the value of the visibility parameter being determined on the basis of an average cloud cover of the geographical area covered by the image segment; more specifically, the value of the visibility parameter will be constrained so as to be above or below a predetermined threshold. The average cloud cover of the surface is determined, for example for a period of one year, from a history of surface images acquired during the same period of one or more years preceding the year of acquisition of the image segment in question, or from re-analysis data produced by a meteorological monitoring organization, for example METEO-FRANCE.

[0045] Furthermore, when the new image segment 5 is also defined by the second cutting parameter, namely a distribution of the useful registration surface, the new image segment 5 comprises a satisfactory distribution of the useful surface over the entire image segment; it will thus be possible to constrain the correction of the geometric models. Indeed, to compensate for the absence of support points on certain surfaces of interest, the geometric corrections are applied to the entire image segment, in order to extrapolate the registration data to all the images of the image segment. Thus, the distribution of the useful surface being assumed to be satisfactory, the distribution of the support points will also be, and this will avoid the effects of error propagation on areas of the image segment where the support points would be absent, fewer in number, or erroneous.

[0046] An indicator of the distribution of the usable surface area over the geographical area can, for example, be defined on the basis of an assessment of the maximum distance in the direction of travel between the useful elementary surfaces included in the new image segment 5, relative to the total length of the new image segment 5. The distribution of the usable area will be considered satisfactory, for example, if the indicator is between 50% and 100%, preferably between 75% and 100%.

[0047] According to these provisions, the choice of an appropriate value for the first and / or the second cutting parameter, i.e. the useful registration surface, and / or the distribution of the useful registration surface, will make it possible to find in the images of the new image segment support points well distributed over the geographical area, so that a good quality spatio-triangulation of the images of this area will be possible, whereas this would not necessarily have been the case in the image segments resulting from the logic directly linked to the passages of the sensor 1 in visibility of the reception stations 6.

[0048] It will also be possible to adapt the spatio-triangulation processing according to the geographical areas covered by the image segments, for example by adapting the complexity of an error model of a spatio-triangulation processing to a geographical area. Thus, in an area where statistically few reference points are found, for example an island or a snowy area, a less complex error model will preferably be used, for example a lower-order polynomial error model.

[0049] Finally, the third segmentation parameter considered by method 100 is an extension of the geographic area along a sensor's scanning direction. This extension must be constrained to avoid, on the one hand, excessive complexity in the geolocation processing of the images in the new image segment, and on the other hand, to allow for the rational allocation of resources available for processing all segments, avoiding the use of these resources on segments covering geographic areas that are too small. Furthermore, when the geographic extension of the area covered by the image segment is too large, the time period associated with the image segment is also significant, which consequently increases the complexity of the error model associated with the geolocation processing by saptio-triangulation of the images in the new image segment.

[0050] Thus, a maximum and a minimum extension, in the direction of the sensor 1's movement, of the geographical area covered by the new image segment 5, are determined for example on the basis of the available computing resources and are taken into account by the process 100 in the step of determining the new image segment 5.

[0051] According to an example of an implementation of the method 100, the step 102 for determining the new image segment 5 includes a process for selecting the new segment 5 images are selected from a set of potential new image segments. The new image segment 5 is selected based on a function calculated for each potential new image segment, such that the value of the function for the selected new image segment 5 is optimized. The selected optimum is, for example, a minimum or a maximum of the calculated function.

[0052] In particular, the function is defined as a linear combination of a first term and a second term, the first term being a first function of the first clipping parameter, i.e. of the useful registration area, of the new image segment, and of the second clipping parameter, i.e. of the distribution of the useful registration area, on the new image segment, the second term being a second function of the length, i.e. of the extension in the direction of movement of the sensor 1, of the geographical area covered by the new image segment.

[0053] Thus, the function in question can, for example, be written according to the following formula:

[0054] [Math 1]

[0055] E=C1+X^C2

[0056] in which Cl is the first term function of the first clipping parameter and the second clipping parameter, and C2 is the second term function of the third clipping parameter, i.e. of the extension in the direction of scrolling of sensor 1, of the geographical area covered by the new image segment,

[0057] and in which X is a scalar which allows to experimentally adjust a compromise between the first term and the second term.

[0058] The step of selecting the new image segment 5 corresponding to the optimum of the calculated function includes, for example, the implementation of a non-linear optimization method such as simulated annealing, graph-cutting, dynamic programming, with the linear optimization method being applied to a set of potential new image segments.

[0059] Fig. 2 is a representation of a succession of geographical areas covered respectively by a series of image segments 10, 11, 12, 13, 14, generated according to a logic other than that related to the optimization of the registration, for example according to a logic related to the passage of the satellite into visibility of the ground receiving stations.

[0060] Fig. 3 is a representation of another succession of geographical areas obtained from the succession of geographical areas of Fig. 2, after optimization of the segmentation according to an implementation method of the process described above.

[0061] It is observed in particular in [Fig. 2] that the island of Java is located at the end of segment 11, and that this same segment only has an extremely small useful area on Australia. Conversely, in [Fig. 3], it is observed that segment 2, proposed by the application of method 100 to image segments 10 and 11, covers more land in Australia, which makes it possible to increase both the useful registration area and the re- partitioning of support points, at the cost of a longer segment.

[0062] Furthermore, it can be seen in [Fig.2] that segment 14 is too long to be processed by a single machine, whereas in [Fig.3], a division of segment 14 is proposed into two shorter segments 5 and 6, by applying method 100 to image segments 10, 11, 12, 13, 14.

Claims

Claims

1. Method (100) for improving the geolocation of images acquired by scrolling an image sensor (1) along a scrolling path of the sensor (1), said sensor being configured to acquire at least one image segment (2, 3) during a period of time, the at least one image segment (2, 3) comprising a sequence of images of the Earth's surface acquired, from successive positions of the sensor along the scrolling path, at successive times of the period of time, the method comprising the following step: - determining (102) at least one new segment (5) of images acquired during a new period of time which respectively begins and ends within the time period of the at least one image segment (2, 3), the at least one new image segment (5) comprising a new sequence of images acquired at successive times of the new period of time time,the new sequence of images covering a geographical area observed by the sensor on the surface of the earth during the new period of time, said observed geographical area being defined by at least one clipping parameter.,

2. Method (100) according to claim 1, wherein the at least one segment of images (2, 3) comprises at least a first segment of images (2) acquired during a first period of time and at least a second segment of images (3) acquired during a second period of time, the method further comprising the following steps, implemented prior to the step of determining (102) at least one new segment (5) of images: - assembling (101) the first segment of images and the second segment of images, to form a merged segment (4) of images acquired during a cumulative period of time corresponding to the joining of the first period of time concatenated with the second period of time; - determination (101bis) that, for the step of determining (102) at least one new segment (5) of images, the at least one segment (2, 3) is the merged segment (4) and that the time period is the cumulative time period.

3. The method of claim 2, wherein the first time period ends at a time when the second time period begins. time.

4. Method according to one of claims 1 to 3, in which the at least one cutting parameter comprises for example: a useful registration surface, a distribution of the useful surface, an extension of the geographical area covered by the new segment of images along a scrolling direction of the sensor.

5. Method according to claim 4, in which the useful recalibration surface is a land surface, the land surface being included within the geographical zone, and being associated with a value of a visibility parameter, the value of the visibility parameter being greater than or less than a predetermined threshold.

6. The method of claim 5, wherein the land area is determined after excluding the area of specific areas.

7. Method according to one of claims 1 to 6, the step of determining (102) the at least one new image segment comprises a selection of the at least one new image segment from a set of potential new image segments, the at least one new segment being selected on the basis of a function calculated for each potential new image segment, so that the function is optimized by the at least one new segment selected.

8. The method of claim 7, wherein the function is defined as a linear combination of a first term and a second term, the first term being a first function of the useful area of the new image segment and the distribution of the useful area over the new image segment, the second term being a second function of the extension of the geographical area covered by the new image segment along a scrolling direction of the sensor.

9. Method according to claim 8, in which the geographical extension of the geographical area covered by the at least one new image segment is bounded by a minimum value and / or by a maximum value.

Images