Method and system for determining the status of a geographic line

Abstract

The present invention relates to a method for determining one or more conditions of a geographic line (GL) or its surroundings, the method comprising the steps of obtaining georeferenced Earth observation data representing one or more spatially overlapping image layers covering a particular region of interest (ROI) on the Earth's surface from an aerial or space perspective, the ROI including the GL, geometrically transforming, at least in part, the georeferenced observation data into a local internal reference frame of the GL within the ROI to obtain a mapping of the georeferenced observation data to respective corresponding coordinates in the local internal reference frame of the GL, and evaluating the mapped Earth observation data represented in the local internal reference frame of the GL according to a classification scheme to obtain evaluation data therefrom representing a classification of one or more characteristics of the GL or its surroundings according to one or more conditions of the GL.

Description

[Technical field]

[0001] The present invention relates to the field of image processing, and more particularly to the field of extracting and classifying features from images for classifying and monitoring conditions along geographic lines. In particular, the present invention relates to a method and system for determining the condition of a geographic line, for example a road, especially an unpaved road, a transportation line such as a railroad, or a border line. [Background technology]

[0002] A prominent example of land transportation lines, found in almost every country in the world, are roads. Specifically, the vast majority of roads around the world are unpaved. Official figures are only available for some countries, but overall estimates reach as high as 80%. This percentage correlates negatively with the development status of a particular country, but even some of the wealthiest and most developed countries have a significant number of roads that are unpaved.

[0003] The condition of such unpaved roads depends heavily on the current season and weather, vegetation, and frequency of use. Snow, ice, floods, landslides, and continued exposure to vehicles and large trucks can make these roads impassable for months. Although the traffic volume on these roads may be much lower, the risk of accidents is much higher than in urban areas. However, these transport lines are also essential, as 50% of the world's population lives along such roads. Their quality is a performance indicator for economic development, tourism and infrastructure accessibility of remote communities to hospitals, schools, and markets.

[0004] While unpaved roads are most susceptible to the aforementioned problems, other transportation lines, especially paved roads, can be subject to problems such as asphalt deterioration, potholes, geological phenomena, and vegetation overgrowth that require maintenance work that is difficult to predict, causing costly delays and damage and potentially even costing lives if, for example, a poorly maintained bridge collapses. At least some of these problems may also be associated with other transportation lines, for example railroad tracks or water transportation lines that run along rivers, pipelines, or power lines.

[0005] Referring again to roads as a non-limiting example of geographic lines, the assessment of road surface quality and its surrounding space in rural areas has some additional complexities when compared to metropolitan areas.

[0006] First of all, it should be noted that these roads are usually less frequently used than the streets and roads of large cities or major arterial roads. Moreover, monitoring systems that rely on on-the-ground quality assessments with regular updates are usually not feasible for technical or economic reasons. Technical equipment, staff, and funds limit road maintenance activities to actions taken after specific disasters or accidents or require long-term infrastructure planning. Thus, in the foreseeable future, up-to-date information collected "on the ground" may not be available for most local transport lines, especially unpaved roads, due to the nature of the problem.

[0007] Second, the quality of the road surface assessment depends heavily on the analysis of historical data, i.e., for the assessment of the current situation, it is important to know how much precipitation the area of ​​interest has received in the past hours, days or weeks. In addition to precipitation (rain, snow or any combination thereof), other meteorological parameters such as temperature and wind can cause road degradation or long-term erosion problems.

[0008] Third, road quality and condition are not only influenced by what is directly present on the drivable road surface, but also by the shape, condition, and events near the road, such as vegetation overgrowth, landslides, avalanches, rock falls, or flash floods. Some of these issues have a very local scope, on the order of a few meters, such as the shape of the road shoulder or the amount of off-road vegetation, while others, such as disastrous landslides or flooding, have a scope of several kilometers. Thus, systems for deriving the current situation and the condition of a particular road segment typically include data from a larger spatial extent in order to be able to provide reliable information.

[0009] Uses of multispectral satellite imagery range from a myriad of civilian applications such as geographical mapping, urban planning, precision agriculture, and traffic management, to pollution monitoring, and climate change analysis. Currently, i.e. in 2020, medium resolution satellite systems typically provide image data at a resolution of up to 10m x 10m per pixel with a revisit time of 3-6 days (e.g. Copernicus Sentinel 2 operated by the European Space Agency), while very high resolution systems typically provide a resolution of up to 30cm x 30cm per pixel. Even higher resolution can be obtained from aerial imagery acquired from drones or airplanes. Medium resolution imagery allows the detection and classification of roads and vegetation, while high resolution systems can resolve lane lines and even individual trees.

[0010] Satellite operators usually provide their data as ortho-images, thus transforming the raw image data to take into account variations in the Earth's surface topography and the tilt of the satellite or airborne sensor used to obtain the images. Satellite images usually provide intensity values ​​in the visible spectrum (red, green, and blue), but often contain other spectral components (e.g. infrared and radar) that are invisible to humans but can derive information about the observed area. Land cover such as "urban", "vegetation", and "water" kindsEach has a characteristic spectral response, so indices such as the Normalized Difference Vegetation Index (NDVI) have been developed to standardize related analyses. Such indices aggregate the spectral information from the various bands to simplify the evaluation of ground features.

[0011] Satellite images have been used excessively to classify land cover and vegetation for precision agriculture or climate change analysis. Known methods for analyzing such satellite images have in common that they are based on classical rectangular raster images that observe the Earth from above as a continuous area. Pre-processing techniques may be applied to reduce image distortions and other image artifacts as much as possible, taking into account the satellite tilt, atmospheric distortions, or altitude. Each pixel in the image thus represents a feature as if it were seen by a camera from directly above (zenith).

[0012] Typically, these images are georeferenced, i.e. their pixels are aligned to a geographic coordinate system, e.g. longitude and latitude related coordinates, or Universal Transverse Mercator (UTM) coordinates. In other scenarios, the satellite's native orientation may be used to avoid costly pre-processing steps. However, transportation lines, whether man-made such as roads, railways, or canals, or natural such as rivers, generally do not coincide with a common geographic coordinate system.

[0013] Furthermore, various methods have been proposed to classify road surfaces based on on-board sensors in order to assess traction or optimize suspension. Such systems have been proposed for aquaplaning mitigation, determining road conditions, classifying winter road surface conditions, or classifying road surface roughness using tire sensors. [Prior art documents] [Patent documents]

[0014] [Patent Document 1] European Patent Application Publication No. 3486608 Summary of the Invention [Means for solving the problem]

[0015] It is an object of the present invention to provide an improved technique for determining the status of a geographic line. In particular, it is desirable to improve one or more of the efficiency, effectiveness, accuracy, and error rate of monitoring one or more conditions of a geographic line.

[0016] A solution to this problem is provided by the teaching of the independent claims.Various preferred embodiments of the invention are provided by the teaching of the dependent claims.

[0017] A first aspect of the invention is directed to a method, particularly a computer-implemented method, of determining one or more conditions at or around a geographic line (GL), comprising the steps of: (i) obtaining georeferenced Earth observation data representing one or more spatially overlapping image layers covering a particular region of interest (ROI) of the Earth's surface from an aerial or space perspective, the ROI including the GL; (ii) geometrically transforming, at least in part, the georeferenced observation data into a local internal reference frame of the GL within the ROI to obtain a mapping of the georeferenced observation data to respective corresponding coordinates in the local internal reference frame of the GL; and (iii) evaluating the mapped Earth observation data represented in the local internal reference frame of the GL according to a classification scheme to obtain evaluation data therefrom representing a classification of one or more properties of the GL or around the GL according to one or more conditions, such as a physical, chemical, or biological condition of the GL.

[0018] The method may therefore be considered as a sophisticated measurement method for measuring one or more, in particular physical (e.g., meteorological, geospatial, topographical, geological, electromagnetic), chemical or biological conditions of the GL, where one or more measurement results represented by the assessment data are determined based on a set of measured georeferenced Earth observation data, e.g., data generated by one or more image sensors (e.g., satellite-based or aircraft-based), and optionally further sensor data. In addition to the current measurements of said conditions, measurements related to their temporal evolution, e.g., annual average values ​​(e.g., for properly modeling permafrost soils), can be determined.

[0019] The term "geographical line", as used herein, refers in particular to a line on the surface of the Earth, i.e. a continuous, primarily one-dimensional geometric structure of longer straight or curved lines. A geographical line may be an artificial object (such as a road, a railway line, a power line, a dam, a bridge, a canal, a pipeline, or a boundary line) or a natural object, such as a river. In particular, a geographical line may be a transportation line, i.e. a line that forms or is part of a transportation infrastructure for transporting, for example, goods, people, or energy. A geographical line may correspond to a section of a more complex network comprising multiple geographical lines, such as, for example, a road network comprising multiple interconnected roads, a railway system comprising multiple interconnected railway tracks, or a pipeline system comprising multiple interconnected pipeline branches.

[0020] The term "surroundings" of a geographic line, as used herein, refers especially to the immediate vicinity of the geographic line, typically within a range of less than a meter up to several hundred meters, measured in a direction perpendicular to the direction of the line at any given point on the line.

[0021] The term "condition of a geographical line" and similar expressions, as used herein, refers to a detected or detectable condition, particularly the current condition of a geographical line at one or more locations on the line. Specifically, each condition or condition is a physical, chemical, or biological condition of the geographical line, or a combination of two or more of these, which condition or combination of conditions can be detected by image analysis or one or more suitable sensors. Without limitation, physical or chemical or biological characteristics of a road pavement, such as asphalt, the location and dimensions of potholes in the road, geological phenomena and vegetation overgrowth, the presence of landslides, avalanches, rock falls, or flash floods, if and to what extent they affect the geographical line, are all examples of conditions or combinations of conditions of a geographical line.

[0022] The term "acquiring" data, e.g., georeferenced Earth observation data, can refer, inter alia, to (i) generating that data by "acquiring" the entity itself, e.g., using one or more sensors or deriving it from input data, or (ii) receiving that data from an external data source.

[0023] The term "geo-referenced reference geometry" as used herein refers to a geometry represented by a respective datum defined or expressed in a geographic coordinate system, such as, for example, longitude and latitude or UTM coordinates. In particular, but not exclusively, the geometry of a particular road represented by an electronic map serving as a reference, such as an electronic map based on map data for a car navigation system, a map or navigation application on a computer or handheld device, such as, for example, a smartphone or tablet computer, is a geo-referenced reference geometry. Of course, the same applies to the respective reference geometry of other geographic lines, such as, for example, railways, national borders, rivers, etc., which may in particular also be represented by respective map data or other geo-referenced geometry data.

[0024] The term "local internal frame of reference of the geographic line (GL)" as used herein refers in particular to a reference frame, such as a coordinate system, and is distinct from a georeferenced reference frame in general. A geographic line is in the local internal reference frame and is represented therein as a line of a predefined selected geometry, which may in particular be a straight line. Specifically, when the geographic line is a road, such a reference frame is often called "road space". Thus, a GL, which is a boundary of a geographic reference frame (geographic coordinate system) that uses, for example, longitude and latitude as coordinates or UTM coordinates, may be represented as a straight line in the local internal reference frame. For example, in the local internal reference frame, the coordinates of a given position on the GL may include (i) a first coordinate representing the distance of this point from a predefined reference point on the line, the distance being measured along the path of the line (e.g., road kilometers), and (ii) a second coordinate representing the distance of the given point from a (physical or virtual) centerline of the GL, measured along a straight line passing through the given point in a direction perpendicular to the GL.

[0025] The term "classification scheme" as used herein refers in particular to kinds By way of example and not by way of limitation, such a scheme refers to a scheme for classifying one or more states of a geographic line according to a set of kinds may be related to various physical conditions, e.g., the presence or absence of pavement, potholes, ice or snow, rocks or other objects on the road surface, various degrees of vegetation overgrowth at a given GL, current wind or precipitation conditions, etc.

[0026] The terms "first," "second," "third," and the like, in this specification and claims are used to distinguish between similar elements and are not necessarily used to describe a sequence or chronology. It is to be understood that the terms so used are interchangeable under appropriate circumstances, and that the embodiments of the invention described herein can operate in sequences other than those described or illustrated herein.

[0027] Where the terms "comprising" or "including" are used in the present description and claims, it does not exclude other elements or steps. Where an indefinite or definite article is used when referring to a singular noun, such as "a" or "an", "the", this includes the plural of that noun, unless something else is specifically stated.

[0028] Furthermore, unless explicitly stated to the contrary, "or" refers to inclusive rather than exclusive or. For example, condition A or B is satisfied by any of the following: A is true (or exists) and B is false (or does not exist), A is false (or does not exist) and B is true (or exists), and both A and B are true (or exist).

[0029] Thus, the method of the first aspect involves an effective, usually non-linear, geometric transformation of georeferenced Earth observation data from world space into a local internal reference frame of this GL in order to obtain a respective view of the GL in a local reference frame (e.g. an image or a combined stack of different images), the observation data referring to the area of ​​interest covering the GL of interest. In particular, this local reference frame and the way of representing the GL therein may form a very suitable and easy-to-use basis for a normalized representation of the GL (e.g. a normalized road space), which may be particularly useful for subsequent classification steps or comparisons with other GL-related data, for example between respective states of two or more different GLs, or between respective states of the same GL, but at different times. Moreover, because of the transformation, the method allows a variety of different Earth observation data (particularly image data) to be used as input, in particular independently of the respective orientation or resolution at which that data was captured. The resulting evaluation data obtained from the classification may be used in many different applications, as will be explained in more detail below.

[0030] In particular, Earth observation data such as satellite imagery can be used to fill gaps in data regarding road surface quality in rural areas where sufficient data from the ground is not available. Thus, the method of the first aspect provides a standard way of using satellite imagery, and optionally supplemental data, to provide a normalized view of road segments and other geographic lines.

[0031] An additional advantage of the method can be achieved in that otherwise unconnected image regions can be transformed into continuous regions based on the geometry of the carrier lines, which may reveal otherwise non-trivial spatial and temporal patterns. Furthermore, some methods, such as machine learning, require input data with a uniform structure resulting from this method, and can project data of different dimensions into a rectified road space.

[0032] In the following, preferred embodiments of the method are described, which can be combined in any way with each other or with other aspects of the invention, unless such combinations are explicitly excluded or technically impossible.

[0033] In some embodiments, the step of geometrically transforming the georeferenced observation data, at least in part, into the local internal reference frame of the GL includes: (i) defining a plurality of control points in the ROI at different locations along the GL; (ii) for each control point, determining a respective local cell grid that covers a limited local spatial neighborhood of the control point, the resulting set of local cell grids defining the local internal reference frame of the GL; and (iii) geometrically transforming the observation data, at least in part, into the local internal reference frame of the GL to obtain a mapping of the observation data to the set of local cell grids of the control points. This approach is particularly flexible in that the local internal reference frames are defined by a combination of local cell grids, each of which may be individually optimized, in particular with respect to their orientation relative to the corresponding control point. Furthermore, this approach is also very efficient, since the set of local cell grids as a whole need not cover the entire GL, but may instead cover only a portion of it, such as a set of selected segments of the GL within the ROI and their respective neighborhoods, such that the total data volume of the transformed observation data is effectively limited, and in particular may be less than the total data volume of the original georeferenced observation data covering the ROI. One or more of the local cell grids may have a rectangular pattern, although other shapes are possible, such as, for example, hexagonal cells within a corresponding hexagonal grid. Indeed, the term "local cell grids" as used herein includes the possibility that the local cell grids may include or even consist of one or more cells separated from each other by intermediate spaces, for example, some unconnected circular cells arranged in a fixed pattern.

[0034] In some related embodiments, the method further comprises the step of concatenating the individual cell grids of the control points to form a data representation of an aligned, in particular regular, global cell grid that defines a local internal reference frame of the GL, the individual cell grids of the control points being arranged in an aligned aligned manner in a sequence corresponding to the sequence of the control points along the GL. This has the advantage of providing a particularly simple, compact and easy to handle reference frame, in particular allowing a particularly simple normalized representation of the transformed observation data or the resulting evaluation data, in particular for the above-mentioned purpose of comparing such data with other respective represented data.

[0035] In some of these embodiments, the evaluation of the respective Earth observation data is performed based on a representation of the Earth observation data mapped to the overall adjusted cell grid. This means that not only is the final result represented based on the entire adjusted cell grid, but also the actual preliminary evaluation of the transformed observation data (and, optionally, additional sensor data) has already been performed based on the entire adjusted cell grid. This further increases the simplicity and efficiency of the evaluation process, since adjusted reference frames can typically be processed with less effort than most other kinds of reference frames, e.g., bound ones. In particular, adjusted reference frames can be very easily represented by simple and storage-efficient data structures such as tables or matrices.

[0036] In some embodiments, at least one of the local cell grids is defined to have one or more of the following characteristics: (i) anisotropic arrangement of at least a subset of cells in the local cell grid; (ii) cell widths or heights or distances vary among at least a subset of cells in the local cell grid; and (iii) orientations of at least a subset of cells in the local cell grid are aligned to local orientations of the GL according to georeferenced reference geometries at the control points corresponding to the local cell grid. All of these particular variations (i) to (iii) have in common that they allow the associated local cell grid or grids to be optimally adjusted to the particular individual geometries of the GL to reflect the desired type of evaluation in the evaluation data. For example, if the GL is highly curved in the vicinity of a particular first control point, the cell grid of that first control point may be defined to have significantly different orientations, cell dimension variations, and anisotropic arrangements of cells in that local cell grid than the cell grid of a second control point elsewhere on the GL, where the GL is significantly straight.

[0037] In some embodiments, the georeferenced Earth observation data represents one or more image point values ​​for each image point of an original image point grid covering at least one image layer. Furthermore, the step of geometrically transforming the georeferenced observation data comprises determining at least one cell value for each cell of the local cell grid based on an aggregation of a subset of image point values ​​corresponding to an area of ​​the ROI covered by a respective cell of the local cell grid. This is a particularly effective method offering high flexibility and the possibility to optimally adjust the transformation to the desired type of evaluation. Furthermore, the use of multiple image point values ​​per image point allows for even more accurate or more complex evaluations, including evaluations that rely on combining more than two image point values ​​per image point to determine relevant evaluation data for that image point or its surrounding area.

[0038] Specifically, in some of these embodiments, aggregating said subset of image point values ​​comprises the following operations: (i) averaging the image point values ​​of all image points that lie entirely within the corresponding cell of the local cell grid; (ii) averaging the image point values ​​of all image points that overlap an area of ​​the ROI covered by the corresponding cell of the local cell grid, where each image point value is weighted according to the size of its overlap with that cell of the local cell grid; and (iii) weighting the image point values, where image points that are closer to the center of a cell of the local cell grid are weighted according to the size of their overlap with that cell of the local cell grid; and (iv) weighting the image point values, where image points that are further away from the center of the cell are weighted according to the size of their overlap with the center of the cell of the local cell grid. (iv) determining a median, minimum and / or maximum image point value among the image point values ​​covered by the corresponding cell of the local cell grid, (v) determining an image point value that represents a certain percentile of the distribution of image point values ​​in the corresponding cell of the local cell grid, (vi) determining one or more of a range, a standard deviation, a variance or any other higher order function that captures a characteristic of the distribution of image point values ​​in the cells of the local cell grid, and (vii) determining a histogram that represents the distribution of image point values ​​in the cells of the local cell grid. The result of the aggregation may be a single value or a data structure depending on the type of operation applied. For example, a histogram according to variant (vii) is typically represented by a data structure with multiple values, while an average of point values ​​according to any of variants (i) to (iii) is typically represented by a single value.

[0039] In some embodiments, at least one of the local cell grids is defined by a set of grid parameters, the method further comprising a step of determining at least one of said grid parameters based on a process including machine learning related to cross-validating intermediate results of the evaluation data obtained during the machine learning process with reference parameter data representing a known correct classification of the relevant one or more characteristics of the GL or its surroundings according to said one or more states. These embodiments in particular allow an automatic definition and even a dynamic adjustment or modification of the local cell grid by a machine learning based process, whereby the reference parameter data serves as training data for the machine learning process in a supervised learning sense.

[0040] In some embodiments, the step of obtaining the georeferenced Earth observation data includes one or more of the following steps: (i) obtaining georeferenced Earth observation data representing one or more of the image layers, each of which represents an image covering the ROI in a respective different spectral range of the electromagnetic spectrum; (ii) obtaining georeferenced Earth observation data representing one or more data layers as further image layers, each of which represents a respective sensory measurement or prediction, for example a simulation result of one or more local conditions of the GL in the ROI or its surroundings; and (iii) obtaining georeferenced Earth observation data representing one or more preprocessed image layers obtained from said image layers of variable (i) or a combination of two or more of the data layers of variable (ii). Both using two or more image layers for each variable (i) or supplementing such image layers with one or more said data layers according to variable (ii) or (iii) makes it possible in particular to extend the range of possible types of evaluations, or the respective achievable quality and accuracy.

[0041] Specifically, in some of these embodiments, the data layers representing respective sensory measurements or predictions of one or more local conditions at or around the GL within the ROI may include the following: (a) precipitation, (b) local radiation intensity or wavelength, (c) temperature, (d) air pressure, (e) wind force or direction, (f) if the GL is a geographic line over water in at least one section of the ROI: water flow speed and direction, (g) if the GL is a geographic line over land accessible to land vehicles in at least one section of the ROI: water flow speed and direction, (h) if the GL is a geographic line over land accessible to land vehicles in at least one section of the ROI: water flow speed and direction, (i) a data layer representing a sensory measurement or prediction of one or more local conditions at or around the GL within the ROI that may provide insight into the particular condition of the transportation line, (j) a data layer representing a sensory measurement or prediction of one or more local conditions at or around the GL within the ROI that may provide insight into the particular condition of the transportation line, (k ... provides insight into the particular condition of the transportation line, (k) a or vibration, orientation, speed, acceleration or any other vehicle parameter measured by a vehicle-based sensor or a number of sensors, (h) if the GL is a transportation line: the frequency of use of the transportation line, the maintenance status of the transportation line or the time since the last maintenance interval, age, or the configuration of the transportation line surface, (i) the climate zone, (j) the climate profile, (k) the expected temperature or precipitation, (l) the susceptibility to unexpected events such as accidents, natural disasters, floods, landslides, avalanches, storms, tornadoes, hurricanes, typhoons, and (m) the electromagnetic field, e.g. the signal strength of the cellular network covering the GL. This allows a very wide range of different assessments depending on the type of measurement or prediction selected from above, thus having a positive effect on the achievable quality and accuracy of the resulting assessment data, respectively.

[0042] In some embodiments, the method comprises a step of aligning the respective geometries of the GL in one or more image layers represented by the Earth Observation data with the georeferenced reference geometry of the GL to form a combined image stack, where at least two of the image layers are assembled in the stack, each of which represents a respective digital image at a given spatial resolution, and the spatial resolutions of at least two of the image layers differ from each other in at least one overlapping area, thus generating a combined multi-resolution image stack. Using such alignment of the respective geometries of the GL allows a particularly simple, efficient and reliable way of performing the geometric transformation, since it can be directly transformed by projection of the respective image data of the layers of the stack into the local internal reference frame of the GL. Furthermore, the multiple different resolutions increase the efficiency of the method, in particular in terms of limiting the amount of data to be processed, thereby improving the performance of the method, while at the same time maintaining the desired accuracy. This can be achieved by selectively using higher or lower of the available resolutions, in particular for the different layers, depending on the impact of the particular choice on the ability to achieve the required accuracy of each of the desired types of evaluations.

[0043] In some embodiments, the method further comprises applying a data compression process to reduce the spatial resolution of a given image layer, the data compression process comprising generating two or more sublayers of the image layer by resampling at least one region of the image layer with a spatial resolution respectively lower than the original spatial resolution and maintaining or resampling the spatial resolution in at least one other region of the image layer such that when the sublayers are superimposed to jointly define a combined multi-resolution representation of the image layer, the average resolution of the combined multi-resolution representation is lower than the original resolution of the image layer. Thus, instead of or in addition to varying the resolution across different image layers, according to these embodiments, data compression is applied within the same image layer. This can be advantageously used to further increase efficiency, for example in terms of speed of the method and data processing and storage efficiency.

[0044] Specifically, in some of these embodiments, the lower spatial resolution to which the at least one region of the image layer is resampled is a function of the distance of the region from the GL. In particular, the spatial resolution is lower at larger distances and higher at smaller distances. In this way, the achievable accuracy of the assessment related to the GL itself or its immediate vicinity may be selected to be higher than for remote areas around the GL. For example, it may be less important to detect vegetation details at a distance from the GL, but it may be important to detect vegetation overgrowth the GL with high accuracy, and the benefits of compression may outweigh the need for high resolution in those remote areas.

[0045] In some of the embodiments using sublayers, overlapping the sublayers comprises aggregating values ​​of spatially overlapping image points of different sublayers according to a predefined aggregation method to obtain, in the resulting combined multi-resolution representation, image point values ​​for image points that represent spatial regions corresponding to the overlapping image points of the different sublayers. In particular, one or more of the various aggregation methods described above in relation to the embodiments involving aggregating a subset of image point values ​​may also be applied here.

[0046] Furthermore, in some of the embodiments using sublayers, the image points of the first one of the sublayers are locally aligned to the respective local orientation of the GL at or near the respective image points, while the image points of the second one of the sublayers correspond to the normal orientation of the image points in the given image layer before compression. This has the advantage that, despite the image information lost due to compression, the most relevant image information, e.g., image information related to the GL or its immediate surroundings, remains available at a sufficiently high resolution and in an optimized orientation corresponding to the root of the GL. The resulting processed image layer may itself be used as the desired compressed version of the original image layer, or may instead be used only as an intermediate image, which is then further processed and in particular adjusted as described below to generate the final compressed version of the original image layer.

[0047] Thus, in some embodiments, overlaying and aggregating values ​​of spatially overlapping image points of different sublayers includes generating a combined multi-resolution representation of the image layer such that the orientation of the image points in the resulting representation is the same for all of these image points (i.e., the adjusted compressed multi-resolution image layer resulting from this operation).

[0048] In some embodiments, the method further comprises performing one or more of the following actions based on the evaluation data: (i) determining a navigation or information output of a vehicle navigation system; (ii) controlling dynamic traffic signs; (iii) adjusting one or more configurable technical parameters of the vehicle or other transportation system; (iv) triggering a maintenance or inspection process on the GL; and (v) predicting future conditions at or around the geographic line.

[0049] In particular, in the case of variant (i), the output of the vehicle navigation system may refer to a particular route that has been selected as the preferred route or as the route selected for navigation purposes among a number of possible routes leading to a particular destination. In the case of variant (ii), the dynamic traffic sign may in particular be a sign indicating whether a particular road segment can be entered or not, or a warning sign that can signal two or more different warning levels or categories depending on the current configuration, which may be controlled, i.e. selected, on the basis of the evaluation data. In the case of variant (iii), the one or more configurable technical parameters may in particular relate to a respective specific suspension setting of the vehicle or transport system. In the case of variant (iv), the maintenance or inspection process may in particular be triggered by causing an associated signaling, for example to notify maintenance personnel accordingly, or these processes may also be fully or partially automated processes that are automatically triggered by a respective trigger signal.

[0050] A second aspect of the present invention is directed to a data processing system configured to perform the method of the first aspect. The data processing system may in particular be configured by one or more computer programs to perform the method of the first aspect. Additionally or alternatively, the configuration may be implemented in whole or in part by respective hardware.

[0051] A third aspect of the present invention is directed to a computer program or computer program product comprising instructions which, when executed on a data processing system according to the second aspect of the invention, cause the system to perform a method according to the first aspect of the invention.

[0052] The computer program (product) may in particular be implemented in the form of a data carrier on which one or more programs for carrying out the method are stored. Preferably, this is a data carrier such as a CD, a DVD or other optical medium, or a flash memory module. This may be advantageous if the computer program product is intended to be traded as a separate product independent of the processor platform on which the one or more programs are executed. In another implementation, the computer program product is provided as a file on a data processing unit, in particular a server, and can be downloaded via a data connection, for example the Internet, or a dedicated data connection such as a dedicated or local area network.

[0053] The system of the second aspect may therefore have a program memory in which a computer program is stored, or the system may also be set up to access computer programs available externally, for example on one or more servers or other data processing units, via a communication link, in particular for exchanging data used during the execution of the computer programs or data representative of the output of the computer programs.

[0054] The descriptions, embodiments and advantages discussed above in relation to the method of the first aspect apply equally to the other aspects of the invention.

[0055] Further advantages, features and applications of the present invention are provided in the following detailed description and accompanying drawings. [Brief description of the drawings]

[0056] [Figure 1]1 illustrates a schematic diagram of an exemplary embodiment of the method according to the invention; [Diagram 2] FIG. 2A is a schematic diagram of an exemplary merged image stack with aligned layers; [Diagram 3] 6 illustrates a set of exemplary outputs of possible evaluation results of the method of FIG. 1 or FIG. 5. [Figure 4] FIG. 2 shows a schematic diagram of a data compression process of an image layer according to a particular embodiment of the method of the present invention; [Diagram 5] FIG. 1 is a detailed flowchart illustrating one embodiment of a method according to the present invention, the method specifically comprising: (i) a model creation step for optimizing a parameterized model of the method; and (ii) a model application step for applying the optimized model as an instantiation of the method to actually perform a determination of evaluation data representative of a classification of one or more properties of the GL or its surroundings according to one or more states of the GL. [Figure 6] 1A-1C are diagrams illustrating various different exemplary embodiments of a sub-grid 111. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0057] With reference to FIG. 1 , an exemplary embodiment 100 of the method of the first aspect of the present invention comprises a step S1 of obtaining georeferenced Earth observation data comprising aerial image data obtained from one or more drones 101, aircraft 102, satellites 103 or any other device capable of recording multispectral observation data of the Earth 104, and in particular its surface.

[0058] In a further step S2, a particular region of interest (ROI) 105 is selected, for example a specific location, a rectangular region, a specific country, state, city, or a region of any shape, or any combination / union of such regions, which may not be contiguous. Further processing of the obtained observation data focuses on that ROI. In this way, a stack 106 of multiple image layers (each image layer forming one image) is obtained, in particular each image layer may correspond to a respective associated spectral band, such that the stack may contain multiple spectral bands in total in different image layers. For example, a typical RGB image layer contains three channels. Satellite images may contain additional infrared bands, for example Sentinel-2 satellite images provide 12 bands. Active sensors (radar or lidar) with an active signal emission component may provide one or more additional image layers independent of natural light sources. Furthermore, these image layers may contain preprocessed images, generated by combining image data from different channels or by preprocessing methods for obtaining a digital elevation model. An image stack may contain images from the same source, e.g., a particular satellite system, but recorded on different dates (time series) or from different source systems, e.g., various satellite systems or aircraft- or drone-based Earth monitoring systems. Further image layers in the stack 106 may be data layers representing sensor data obtained from sensors other than image sensors, e.g., vehicle-based sensors such as land sensors or vibration sensors.

[0059] In a further step S3, one or more georeferenced geographic lines GL or a network of GLs 107, such as a road network (or any other transport line network or geographic boundary), are aligned with the georeferenced image stack 106 to obtain a combined image stack 108. The transport line network can be obtained (a) manually, for example by an operator who defines the geographic coordinates of the transport lines to be evaluated, (b) from an existing database such as "OpenStreetMaps", (c) via any image processing method that derives vector data from images, or (d) automatically via GPS (or Galileo or Glonass, etc.) tracks of vehicles recorded. The transport line network can be defined by a graph consisting of a series of nodes (defined by latitude and longitude coordinates on the Earth) and their connections. This is the most common case, but does not exclude other ways of defining lines, such as Bézier curves, splines, or any other parametric curves.

[0060] FIG. 2 shows an exemplary combined image stack 200 (i.e., an exemplary instantiation of the combined image stack 108 described above) including six different image layers 210 to 260. The combined image stack 200 has already been aligned by step S3 with respect to the GL in which the evaluation is performed using the method 100. By way of example, the layer 210 may represent an image of the visual part of the electromagnetic spectrum, represented for example in the RGB color space, or any other color space suitable for this spectral range. The image data of the layer 210 may be provided by a satellite system, for example Landsat or Sentinel. The layer 220 may represent an elevation / topology image provided by a digital elevation model, possibly generated by combining image data from various channels, for example also from radar channels. A further layer 230 may represent an image showing the spatial distribution of vegetation, and optionally the type of vegetation, for a covered region of interest (ROI). Yet another layer 240 may represent an image showing the type of surface present at various locations within the ROI. This layer may be derived in particular from image analysis of layer 210, or by combining data from several image layers and / or other data sources, such as ground sensors, to determine the type of surface of the GL itself, especially if the GL is a road. Layer 250 may represent evaporation data over the ROI, which may be particularly relevant for detecting the characteristics of the GL's surroundings, for example, areas covered by forests, meadows, rivers, lakes, snow, etc. Layer 260 may represent meteorological data of the ROI, such as precipitation forecast data. Further layers may represent, for example, historical meteorological data of the ROI. It should be noted that image layers ("layers" for short) are therefore not limited to representing optical images, but may also represent other data that can be used to characterize the conditions of the ROI.This data may be data provided directly by one or more image sensors or other sensors, or it may be derived solely from such sensor data, e.g., by calculation and / or by combining two or more different types of input data, e.g., two or more other image layers, to derive data for the image layer in question.

[0061] Referring again to FIG. 1, the method 100 further comprises a step S4, in which a sample of control points 109 is defined for every continuous line segment 110 of the network 107, i.e. for at least one GL in the network 107, for example by selecting evenly distributed points on the line segments 110. Depending on the use case, other sampling schemes may be suitable. For example, if the GL to be monitored contains inherent control points, these existing control points can be used instead of a mathematical sampling scheme. A typical scenario is the monitoring of overhead power lines, where the control points 109 can be sampled at the locations of the transmission towers / electricity pylons.

[0062] In step S5, a respective sub-grid 111 is placed at each control point 109, and the orientation of each sub-grid is determined from the local orientation of the GL at that control point 109. Specifically, the sub-grid 111 may be defined as a regular linear grid that is placed in a direction perpendicular to the path of the GL according to a georeferenced reference frame of the image data. The captured spectral variability and visual complexity around the monitored GL are controlled by the grid parameters of the sub-grid 111 (number of cells, width and height of each cell) and the number of control points 109.

[0063] In one exemplary implementation, these parameters are received via a user interface. They may therefore be selected manually based on expertise in a particular field. In another exemplary implementation, these parameters may be optimized by machine learning and cross-validation methods, i.e., the method is trained with various parameter sets and the results are compared with known reference data (supervised learning). The parameter set that achieves the highest accuracy may then be selected as the final model. In yet another exemplary implementation, multiple parameter sets are combined and used simultaneously for a particular prediction method. Thus, at each control point 109, a multi-resolution image stack is acquired. A typical example is road monitoring, where a high-resolution grid can detect certain road marking patterns and a low-resolution grid can capture features of non-road vegetation.

[0064] Now, in a process that includes further steps S6 and S7, the image stack is geometrically transformed into the local internal reference frame of the GL. To that end, in step S6, in each sub-grid 111, image features from the source image 112 are aggregated by a suitable aggregation function 114 over the image region where the source image 112 overlaps with the respective sub-grid 111 to obtain either a single value or data structure for each grid cell 113 of the respective sub-grid 111.

[0065] Exemplary suitable aggregation functions 114 are: (a) averaging the pixel intensities of all pixels of the respective source image 112 that lie entirely within a cell 113 of the subgrid 111; (b) averaging the pixel intensities of all pixels of the respective source image 112 that overlap with the cell 113, modified by the proportion of space they contribute to the cell; (c) a weighted average of pixel intensities, where pixels at the center of the cell 113 get a higher weight than pixels at the border; (d) median, minimum, and maximum intensities; (e) intensities that represent a certain percentile of the intensity distribution within the cell 113, e.g., to account for outliers; (f) range, standard deviation, variance, or any other higher order function that captures characteristics of the intensity distribution within the cell 113; (g) histograms of intensities; (h) histograms of orientation or other image features. The latter two are special cases that increase the dimensionality of the output space (e.g., road space), but provide a more detailed characterization. This is not an exhaustive list of potential feature extraction methods, but other methods may be appropriate for specific use cases.

[0066] The subgrid 111 may be anisotropic, i.e., the width and height of the grid cells may not be equal, to model that certain characteristics of the transport line vary significantly with lateral distance from the center of the transport line, but remain largely the same along the same transport line. Additionally, the subgrid 111 may not be equally spaced, i.e., the width and height of the grid cells may differ, to model stronger smoothing when far away from the transport line. Other variations between cells are also possible.

[0067] FIG. 6 specifically illustrates a set 600 of various different exemplary embodiments of the subgrid 111. Specifically, FIG. 6(a) and FIG. 6(b) respectively illustrate examples of elements of the first category ("regular") of subgrids 610 and 620, in which the individual cells 615, 625 are arranged in a regular order, such as a one-dimensional matrix arrangement 610 or a two-dimensional matrix arrangement 620, in either case all the cells 615, 625 have the same shape, size, orientation and distance (center-center) from their respective neighbors. FIG. 6c illustrates an exemplary subgrid 630 of the second category ("anisotropic"), characterized in that the shapes and sizes of the various cells 635 may vary. In particular, for rectangular cells 635, their widths and heights may vary across the set of cells in the subgrid. 6d and 6e show exemplary subgrids 640 and 650 of a third category ("Non-gapless"), respectively, characterized in that at least some cells of the subgrid are spaced apart. In FIG. 6d, the cells are equally spaced, while in FIG. 6e, the distance between adjacent cells may vary ("non-equidistant"). Further embodiments of the subgrid 111 may be obtained by simultaneously implementing two or more of the above examples (unless inconsistent). For example, the two-dimensional aspect of the array 620 may be mixed with any of the embodiments 630, 640, and 650 to define a two-dimensional version of a cell array according to each of the embodiments 630, 640, or 650.

[0068] Referring again to FIG. 1, in a subsequent step S7, the obtained values ​​for each cell 113 in each sub-grid 111 are: Regular The layers are then arranged on a grid 115 (e.g., in a rectified road space if the GL corresponds to a road), thus generating, for each layer, a geometrically transformed rectified image of the GL and its surroundings, which can be used for further processing and visualization purposes. RegularGrid 115 defines a local internal reference frame for the GL, and the locations of individual cells of grid 115 can be considered as coordinates in this local internal reference frame.

[0069] The method 100 further includes an evaluation step S8, in which the stack of transformed adjusted images of the GL and its surroundings is evaluated by applying a classification scheme to obtain evaluation data representative of a classification of one or more properties of the GL or its surroundings according to one or more conditions, e.g., a physical, chemical, or biological condition of the GL.

[0070] Fig. 3 shows exemplary assessment data obtained from performing a method of the invention, e.g. the method of Fig. 1 or Fig. 5, relating to several different classifications of GL in road morphology. In this example, the classifications relate, among others, to the type of road surface, the road gradient, flood risk, avalanche risk, overgrowth index (indicating the level of vegetation on the road, see exemplary assessment data 116 in Fig. 1), smoothness of the road surface, air temperature along the road and temperature of the road surface. In Fig. 3, the spatial expansion of the GL, i.e. the adjacent road segments of interest, is expanded from left to right in the top RGB image of Fig. 3, and the various rows of data items shown below are arranged accordingly.

[0071] Thus, the method of FIG. 1 (the same applies to the method of FIG. 5 described below) is configured to provide a normalized image stack including one or more image layers such that, with respect to the geographic lines (GLs) of interest, these GLs are always displayed in a straight line, moving from left to right along the horizontal centerline of the normalized image. Other arrangements of sub-grids are similarly covered, for example, arranged row by row so that roads become columns in the image. Those skilled in the art will take into account state-of-the-art considerations to select an appropriate memory layout, CPU architecture, and file system, for example, storing each image layer in a single file may cause significant overhead in certain file systems. The aforementioned method may be implemented in software or with dedicated hardware, for example, semiconductor chips.

[0072] Figure 4 illustrates an image compression process 400 that may be used, particularly in conjunction with any one of methods 100 of Figure 1 or 500 of Figure 5, to compress image data representing an original image layer 410. In particular, process 400 may be used to provide GL-focused compression of satellite imagery.

[0073] In this scenario, an aerial or satellite image covering the ROI is processed in an unrelated reduction-based approach that focuses on image points (pixels) in the original image layer 410 that represent the ROI belonging to a GL of interest. In this example, the GL of interest is selected to be a particular transportation line in the transportation line network 411, e.g., a road network represented in the image 410. However, the compression process of FIG. 4 does not depend on the selection of a particular GL with the transportation line network 411, but rather relates to the entire ROI, and thus the entire transportation line network 411 and its surroundings.

[0074] In this example, all GLs (transport lines) from the original image 410 are densely sampled in the manner described above for transport line alignment at multiple resolution levels, e.g., the GL itself may be stored at native resolution 430, the immediate surroundings may be stored at half the native resolution 440, etc. Thus, all satellite images are represented by a low-resolution base image and a series of high-resolution GL-related images. Specifically, in this example, the pixels representing the transport line network 411 itself are densely sampled using the techniques described above (see FIG. 1 ) including subgridding, and furthermore, the spatial resolution of the surrounding pixels varies depending on the lateral and vertical distances of their surrounding pixels to a reference pixel, e.g., pixel 431 that is part of the transport line network. For example, the transport line pixels 431 may be stored at the native resolution of the original image 410, while the pixels 441 representing the immediate surroundings of the transport line network 411 may be subsampled, e.g., at a lower sampling rate, e.g., half the native resolution, and even more remote pixels 421 may be sampled at an even lower resolution, etc. Here, each different resolution may correspond to an associated subsampling layer that is sampled at the respective resolution from the original image 410. Finally, a much lower resolution base image 420 is added to the stack of subsampling layers.

[0075] Thus, multiple different solutions may be used to collectively represent the resampled version of the original image 410. The orientation of the pixel grid of the base image 420 may specifically correspond to the orientation of the pixel grid of the original image 410 (rather than being aligned to the GL path). The native resolution and subsampling window of each subsampling layer 420, 430, 440, as well as the total number of subsampling layers 420, 430, 440, comprise the parameters that control the amount of extraneous reduction performed by this compression process 400.

[0076] The new data structure effectively compresses the original data 410, which typically contains a large amount of irrelevant information with respect to the application focus that lies only in the region of the original data 410 that contains the GL, by successively decreasing the information density the further the data points are from these GL-centric regions of interest.

[0077] To reconstruct a single compressed adjusted multi-resolution image 460 from the various individual layers 420, 430, and 440, the low-resolution base image 420 is overlaid with all GL-focused subsampling layers 430 and 440 of increasing native resolution to reconstruct an intermediate image 450 that yields high levels of detail only in those regions of the GL and their immediate surroundings, while only coarse levels of detail are preserved in sparsely populated regions of the GL.

[0078] The desired compressed adjusted multi-resolution image 460 is then restored by combining and mapping the overlay pixel values ​​from each of the overlapping 420, 430, 440 layers of the intermediate image 450 onto the pixel grid of the image 460, thus achieving the adjustment as well. Specifically, the orientation of the pixel grid of the image 460 may be the same as that of the base image 420. Some areas of the intermediate image 450 may be covered by pixels from different layers 420, 430, or 440. In this case, an application-specific combination function is applied to determine a single pixel value for each pixel of the image 460, e.g., the median, average, or weighted average of the contributing pixels of the sub-sampling layers 420, 430, and 440.

[0079] This scenario allows very high compression rates and therefore makes storing originally large amounts of data feasible in storage-constrained systems, and enables transmission of such data over bandwidth-limited channels such as cellular networks, reducing the memory footprint of machine learning-centric algorithms and enabling their applications in memory-constrained systems. Particularly in geographical regions, e.g. rural areas with low density of GLS, of the transportation network, this allows very high compression rates and therefore makes it possible to store such images on devices or transmit them over cellular networks without losing important information about the transportation network.

[0080] The methods and processes described above may be particularly used in connection with machine learning or pattern recognition frameworks, an example of which is shown in Figure 5, which illustrates an embodiment of a model-based method 500 according to the present invention, the method notably having (i) a model creation stage for optimizing the parameterized model of method 100 of Figure 1, and (ii) a model application stage for applying the model as an instantiation of method 100 to actually perform the determination of evaluation data representative of a classification of one or more properties of or surrounding the GL according to one or more conditions of the GL.

[0081] Referring to the model creation stage of method 500, in a first step 505, an image is obtained and stored in a database or file system (the "imagery data base"). For example, as noted above, the image may resemble the images in image stack 200 of FIG.

[0082] In a further step 510, the GL geometry is obtained from an open or proprietary data source, for example electronic map data of the ROI of interest, in which the GL is located and stored in a database or equivalent system ("GL data base"). This step 510 may optionally include augmentation of GL features obtained by other sensory inputs or predictions, for example simulation results, related to the particular GL segment. For example, if the GL is a road, such other sensory inputs or predictions may relate to, but are not limited to, sensor data representing: (a) vibration, orientation, speed, acceleration, or any other vehicle parameters measured by one or more vehicle-based sensors that may provide insight into a particular line condition; (b) measurements from road weather sensors to measure temperature, precipitation, air pressure; (c) frequency of use of the line; (d) maintenance status or last maintenance interval of the line; (e) age and configuration of the surface; (f) climate zone, climate profile, expected temperature and precipitation; (g) susceptibility to unexpected events such as accidents, natural disasters, floods, landslides, avalanches, storms, tornadoes, hurricanes, typhoons, etc.; (h) electromagnetic fields, e.g., cellular network coverage (such data may be generated, for example, by a sensory assessment of network coverage, by use of test devices, e.g., in the field, or by simulation).

[0083] The term "feature" as used herein refers to a particular measurable state of a GL or a GL segment. As mentioned above, a non-limiting set of examples of various features, such as type of road surface, gradient of the GL (e.g., road or railroad track), etc., are shown in FIG.

[0084] In a further step 515, adjustment parameters for a subsequent image adjustment / normalization step 520 are defined (i) by receiving respective parameter data via an interface, e.g. a user interface, or (ii) by systematic evaluation. In the latter case (ii), a number of parameter sets are used to predict the characteristics of a particular target (i.e. the state of a particular type of GL), and the best performing parameter set is finally selected based on a comparison with known reference image data describing the correct information collected and / or verified individually for its features. Such a parameter set may include in particular parameters relating to the number of pixels, their size, arrangement and orientation. The adjustment step 520 may in particular comprise a step of generating a set of image patches stored in a file system or database ("Patch database") and applying the compression method 400 of FIG. 4 to at least one of the image layers of the image obtained in step 505.

[0085] The term "image patch" or "patch" for short, as used herein, refers specifically to a rectified image of a contiguous section of the relevant GL. Specifically, the result 115 of step S7 of FIG. 1 may be an image patch or a set of two or more separate image patches, each image page relating to a particular contiguous section of the GL considered. If in addition to image data, sensor data is also used to predict the target features, the concept of an "image patch" may be extended to become a "rectified GL datagram" that includes one or more layers of sensor data covering the same GL or GL segment in addition to one or more image patches. For simplicity, the following procedure is described with reference to image patches only, but the same concepts may be applied to rectified GL datagrams.

[0086] Depending on the desired use case, in step 525, the target features to be modeled or predicted are selected. Then, in a further step 530, pattern recognition methods are used to derive the model. A subset of the patches previously stored in the patch database is selected for training, and the results are verified on the retained subset. The model parameters describe the function that maps any image patch to the target features. The model parameters may form the parameters, for example, weights of an artificial neural network or regression values ​​in a regression analysis process that serve as the basis of the mathematical framework or model. Once the model parameters are determined in step 530, the established model that they represent is ready to be applied.

[0087] Referring now to the model application stage of the method 500, this stage may start by firstly defining a geographical region of interest (ROI) in step 540, obtaining a corresponding image in a further step 545, and finding data related to a particular desired GL or GL segment, e.g. a particular transport line segment, located within the ROI in a GL database. Thus, all information is available to apply an image rectification in a further step 555, using the same set of parameters ("rectification parameters") as in step 520 of the model creation stage. Then, in a further step 560, target features are predicted using the relevant acquired image patch or patches and the stored model parameters. Results in the form of respective evaluation data are output in step 565 for visualization or as input to any further processing steps, e.g. to derive control signals for controlling devices such as dynamic traffic signs or barriers based thereon.

[0088] The present invention may be used for a variety of different applications. By way of example and not limitation, the present invention may be used in conjunction with the following applications and related implementations of the present invention (see FIG. 3).

[0089] In one implementation, the weather profile of each GL or GL segment in a particular time frame is determined: (a) using live weather forecast imagery layers to model the amount of rain, snow, or wind expected in the next few minutes, hours, or days; (b) estimating the weather for a particular time of the year based on historical data, i.e., modeling seasonal weather phenomena along the GL, e.g., roads. Thus, to avoid dangerous situations, route planning and navigation obtain road segment specific weather risk information.

[0090] In another implementation, the maintenance requirements of a particular line segment can be evaluated: the spectral information of road surfaces such as asphalt or concrete has been shown to vary significantly with age and deterioration, meaning that maintenance actions can be triggered when certain thresholds are exceeded.

[0091] Yet another implementation takes into account the location and severity of disasters. In some countries, very detailed information on the locations of vehicle accidents is available. This allows assigning an accident risk level to each road segment or image patch and also allows predicting whether other areas are exposed to the same risk, for example, an alley with trees close to the road may have a higher risk of fatal accidents than a road without trees along the road. The same method can be applied in the case of natural disasters. Almost all disasters that result in human damage are known by local authorities. By utilizing meteorological information, elevation data, and disaster locations in the image stack, it becomes possible to predict other transportation lines that are prone to the same type of disaster. Thus, the current method can even support the planning of new transportation lines by assessing the risk of disasters along the intended route.

[0092] In another implementation, the method can be used to evaluate the expected amount of motion along or across a transportation line. Motion is defined as one or more vectors per patch and can represent motion information such as wind, traffic, or water flow.

[0093] At least one exemplary embodiment of the present invention has been described above, but it should be noted that there are many variations thereof. Furthermore, it should be understood that the exemplary embodiment described is merely a non-limiting example of how the present invention can be implemented, and is not intended to limit the scope, application, or configuration of the apparatus and methods described herein. Rather, the foregoing description provides one skilled in the art with an arrangement for implementing at least one exemplary embodiment of the present invention, and it should be understood that various changes in the function and arrangement of the elements of the exemplary embodiment can be made without departing from the subject matter defined by the appended claims and their legal equivalents. [Explanation of symbols]

[0094] 100 Method for determining a geographic line (GL) or one or more states around it 101 Drone 102 Aircraft 103 satellite 104 Earth 105 Regions of Interest (ROI) 106 Image Stacks Georeferenced reference geometry in 107 GL Merged image stack aligned to 108 GL reference geometry 109 Control Points 110 Continuous GL Segments 111 Subgrid 112 Image features of georeferenced source images from an image stack 113 grid cells 114 Aggregation Functions 115 Image adjusted using regular grid 116 Evaluation Data Merged image stack aligned to 200 GL reference geometry 210 RGB image layers 220 Elevation / Topology Image Layer 230 Vegetation Image Layer 240 Surface Type Image Layers 250 Evaporation Image Layers 260 Precipitation Forecast Image Layer 300 Exemplary Output of Possible Evaluation Results of the Method of FIG. 1 or FIG. 400 Compression Method 410 Original Image 411 GL Network 412 Snow-free ground 413 Forest 414 Snow-covered Ground 420 Low-res base image 421 image points for low-resolution base image Original high resolution image of the 430 GL 431 image points for high resolution base image 440 Reduced resolution 441 image points for low resolution image 450 intermediate multi-resolution image obtained by superimposing images 420, 430, and 440 460 Compressed and scaled multi-resolution images 500 A method for creating and applying a parameterized model of a method for determining one or more states at or around a geographic line (GL) 600 Subgrid 610 Regular 1D Subgrid 615 Cell of subgrid 610 620 Regular 2D Subgrids 625 Cell of subgrid 620 630 Anisotropic 1D Subgrid 635 Cell of subgrid 630 640 Non-gapless, equally spaced 1D subgrids 645 Cell of subgrid 640 650 Non-gapless, non-uniformly spaced 1D subgrids 655 Subgrid 650 Cells

Claims

1. A method (100) for determining one or more conditions at or around a geographic line (GL), the GL being a primarily one-dimensional geometric structure, either manmade or natural, on the surface of the Earth, and the conditions of the GL being physical, chemical, or biological conditions at one or more locations on the GL, the method (100) comprising: obtaining (S1, S2) georeferenced Earth observation data representing one or more spatially overlapping image layers (106) covering a particular region of interest (ROI) (105) on the Earth's surface from an aerial or space perspective, said ROI (105) including said GL, each image layer (106) representing imagery corresponding to an associated spectral band or data obtained from an associated sensor, said georeferenced Earth observation data being Earth observation data expressed in a geographic coordinate system; - geometrically transforming (S4 to S7) said georeferenced Earth observation data, at least in part, by a non-linear geometric transformation means into a local internal reference frame (115) of said GL within said ROI (105) to obtain a mapping of said georeferenced Earth observation data to respective corresponding coordinates in said local internal reference frame of said GL, said GL being in said local internal reference frame and represented therein as a line of a predetermined selected geometry; evaluating the mapped Earth observation data represented in the local internal reference frame of the GL according to a classification scheme to obtain evaluation data representative of a classification of one or more characteristics of the GL or its surroundings according to one or more conditions of the GL (S8), the classification scheme being a scheme for classifying one or more conditions of the GL according to a set of types, each condition being a detected or detectable condition of the GL at one or more locations of the GL; The method (100) comprises:

2. Geometrically transforming the georeferenced Earth observation data, at least in part, into a local internal reference frame (115) of the GL within the ROI (105), defining a plurality of control points within the ROI (105) at different positions along the GL; determining, for each control point, a respective local cell grid containing cells covering a neighborhood of that control point, the resulting set of local cell grids defining the local internal reference frame of the GL; geometrically transforming the georeferenced Earth observation data at least in part into the local internal reference frame of the GL to obtain a mapping of the georeferenced Earth observation data to the set of local cell grids of control points; The method (100) of claim 1, comprising:

3. 3. The method of claim 2, further comprising: concatenating individual cell grids of the control points to form a data representation of an adjusted overall cell grid that defines the local internal reference frame of the GL, the individual cell grids of the control points being arranged in an aligned and adjusted manner in a sequence that corresponds to a sequence of the control points along the GL.

4. The method (100) of claim 3, wherein the evaluation of each Earth observation data is performed based on a representation of the Earth observation data mapped to the adjusted global cell grid.

5. At least one of the local cell grids has the following characteristics: - anisotropic positioning of at least a subset of cells in said local cell grid; - the width or height or distance of the cells varies among at least a subset of the cells in the local cell grid; The method (100) according to any one of claims 2 to 4, wherein the method is defined to have one or more of the following characteristics:

6. the georeferenced Earth observation data represents one or more image point values ​​for each image point of at least one image layer, the image points being pixels of the image layer; Geometrically transforming the georeferenced Earth observation data includes:

6. The method (100) of claim 2, comprising determining at least one cell value for each cell of the local cell grid based on an aggregation of a subset of the image point values ​​corresponding to an area of ​​the ROI (105) covered by each cell of the local cell grid.

7. Aggregating said subset of said image point values ​​comprises the following operations: a) averaging the image point values ​​of all image points that lie entirely within a corresponding cell of the local cell grid; b) averaging the image point values ​​of all image points that overlap the area of ​​the ROI (105) covered by the corresponding cell of the local cell grid, each image point value being weighted according to the size of its overlap with the corresponding cell of the local cell grid; c) weighting the image point values, such that image points located closer to a center of a cell of the local cell grid get a higher weight than image points located further away from the center of the cell; d) determining a median, minimum and / or maximum image point value among the image point values ​​covered by the corresponding cells of the local cell grid; e) determining the image point value that represents a particular percentile of a distribution of the image point values ​​within the corresponding cell of the local cell grid; f) determining one or more of a range, a standard deviation, a variance, or any other higher order function that captures characteristics of the distribution of image point values ​​within cells of the local cell grid; g) determining a histogram representing the distribution of image point values ​​within cells of the local cell grid; The method (100) of claim 6, comprising one of the following operations:

8. At least one of the local cell grids is defined by a set of grid parameters; The method (100), The method (100) according to any one of claims 2 to 7, further comprising a step of determining at least one of said grid parameters based on a process including machine learning associated with cross-validating, per supervised learning, reference parameter data representative of a known correct classification of one or more relevant characteristics of said GL or its surroundings according to said one or more conditions, and intermediate results of said evaluation data obtained during a machine learning process.

9. Data layers (230 to 260) representing respective sensor measurements of one or more local conditions of the GL or its surroundings within the ROI (105) are provided, which are: -precipitation, - the intensity or wavelength of the local radiation, -temperature, -Atmospheric pressure, -Wind force or direction, - the speed and direction of water flow in at least one section of said ROI (105) when said GL is a geographical line on water; - in at least one section of the ROI (105), vibration, orientation, speed, acceleration, or any other vehicle parameter measured by one or more vehicle-based sensors that may provide insight into a particular state of a transportation line when the GL is a land geographic line accessible to land vehicles; - if the ground line is a transportation line, the frequency of use of the transportation line, the maintenance status of the transportation line or the time since the last maintenance interval, the age, or the composition of the transportation line surface; -Climate zones, climate profile, expected temperatures, and precipitation, - Susceptibility to unexpected events such as accidents, natural disasters, floods, landslides, avalanches, storms, tornadoes, hurricanes, typhoons, etc., as well as -Electromagnetic field The method (100) of claim 8, wherein the method (100) represents one or more of:

10. 10. The method (100) according to any one of claims 1 to 9, further comprising a step (S3) of aligning a respective geometry of the GL in the one or more image layers (210 to 260) represented by the Earth observation data with a georeferenced reference geometry (107) of the GL to form a combined image stack (200), wherein at least two image layers of the image layers (210 to 260) are incorporated into the image stack (200), each of the image layers (210 to 260) representing a respective digital image at a given spatial resolution, the spatial resolutions of at least two image layers of the image layers (210 to 260) differing from each other in at least one overlapping area, thus generating a combined multi-resolution image stack (106).

11. The method further comprises applying a data compression process (400) to reduce the spatial resolution of a given image layer (410), the data compression process (400) comprising:

11. The method (100) of claim 1, comprising generating the sublayers (420, 430, 440) by resampling at least one region of the image layer (410) at a spatial resolution respectively lower than the original spatial resolution, and maintaining or resampling the spatial resolution of at least one other region of the image layer (410), such that when two or more sublayers (420, 430, 440) of the image layer (410) are overlaid to jointly define a combined multi-resolution representation (450, 460) of the image layer (410), an average resolution of the combined multi-resolution representation is lower than the original resolution of the image layer (410).

12. 12. The method of claim 11, wherein the lower spatial resolution to which the at least one region of the image layer is resampled is a function of a distance of the region from the GL.

13. overlapping the sublayers (420, 430, 440); 13. The method (100) of claim 11 or 12, comprising aggregating values ​​of spatially overlapping image points of different sublayers (420, 430, 440) according to a predefined aggregation method to obtain, in the resulting combined multi-resolution representation (450, 460), image point values ​​of image points representing spatial regions corresponding to the overlapping image points of the different sublayers, wherein the image points are pixels of the image layers.

14. Overlapping and aggregating values ​​of the spatially overlapping image points of different sub-layers (420, 430, 440) 14. The method (100) of claim 13, comprising generating the combined multi-resolution representation (460) of the image layers such that an orientation of image points in the resulting representation is the same for all of these image points.

15. Based on the evaluation data, the following actions are taken: - determining a navigation or information output of a vehicle navigation system; and -Controlling dynamic traffic signs; Adjusting one or more configurable technical parameters of a vehicle or other transportation system; and - triggering a maintenance or inspection process on said GL, said maintenance or inspection process being a fully or partially automated process and / or being a process triggered by causing relevant signaling to notify maintenance personnel; - predicting future conditions of said GL or its surroundings; 15. The method (100) according to any one of claims 1 to 14, further comprising the step of performing one or more of the following operations:

16. A data processing system configured to perform the method (100) of any one of claims 1 to 15.

17. A computer program comprising instructions which, when executed on one or more processors of a data processing system according to claim 16, cause said data processing system to perform the method (100) of any one of claims 1 to 15.

Images