A method and system for terrain accuracy evaluation based on image shadows

Abstract

The present application relates to a kind of topographic precision evaluation method and system based on image shadow, method includes: first, based on digital terrain simulation the solar reflection light intensity at the moment of image shooting, make simulation image, real image is projected to terrain surface and is registered to simulation image;Then in real image and simulation image respectively, extract shadow, and calculate the deviation at the end point of shadow;Finally, based on the optical and statistical connection between shadow end point deviation and terrain elevation error, the mean error of terrain elevation relative to elevation true value is quantitatively calculated.Compared with prior art, the present application realizes the quantitative mean error evaluation of digital terrain relative to real surface in the absence of high-precision ground test point, and has statistical significance.

Description

Technical Field

[0001] This invention relates to the field of terrain quality assessment, and in particular to a method and system for terrain accuracy assessment based on image shadows. Background Technology

[0002] Due to the extreme scarcity of high-precision ground verification points on the surfaces of extraterrestrial objects—for example, only a few laser reflectors from the Apollo program can serve as verification points on the lunar surface—the accuracy evaluation of digital terrain on extraterrestrial surfaces is quite difficult. Currently, the quantitative evaluation of the accuracy of terrain products on extraterrestrial surfaces mainly adopts cross-validation. However, this method can only obtain the relative accuracy between terrain products and cannot directly obtain quantitative accuracy that is statistically significant relative to the true terrain value, thus raising doubts about the reliability of the evaluation results.

[0003] In existing planetary surface topographic mapping based on orbiter imagery, shadows are typically considered interference signals and not involved in the generation of 3D terrain. Therefore, for current digital terrain products, shadow signals and digital terrain signals can be considered independent. However, in reality, there is a clear connection between shadows and terrain. Based on the distribution of shadows in the image, it is possible to infer the accuracy of digital terrain creation. Methods and products that simulate shadow distribution at the time of image capture based on digital terrain and solar azimuth from planetary ephemeris are now relatively mature. When the solar altitude angle is small, even slight differences in altitude can cause significant differences in shadow length; therefore, terrain accuracy assessment methods based on image shadows have the potential value of high sensitivity.

[0004] In summary, there is currently no method for quantitatively evaluating terrain accuracy using image shadow information, and this invention is of great significance in the field of three-dimensional mapping of extraterrestrial objects. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method and system for evaluating terrain accuracy based on image shadows, so as to realize the quantitative evaluation of digital terrain accuracy directly relative to the terrain true value based on image shadow information.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for evaluating terrain accuracy based on image shadows includes the following steps:

[0008] Acquire real images, and obtain the relative position of the sun with respect to the three-dimensional shape of the lunar surface according to the time of the real image capture. Back-project the digital terrain corresponding to the real image to obtain the digital three-dimensional shape. For each unit of the digital three-dimensional shape, calculate the incident light information, and further calculate the reflected light information according to the physical reflection model. Define the rasterized reflected light information as the simulation image. Project the real image onto the surface of the digital three-dimensional shape, register it with the simulation image, and then rasterize it according to the same rules as the simulation image to obtain the projected real image.

[0009] The projected real image is pixel-level registered to the simulation image, and an adaptive threshold for extracting shadows is calculated. The projected real image and the simulation image are binarized according to the adaptive threshold, and the shadow endpoint coordinates are extracted respectively. For each shadow endpoint coordinate of the projected real image, the nearest shadow endpoint coordinate is searched in the simulation image along the direction of the solar azimuth angle, and the distance between the two is used as a sampling element of the shadow deviation.

[0010] For each sampling element of the shadow deviation, based on the solar elevation angle, and the height difference between the shadow deviation and the shadow endpoint of the simulated image and the projected real image, the difference in elevation error between the shadow occlusion point and the real shadow endpoint is calculated. Then, based on the statistical standard deviation of this difference, the elevation mean square error of the terrain product is calculated, and the terrain accuracy evaluation result of the terrain product is obtained.

[0011] Furthermore, the back-projection of the digital terrain corresponding to the real image includes:

[0012] The digital terrain is converted into a 3D point cloud representation in a Cartesian coordinate system, where the negative z-axis points to the center of the moon and the negative x-axis points to the light source. The terrain is then re-rasterized along the xOy plane in the Cartesian coordinate system, and the z-coordinates are filled into the raster elements to obtain the digital 3D shape.

[0013] Furthermore, the calculation process for the incident light information includes:

[0014] For each unit of the digital 3D topography, a ray is drawn along the solar azimuth angle, and the terrain elevation angle is searched along the ray. The horizontal line along the terrain elevation angle is defined as the horizon. Combining the horizon, the solar altitude angle, and the radius of the solar disk, the visible proportion of the solar disk is calculated as the incident light information.

[0015] Furthermore, the calculation process for the reflected light information includes:

[0016] The reflected light information is calculated based on the incident light information, surface albedo, and the dot product between the solar azimuth vector and the terrain surface normal vector.

[0017] Furthermore, the surface albedo is obtained from the albedo map in the physical reflection model, which is preset by the physical reflection model or selected as needed.

[0018] Furthermore, the calculation process of the adaptive threshold includes:

[0019] Set an initial adaptive threshold, normalize the pixel values ​​of the registered simulation image and the projected real image, and then use the adaptive threshold as the binarization threshold to binarize the simulation image and the projected real image respectively, and extract the shadows.

[0020] Then, the adaptive threshold is adjusted according to a fixed step size and used as a binary threshold to extract shadows, resulting in multiple sets of shadow distribution binary maps;

[0021] For each group of shadow distribution binary maps, count the total number of grid cells with the same value at the corresponding grid cells, and record it as the shadow similarity. Select the adaptive threshold corresponding to the shadow distribution binary map with the highest shadow similarity as the adaptive threshold for subsequent shadow extraction.

[0022] Furthermore, the elevation error at the shaded point is the terrain elevation minus the true elevation value; both the elevation error at the shaded point and the elevation error at the true shadow endpoint are random samples of the digital terrain elevation error. Based on the statistical relationship between the difference of random sampling errors and the standard error, the elevation standard error of the terrain product is calculated from the statistical standard deviation.

[0023] The present invention also provides a terrain accuracy evaluation system based on image shadows, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to perform the following steps:

[0024] Acquire real images, and obtain the relative position of the sun with respect to the three-dimensional shape of the lunar surface according to the time of the real image capture. Back-project the digital terrain corresponding to the real image to obtain the digital three-dimensional shape. For each unit of the digital three-dimensional shape, calculate the incident light information, and further calculate the reflected light information according to the physical reflection model. Define the rasterized reflected light information as the simulation image. Project the real image onto the surface of the digital three-dimensional shape, register it with the simulation image, and then rasterize it according to the same rules as the simulation image to obtain the projected real image.

[0025] The projected real image is pixel-level registered to the simulation image, and an adaptive threshold for extracting shadows is calculated. The projected real image and the simulation image are binarized according to the adaptive threshold, and the shadow endpoint coordinates are extracted respectively. For each shadow endpoint coordinate of the projected real image, the nearest shadow endpoint coordinate is searched in the simulation image along the direction of the solar azimuth angle, and the distance between the two is used as a sampling element of the shadow deviation.

[0026] For each sampling element of the shadow deviation, based on the solar elevation angle, and the height difference between the shadow deviation and the shadow endpoint of the simulated image and the projected real image, the difference in elevation error between the shadow occlusion point and the real shadow endpoint is calculated. Then, based on the statistical standard deviation of this difference, the elevation mean square error of the terrain product is calculated, and the terrain accuracy evaluation result of the terrain product is obtained.

[0027] Furthermore, the calculation process of the adaptive threshold includes:

[0028] Set an initial adaptive threshold, normalize the pixel values ​​of the registered simulation image and the projected real image, and then use the adaptive threshold as the binarization threshold to binarize the simulation image and the projected real image respectively, and extract the shadows.

[0029] Then, the adaptive threshold is adjusted according to a fixed step size and used as a binary threshold to extract shadows, resulting in multiple sets of shadow distribution binary maps;

[0030] For each group of shadow distribution binary maps, count the total number of grid cells with the same value at the corresponding grid cells, and record it as the shadow similarity. Select the adaptive threshold corresponding to the shadow distribution binary map with the highest shadow similarity as the adaptive threshold for subsequent shadow extraction.

[0031] Furthermore, the elevation error at the shaded point is the terrain elevation minus the true elevation value; both the elevation error at the shaded point and the elevation error at the true shadow endpoint are random samples of the digital terrain elevation error. Based on the statistical relationship between the difference of random sampling errors and the standard error, the elevation standard error of the terrain product is calculated from the statistical standard deviation.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] This invention compares the differences in shadow distribution between real and simulated images. When the solar altitude angle is small, even slight differences in altitude can cause significant differences in shadow length, thus providing higher sensitivity for terrain accuracy assessment. It also enables a quantitative and statistically significant mean error evaluation of digital terrain relative to the real surface in the absence of high-precision ground verification points. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating a terrain accuracy evaluation method based on image shadows provided in an embodiment of the present invention;

[0035] Figure 2(a) is a schematic diagram of a 5-meter resolution simulated image of the edge of Shackleton Crater in the South Pole of the Moon provided in an embodiment of the present invention;

[0036] Figure 2(b) is a schematic diagram of a 5-meter resolution projected image of the edge of Shackleton Crater in the South Pole of the Moon, provided in an embodiment of the present invention.

[0037] Figure 2(c) is a schematic diagram of the shadow endpoint deviation at a resolution of 5 meters at the edge of the Shackleton Crater in the South Pole of the Moon, provided in an embodiment of the present invention.

[0038] Figure 2(d) is a 5-meter resolution elevation error histogram of the edge of the Shackleton Crater in the South Pole of the Moon, provided in an embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0040] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0041] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0042] Example 1

[0043] This embodiment provides a terrain accuracy evaluation method based on image shadows, which mainly includes the following aspects:

[0044] I. Production of Simulated Images

[0045] A: Calculation of incident light: including finding the relative position of the sun at the time of image capture from planetary ephemeris; including traversing terrain units to calculate the visible scale of the solar disk, and raster reconstruction;

[0046] B: Calculation of reflected light: This includes estimating the terrain neighborhood normal vector and calculating the reflection ratio of the incident light intensity based on the light source vector and the terrain normal vector.

[0047] II. Statistics on Shading Deviation

[0048] This includes image projection and alignment onto the terrain, calculation of the adaptive shadow threshold, determination of shadow endpoints, and calculation of the distance between shadow endpoints.

[0049] III. Calculation of Topographic Error

[0050] A: Extracting shadow endpoints: This includes image projection and alignment onto the terrain, calculation of the shadow adaptive threshold, and determination of shadow endpoints;

[0051] B: Calculation of terrain mean square error: including statistics on the error at the end of the shadow and extraction of the terrain elevation at the end of the shadow.

[0052] like Figure 1 As shown, the steps in this embodiment are detailed below:

[0053] S1: Acquire real images, consult ephemeris at the time the real images were captured, obtain the relative position of the sun with respect to the three-dimensional shape of the lunar surface, back-project the digital terrain corresponding to the real images to obtain the digital three-dimensional shape; for each unit of the digital three-dimensional shape, calculate its solar disk visibility and record it as incident light information; for each unit of the digital three-dimensional shape, calculate its surface normal vector in the neighborhood, input the surface normal vector and incident light information into the physical reflection model (e.g., Lambert model) to obtain reflected light information, and define the rasterized reflected light information as the simulation image; project the real images onto the surface of the digital three-dimensional shape, register them with the simulation images, and then rasterize them according to the same rules as the simulation images to define the projected real images;

[0054] In this embodiment, the above process specifically includes:

[0055] The terrain data is converted into a 3D point cloud in a Cartesian coordinate system, defined as follows: the negative z-axis points to the center of the moon, and the negative x-axis points to the light source, forming a right-handed rectangular coordinate system. The data is then re-rasterized along the xOy plane, and the raster elements are filled with z-coordinates. The solar azimuth at the time of the actual image capture is obtained by querying the planetary ephemeris, as well as the azimuth and elevation angles in the current coordinate system. For each terrain raster unit, a ray is drawn along the solar azimuth, and the terrain elevation angle is searched along the ray. The horizontal line along the terrain elevation angle is defined as the horizon. Combining the horizon, solar elevation angle, and solar disk radius, the visible proportion of the solar disk is calculated and defined as the incident light intensity. The incident light intensity is then substituted into the Lambert reflection model to obtain the reflected light intensity, i.e., the simulated image. The calculation process is shown in Equation (1).

[0056] Intensity=Incident×Albedo×dot(VecS,VecT) (1)

[0057] In Equation (1), Intensity is the pixel intensity of the simulated image, Incident is the incident light intensity, Albedo is the surface albedo, which is 1 by default if not specified, and dot(VecS,VecT) represents the dot product between the solar azimuth vector and the terrain surface normal vector.

[0058] S2: Perform pixel-level registration between the projected real image and the simulation image, calculate the adaptive threshold for extracting shadows, binarize the projected real image and the simulation image according to the adaptive threshold, and extract the shadow endpoint coordinates respectively; for each shadow endpoint coordinate of the projected real image, search for the nearest shadow endpoint coordinate in the simulation image along the direction of the solar azimuth angle, and use the distance between the two as a sampling element of the shadow deviation.

[0059] In this embodiment, the above process specifically includes:

[0060] The image is projected onto the terrain surface according to intrinsic and extrinsic parameters. Then, the projected image is pixel-level registered to the simulated image, and resampled according to the raster of the simulated image. The registered real and simulated images are first normalized in pixel value, and then shadows are extracted sequentially using a fixed step size, such as 0.01, as a binarization threshold. Each threshold combination corresponds to a set of shadow distribution binary maps. For each set of shadow distribution binary maps, the total number of raster values ​​with the same value is counted, and this is recorded as the shadow similarity. All threshold combinations are iterated through, and the threshold combination with the highest shadow similarity is found, which becomes the adaptive threshold for subsequent shadow extraction.

[0061] A pair of shadow maps are obtained by adaptive thresholding. First, starting from the first bright image point along the incident light direction, the odd-numbered shadow image point is recorded as the shadow start point and the even-numbered shadow image point is recorded as the shadow end point. For each shadow end point in the real image, the nearest shadow end point is searched along the x-axis in the simulated image, and the distance between them is recorded as the shadow deviation.

[0062] S3: For each sampling element of the shadow deviation, based on the solar altitude angle, and the height difference between the shadow deviation and the shadow endpoint of the simulated image and the projected real image, calculate the difference in elevation error between the shadow occlusion point and the real shadow endpoint. Then, based on the statistical standard deviation of this difference, calculate the elevation mean square error of the terrain product and obtain the terrain accuracy evaluation result of the terrain product.

[0063] This is equivalent to calculating the difference between the digital elevations of the shadow occlusion point and the shadow endpoint and the true elevation value; then, based on the statistical relationship between the difference in random sampling error and the mean square error, the mean square error of the digital terrain relative to the true terrain value is calculated.

[0064] In this embodiment, the above process specifically includes:

[0065] For each set of shading deviations, equation (2) can always be written:

[0066] E h1 -E h2 =E s ×tan(α)+(h r -h i (2)

[0067] Among them, E h1 E represents the elevation error at the shaded point (topographic elevation minus the true elevation value); h2 E represents the elevation error at the actual endpoint of the shadow in the image. s The shadow deviation is calculated as the x-coordinate of the shadow endpoint in the simulated image minus the x-coordinate of the shadow endpoint in the real image; α is the solar altitude angle; h r h is the terrain raster value (i.e., z-coordinate) corresponding to the end of the shadow in the simulated image; i Let be the terrain raster value corresponding to the end point of the shadow in the real image. Thus, since the right side of equation (2) is known, a statistic about the left side of equation (2) can be obtained, and the standard deviation σ can be calculated. s And E h1 and E h2 Both can be regarded as a random sampling of the elevation error of digital terrain. The elevation error σ of the terrain product and the standard deviation calculated according to equation (2) conform to the relationship shown in equation (3).

[0068]

[0069] Therefore, the elevation error σ of the terrain product can be obtained by equation (3).

[0070] Experiments and Analysis

[0071] The platform used in this experiment was the LRO NAC image M139811097L / R, which was selected as the image data source. An accuracy evaluation experiment was conducted on a 1km × 1km LOLA DEM with a resolution of 5 meters near the lunar south pole. At the time of image capture, the solar azimuth angle under the ME frame at the lunar south pole was 6.99°, and the local solar altitude angle was approximately 1.57°. The image was projected onto the terrain surface, and a simulated image was created based on the terrain and aligned to a unified coordinate system, as shown in Figures 2(a) and (b). The shadow endpoint deviation was extracted, as shown in Figure 2(c). The elevation error sampling value was calculated according to equation (2), as shown in Figure 2(d). The final elevation mean square error σ of the terrain within the target area was approximately 0.31 meters.

[0072] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for evaluating the accuracy of a terrain based on the shadow of an image, characterized in that, The method comprises the following steps: acquiring a real image, obtaining the relative position of the sun to the three-dimensional topography of the moon surface at the time when the real image is taken, and performing back projection on the digital terrain corresponding to the real image to obtain a digital three-dimensional topography; calculating incident light information for each unit of the digital three-dimensional topography, and further calculating reflected light information according to a physical reflection model, defining the reflected light information after gridding as a simulation image; projecting the real image onto the surface of the digital three-dimensional topography, and performing registration with the simulation image, and then gridding according to the same rule as the simulation image to obtain a projected real image; performing pixel-level registration on the projected real image and the simulation image, calculating an adaptive threshold for extracting a shadow, binarizing the projected real image and the simulation image according to the adaptive threshold, and extracting shadow end point coordinates respectively; for each shadow end point coordinate of the projected real image, searching for the closest shadow end point coordinate in the simulation image along the solar azimuth direction, and taking the distance between the two as a sampling element of shadow deviation; for each sampling element of the shadow deviation, calculating the difference between the elevation error at the shadow blocking point and the real shadow end point according to the solar elevation angle, the shadow deviation of the simulation image and the projected real image, and the height difference corresponding to the shadow end point, and then calculating the elevation mean error of the terrain product according to the statistical standard deviation of the difference, to obtain the terrain precision evaluation result of the terrain product.

2. The method according to claim 1, wherein, The back projection on the digital terrain corresponding to the real image comprises: converting the digital terrain into a three-dimensional point cloud representation in a Cartesian coordinate system, wherein the z-axis of the Cartesian coordinate system points to the moon center in the negative direction, and the x-axis points to the light source in the negative direction; regridding along the xOy plane in the Cartesian coordinate system, and filling in the z coordinate in the grid element to obtain a digital three-dimensional topography.

3. The method according to claim 1, wherein, The calculation process of the incident light information comprises: for each unit of the digital three-dimensional topography, a ray is drawn along the solar azimuth, the terrain view elevation is searched along the ray, the horizontal line along the terrain view elevation is defined as the horizon, and the visible proportion of the sun disk is calculated as the incident light information in combination with the horizon, the solar elevation angle and the solar disk radius.

4. The method according to claim 1, wherein, The calculation process of the reflected light information comprises: the reflected light information is calculated according to the point product between the incident light information, the surface albedo and the dot product between the solar azimuth vector and the terrain surface normal vector.

5. The method according to claim 4, wherein, The surface albedo is obtained from the albedo map in the physical reflection model, and the albedo map is pre-set or selected according to requirements in the physical reflection model.

6. The method of claim 1, wherein, The calculation process of the adaptive threshold comprises: setting an initialized adaptive threshold, performing pixel value normalization on the registered simulation image and the projected real image, then taking the adaptive threshold as a binarization threshold to binarize the simulation image and the projected real image respectively, and extracting shadows; then adjusting the adaptive threshold according to a fixed step length, and taking the adaptive threshold as a binarization threshold to extract shadows, to obtain a plurality of groups of shadow distribution binary images; The total number of grids with the same value at the corresponding grid of each group of shadow distribution binary graph is counted, and is recorded as a shadow similarity, and the adaptive threshold corresponding to the shadow distribution binary graph with the highest shadow similarity is selected as the adaptive threshold for subsequent extraction of the shadow.

7. The method of claim 1, wherein, The elevation error at the shadow blocking point is the difference between the terrain elevation and the true value of the elevation; the elevation error at the shadow blocking point and the elevation error at the true shadow end point are both random samplings of the digital terrain elevation error, and the elevation mean error of the terrain product is calculated from the statistical standard deviation of the difference between the random sampling errors and the mean error according to the statistical relationship between the random sampling errors and the mean error.

8. An image shadow-based terrain accuracy evaluation system, comprising: The method comprises the following steps: a true image is obtained, the relative position of the sun with respect to the three-dimensional lunar surface at the time when the true image is taken is obtained, and the digital terrain corresponding to the true image is back projected to obtain a digital three-dimensional terrain; for each unit of the digital three-dimensional terrain, incident light information is calculated, and further, reflected light information is calculated according to a physical reflection model, the reflected light information after gridding is defined as a simulation image; the true image is projected onto the surface of the digital three-dimensional terrain, and is registered with the simulation image, and is gridded according to the same rule as the simulation image to obtain a projected true image; pixel-level registration is performed between the projected true image and the simulation image, an adaptive threshold for extracting a shadow is calculated, the projected true image and the simulation image are binarized according to the adaptive threshold, and shadow end point coordinates are extracted respectively; for each shadow end point coordinate of the projected true image, the shadow end point coordinate closest to the shadow end point coordinate in the simulation image is searched along the solar azimuth direction, and the distance between the two is taken as a sampling element of the shadow deviation; for each sampling element of the shadow deviation, the difference between the elevation error at the shadow blocking point and the elevation error at the true shadow end point is calculated according to the solar elevation angle, the shadow deviation of the simulation image and the projected true image, and the height difference corresponding to the shadow end point, and then the elevation mean error of the terrain product is calculated from the statistical standard deviation of the difference, to obtain a terrain precision evaluation result of the terrain product.

9. The system for terrain accuracy assessment based on image shadows according to claim 8, characterized in that, The calculation process of the adaptive threshold comprises: an initialized adaptive threshold is set, pixel value normalization is performed on the registered simulation image and the projected true image, then the adaptive threshold is taken as a binarization threshold to binarize the simulation image and the projected true image respectively, and the shadow is extracted; then the adaptive threshold is adjusted according to a fixed step, and is taken as a binarization threshold to extract the shadow, to obtain a plurality of groups of shadow distribution binary graphs; the total number of grids with the same value at the corresponding grid of each group of shadow distribution binary graph is counted, and is recorded as a shadow similarity, and the adaptive threshold corresponding to the shadow distribution binary graph with the highest shadow similarity is selected as the adaptive threshold for subsequent extraction of the shadow.

10. The system for terrain accuracy assessment based on image shadows according to claim 8, wherein, The elevation error at the shadow occlusion point is the terrain elevation minus the true elevation; both the elevation error at the shadow occlusion point and the elevation error at the true shadow end point are a random sample of the digital terrain elevation error, and the elevation mean square error of the terrain product is calculated from the statistical standard deviation based on the statistical relationship between the difference of the random sample errors and the mean square error.

Images