Method and system for detecting a structure embedded in an object under investigation

Abstract

The invention relates to a method for detecting a structure embedded in an object under investigation, wherein the method (1, 10) comprises the following steps: providing data of the object under investigation acquired by a high-frequency sensor, wherein the provided data comprise a three-dimensional volume data set of the object under investigation (2, 11); generating voxels based on the provided volume data set, wherein each voxel is assigned a voxel value (3, 12); generating one or more two-dimensional projections corresponding to an assigned projection direction, wherein for this purpose the voxel values ​​of consecutive voxel values ​​in the assigned projection direction are combined to form a pixel value of the respective two-dimensional projection; detecting the embedded structure by evaluating the pixel values ​​of the one or more two-dimensional projections (4, 13);and providing information about the embedded structure (5,14).;

Description

[0001] The present invention relates to a method for detecting a structure embedded in an object to be examined, and in particular to a method with which a structure embedded in an object to be examined can be detected simply and very precisely. Furthermore, the invention relates to a system suitable for this purpose and a computer program product suitable for this purpose.

[0002] For example, during civil engineering work, buried or embedded structures, such as cables or pipes, are repeatedly damaged. Such damage can not only lead to high repair costs but also to further property damage or even personal injury. Furthermore, damage to pipes or cables regularly impairs their functionality. In addition to detecting pipes and cables, it can also be important to detect other structures or anomalies in the subsurface, such as boulders or unexploded bombs. Consequently, there is a need for methods for the precise detection of such embedded structures.

[0003] One option for locating such structures is radar measuring devices, such as ground-penetrating radar. Ground-penetrating radar allows for the non-destructive exploration and characterization of subsurface or soil. Transmission signals are usually emitted into the ground via a transmitting antenna in the form of electromagnetic pulses or in the form of a frequency-modulated signal (in the case of a Frequency Modulated Continuous Wave (FMCW) radar). Part of the energy of the electromagnetic wave is reflected at interfaces between layers with different dielectric properties and received by a receiving antenna. In particular, the radar waves can penetrate the soil, with part of the corresponding energy being reflected at interfaces between different objects.

[0004] For example, by measuring the signal propagation times between the transmitting and receiving antennas, conclusions can then be drawn about the position and geometry of the objects. However, a problem with ground-penetrating radar is that the received signal captured by one or more receiving antennas exhibits a high noise component or a low signal-to-noise ratio due to the low power of the reflected signal. A general problem is direct ground reflection, as its signal power exceeds the power of the signal components of interest by several times. Ground irregularities and stones, vegetation, or other surface structures can generate further reflections and scattered components that overlap the signal components of interest.In addition, reflections from non-interesting structures in the ground, such as stones or the like, complicate signal analysis, as these unwanted reflections are superimposed on the reflections generated by the structures to be detected. Furthermore, such structures can cause shadowing.

[0005] From DE 10 2016 207 973 A1 a system for detecting objects in the ground is known, which system has at least one receiver for an acoustic, electrical, magnetic and / or electromagnetic signal emerging from the ground and a radio interface, wherein the receiver and the radio interface are transportable and are designed to be mounted on a construction machine.

[0006] Based on this prior art, the invention is based on the object of specifying an improved method for detecting a structure embedded in an object to be examined, as well as a device and a computer program product suitable for this purpose.

[0007] This problem is solved by the subject matter of the subordinate claims. Further advantageous developments are the subject matter of independent claims 1 and 8, respectively 15.

[0008] According to one embodiment of the invention, this object is achieved by a method for detecting a structure embedded in an object to be examined, the method comprising the following steps: • Providing data of the object to be examined acquired by a high-frequency sensor, wherein the provided data comprises a three-dimensional volume data set of the object to be examined; • Creating voxels based on the provided volume dataset, with each voxel being assigned a voxel value; the voxel value can be a complex value; • Creating one or more two-dimensional projections corresponding to an assigned projection direction, whereby for this purpose the voxel values ​​of voxel values ​​lying one after the other in the assigned projection direction are combined to form a pixel value of the relevant two-dimensional projection; • Detecting the embedded structure by evaluating the pixel values ​​of the one or more two-dimensional projections; and • Providing information about the embedded structure.

[0009] A sensor, which is also called a detector, (measured variable or measuring) transducer or (measured) probe, is a technical component that can detect certain physical or chemical properties and / or the material properties of its environment qualitatively or quantitatively as a measured variable.

[0010] A high-frequency sensor is a sensor that emits electromagnetic radiation (or electromagnetic waves), in particular electromagnetic radiation in the megahertz to terahertz range, and detects radiation reflected at interfaces between materials with different relative permittivity and conductivity. The high-frequency sensor can be designed, in particular, as a radar sensor, comprising at least one transmitter for the radar radiation and at least one receiver for the reflected radar radiation to be detected.

[0011] Furthermore, a three-dimensional volume data set is understood to be the sum of the detected (received) data of a three-dimensional recording or measurement of the object to be examined.

[0012] A structure or volume structure is understood to be a structure which has a certain volume, and in particular a structure (i.e. a certain geometry) which is composed of several voxels lying one behind the other as seen in a viewing direction. In particular, this can be an elongated structure, for example a tube, a vein or a cavity, which has a longitudinal direction, wherein a cross-section of the structure can vary in planes perpendicular to the longitudinal direction or longitudinal axis. The elongated structure can also represent a part or section of a larger structure, i.e. can be integrated into a larger structure.

[0013] An object to be examined is understood to be the volume in which the embedded structure is suspected, for example a floor or subsurface or a human or animal body or a partial volume thereof.

[0014] The projection direction is understood to be any direction from which the object to be examined, for example a specific three-dimensional section of a floor or subsurface, can be "viewed" for the signal evaluation according to the invention. In this case, a two-dimensional image of the object to be examined is generated in a fictitious projection plane perpendicular to the viewing direction. Figuratively speaking, a projection of each voxel in a viewing direction referred to as the projection direction is generated by imaging the voxel in a projection plane which, viewed in the viewing direction, is located behind the object to be examined. In this way, a two-dimensional image can be generated from the three-dimensional object to be examined divided into voxels, wherein the two-dimensional image is divided into a plurality of pixels with a specific pixel area and each pixel is assigned a pixel value.The pixel values ​​are determined from the voxel values ​​of those voxels that are located one behind the other in the respective projection direction.

[0015] A voxel is defined as a three-dimensional spatial region of a given size and structure or geometry, with each voxel being described by at least one associated voxel value (in this case, the assigned signal value of the received signal), its affiliation with a modeled object (in this case, the object to be examined), and its position within the modeled object. A voxel can, in particular, have the geometry of a cube, with each cube being assigned a voxel value at a grid point (i.e., in the geometric center of the cube).

[0016] In this case, a voxel value is a data value that describes the detected signal value corresponding to the signal component reflected within the respective voxel volume.

[0017] For example, a pixel of a two-dimensional projection may have a square area corresponding to the surface area of ​​one side of a cube-shaped voxel. Combining the voxel values ​​into a single pixel value can be achieved by simply summing them.

[0018] The evaluation of the two-dimensional projections generated according to the invention can be carried out by known methods of two-dimensional image evaluation.

[0019] Common high-frequency sensors, such as ground-penetrating radar, often provide data that can be interpreted as voxels with little or no further processing. Thus, the data acquired for an object under investigation can be stored as voxel values, with the voxel coordinates determined by the position of the corresponding voxel values ​​in a three-dimensional field.

[0020] By creating a two-dimensional projection, i.e., combining voxels or voxel values, the influence of outliers and other disturbances, such as noise caused by reflections from stones in the ground, can be reduced. Consequently, a structure embedded in an object under investigation can be easily and very precisely detected in a two-dimensional projection.

[0021] Overall, a method for detecting a structure embedded in an object to be examined is thus provided which is improved compared to the prior art explained above.

[0022] The step of evaluating the generated two-dimensional projections or the associated pixel values ​​can in particular comprise determining and evaluating the contrast in the corresponding two-dimensional image.

[0023] Contrast refers to a difference in the pixel values ​​in the two-dimensional projection, where the pixel values ​​can be interpreted as brightness values. For visual inspection, an image of the two-dimensional projection can also be displayed on any output device, for example, on a screen, with the corresponding intensity values ​​of the individual pixels (in a specified color or with a color code).

[0024] For example, high signal values ​​can be represented with higher brightnesses and lower signal values ​​with lower brightnesses. In this case, the bright parts of a two-dimensional projection image represent the surface of a structure embedded in an object under investigation and possibly also the volume of the embedded structure if the material of the structure generates higher reflections within a volume part, whereas the surroundings of the structure are represented by the darker parts of the two-dimensional projection image. Due to the resulting increased contrast in the two-dimensional projection compared to the three-dimensional voxel volume, a structure embedded in the object under investigation can be detected easily and precisely.

[0025] In the case of an elongated structure within an object under investigation, the contrast in a projection will be greatest when the projection direction coincides with the longitudinal extension of the elongated structure. In this case, most voxels are located one behind the other in the projection direction, so that summing the corresponding voxel values ​​results in a correspondingly high pixel value for a pixel assigned to the respective voxels.

[0026] If, on the other hand, a projection direction runs perpendicular to the longitudinal direction of the embedded structure, a correspondingly smaller number of voxels, seen in projection directions, lie one behind the other, so that the pixel value of the respectively assigned pixel generated by summarizing the associated voxel values ​​is correspondingly lower.

[0027] According to one embodiment of the invention, for generating each two-dimensional projection, i.e., each corresponding two-dimensional image, the voxel values ​​of those voxels whose projections overlap in a projection plane perpendicular to the relevant projection direction can be combined. This means that the projection of a center point of a corresponding voxel in the projection plane is spaced by a maximum of half a pixel width from a straight line passing through the (geometric) center of a pixel in the projection direction. The pixel size can, in principle, be freely determined. However, it can, in particular, correspond to the size of an area of ​​one side of a cube-shaped voxel, so that the pixels have a square area.

[0028] According to one embodiment, combining the voxel values ​​into one pixel value each comprises a normalization.

[0029] Normalization means converting the corresponding voxel values ​​into a standardized form, for example, by summing the voxel values ​​to be combined and then dividing them by the number of combined voxel values. This further reduces the influence of a different number of voxels or voxel values, which are combined into a single pixel value and represent values ​​caused solely by disturbances, particularly noise.

[0030] Furthermore, when voxel values ​​are combined to form a pixel value, the voxel values ​​can also be weighted. In this case, weighting means taking individual voxel values ​​into account based on their significance for the respective pixel. In particular, the proximity of a voxel center to the straight line that runs through the center of the pixel in question and parallel to the projection direction can be taken into account. For example, weighting (e.g. multiplying the relevant voxel value by a weighting factor) can be carried out with a maximum value (e.g. one) if the voxel center lies on the straight line, and weighting with a minimum value (e.g. zero) if the voxel center is half a pixel width (assuming a square pixel area) from the straight line. A predefined course, e.g. linear or quadratic, of the weighting (ora weighting factor).

[0031] This allows the evaluation of the generated voxels to be further optimized and, in turn, the influence of voxels that are a large distance from the straight line running in the projection direction to be reduced.

[0032] In one embodiment, the step of generating one or more two-dimensional projections (13) comprises the following steps: • Specify an initial projection direction; • Creating an initial two-dimensional projection by summarizing voxel values ​​of the voxels along the initial projection direction into pixel values; • Generating one or more further projection directions by rotating the initial projection direction by a predetermined angle about a rotation axis perpendicular to the initial projection direction; and • Creating a further two-dimensional projection for each further projection direction by combining voxel values ​​of the voxels along the respective further projection direction into pixel values.

[0033] The various projection directions thus generated can be selected such that adjacent projection directions have a constant angular separation, for example, 10°, 5°, or 1°. In this way, a two-dimensional projection can be generated for each of the projection directions and evaluated for the detection of an embedded structure.

[0034] This allows the process to be further optimized, especially since the embedded structure, depending on the design of the structure or the arrangement of the individual volume components, can only be detected at all or with sufficient precision from certain angles. Furthermore, the orientation of the embedded structure can also be determined, especially if the structure is elongated.

[0035] The evaluation of the two-dimensional projections generated in this way (or the corresponding two-dimensional pixel values) can be carried out, for example, by using a Fourier transformation and / or a Hough transformation.

[0036] The Fourier transform is a mathematical method from the field of Fourier analysis that decomposes aperiodic signals into a continuous spectrum. Using the Fourier transform, specific regions in the corresponding images, such as areas of comparatively high brightness, can be searched for or identified.

[0037] The Hough transform is a robust global method for detecting lines, circles, or any other parameterizable geometric figures in a binary gradient image, for example a black and white image (or an image with two predetermined colors), after edge detection.

[0038] Any other pattern recognition methods can also be used. Machine learning methods can also be used for this purpose.

[0039] Thus, based on known and common methods, conclusions can be drawn about the geometry and / or orientation of the embedded structure without the need for complex and resource-intensive calculations.

[0040] The possibility described above of creating a plurality of two-dimensional projections, where all projection directions are perpendicular to a rotation axis, is suitable, for example, for detecting pipes in the subsurface if it can be assumed that a pipe to be detected runs horizontally, i.e. parallel to the surface. Since a pipe represents an elongated structure and can therefore be best detected in a two-dimensional projection if the projection direction is aligned with the pipe axis, in this case detection can be restricted to projection directions that are also horizontal or parallel to the surface of the subsurface. In other words, if the longitudinal direction of the structure to be detected is known to lie in a certain plane (orparallel to it), the rotation direction for the projection directions can be chosen perpendicular to this plane.

[0041] It is, of course, fundamentally possible to select a multitude of arbitrary projection directions and thus determine the geometry and position of an embedded structure using the corresponding two-dimensional projections. For example, two mutually perpendicular rotation axes can be defined and projection directions can be determined for any pair of rotation angles φ, θ. In this case, a fixed angular distance from adjacent projection directions (for a rotation around one of the rotation axes) of, for example, 10°, 5°, or 1° can be selected.

[0042] In general, when detecting an embedded structure, each of the two-dimensional projections (or the associated pixel values) can be evaluated using appropriate image processing or pattern recognition methods. In practice, with data or voxels or voxel values ​​generated by ground-penetrating radar, due to the low signal-to-noise ratio, even in the two-dimensional projections, only those embedded structures can be detected that have a sufficient extent in the respective projection direction. The greater this extent, the greater the contrast improvement that can be achieved with the method, since combining (in particular adding up) the voxel values ​​creates a correspondingly large distance (a sufficiently large difference) to those pixel values ​​that result from combining voxel values ​​that exclusively have a noise component ornoise component and no signal component (i.e., no component caused by an embedded structure to be detected). In this case, the longitudinal extension direction of an elongated embedded structure can be determined from the angle of the projection direction associated with the two-dimensional projection that represents the embedded structure with the best contrast.

[0043] In principle, in this case, it is sufficient to generate projections in a given coordinate system in an angular range for the projection direction from 0° (starting from an initial angular position) to 180° around the rotation axis, since the pixel values ​​of two-dimensional projections that have an angular distance of 180° are identical.

[0044] As already mentioned above, the high-frequency sensor can also be a ground-penetrating radar. Ground-penetrating radar is used in practice, particularly for detecting pipes and cables, or in archaeology for detecting archaeological finds underground or in objects such as pyramids. For example, it can be used to search for underground wall sections or objects in pyramids.

[0045] It should be emphasized again at this point that the method can, of course, detect structures of any shape. In the case of elongated structures, their orientation can also be easily determined.

[0046] In general, the high-frequency sensor can be any high-frequency sensor, such as those used in a computer tomography scanner, magnetic resonance imaging scanner, or an X-ray inspection device. Even with such an application of a high-frequency sensor, the method according to the invention can be used to detect structures embedded in an object to be examined. In this case, it is fundamentally irrelevant whether the high-frequency sensor generates electromagnetic radiation that penetrates the object to be examined and detects the radiation that has passed through the object to be examined, or whether the high-frequency sensor, like radar, operates according to the reflection principle. However, the method according to the invention will achieve a greater effect when evaluating data that was acquired using the reflection principle.become more important in such cases, since methods that work according to the transmission principle usually have a better signal-to-noise ratio.

[0047] A system according to the invention for detecting a structure embedded in an object to be examined comprises: • a first provision unit which is designed to provide data of the object to be examined which are acquired by a high-frequency sensor, wherein the provided data comprise a three-dimensional volume data set of the object to be examined; • a generation unit configured to generate voxels based on the provided volume data set, wherein each voxel is assigned a voxel value; • an evaluation unit which is designed to • to generate one or more two-dimensional projections corresponding to an assigned projection direction, whereby for this purpose the voxel values ​​of consecutive voxel values ​​in the assigned projection direction are combined to form a pixel value of the respective two-dimensional projection; and • to detect the embedded structure by evaluating the pixel values ​​of the one or more two-dimensional projections; and • a second provision unit which is designed to provide information about the recognized structure.

[0048] The method described above can be carried out using such a device.

[0049] The evaluation unit can be designed in such a way that, for the generation of each two-dimensional projection, the voxel values ​​of those voxels are combined whose projections overlap in a projection plane perpendicular to the relevant projection direction.

[0050] In one embodiment, the evaluation unit is designed to carry out the normalization described above, ie when combining the voxel values ​​to form a pixel value, to carry out a normalization to the number of voxel values ​​respectively combined.

[0051] Furthermore, the evaluation unit can also be designed to carry out the weighting of the voxel values ​​described above when combining the voxel values ​​into a pixel value.

[0052] In one embodiment, the evaluation unit comprises: • a setting unit configured to set an initial (virtual) projection direction; • a rotation unit designed to generate one or more additional projection directions by rotating the initial projection direction by a predetermined angle around a rotation axis perpendicular to the initial projection direction. This unit can, of course, also be implemented (virtually) in the form of software. • a summarization unit, which is designed to generate an initial two-dimensional projection of one or more further two-dimensional projections by summarizing voxel values ​​of the voxels along the respective projection direction into pixel values. This unit can, of course, also be implemented (virtually) in the form of software.

[0053] In addition, the evaluation unit can be configured to apply a Fourier transform and / or a Hough transform to evaluate the generated pixel values ​​of a two-dimensional projection. Thus, the evaluation unit can apply known and common methods to draw conclusions about the geometry and orientation of the embedded structure to be detected, without the need for complex and resource-intensive adaptations.

[0054] The invention further provides a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out a method described above for detecting structures embedded in an object to be examined.

[0055] Such a computer program product is designed to be suitable for implementing the method according to the invention. The computer program product can be designed such that the step of providing the data acquired by a high-frequency sensor merely consists of supplying existing voxel values ​​of a volume data set to the further method steps. However, the computer program product can also be designed such that this step includes processing raw data supplied by a high-frequency sensor.

[0056] In summary, it can be stated that the present invention provides a method for detecting a structure embedded in an object to be examined, a system suitable for this purpose and a computer program product suitable for this purpose, wherein a structure embedded in an object to be examined can be detected in a simple manner and very precisely.

[0057] The invention is explained in more detail below with reference to the accompanying figures. They show: Fig. 1 is a flowchart of a method for detecting a structure embedded in an object to be examined according to a first embodiment of the invention; Fig. 2 is a flowchart of a method for detecting a structure embedded in an object to be examined according to a second embodiment of the invention; Fig. 3 is a schematic block diagram of a system for detecting a structure embedded in an object to be examined according to embodiments of the invention; Fig. 4 schematic representations of a simulated voxel volume to be evaluated, which contains an embedded structure in the form of a tube, where Fig. 4a shows the voxel volume with voxel values ​​that do not contain any noise components, and Fig. 4b the voxel volume with voxel values ​​that contain significant noise; and Fig. 5 a schematic representation of two-dimensional projections of the voxel volume according to Fig. 4 in different projection directions parallel to the longitudinal axis of the tube, where Fig. 5a the pixel values ​​of projections of the voxel values ​​according to Fig. 4a and Fig. 5b the pixel values ​​of projections of the voxel values ​​according to Fig. 4b.

[0058] Fig. 1 shows a flowchart of a method for detecting a structure 1 embedded in an object to be examined according to a first embodiment of the invention.

[0059] For example, during civil engineering work, damage to buried or embedded structures, such as cables or pipes or other objects embedded in the ground, often occurs.

[0060] One option for locating such structures is radar measuring devices, such as ground-penetrating radars. Ground-penetrating radars allow for non-destructive exploration and characterization of subsurface or soil. However, as already mentioned above, a problem with ground-penetrating radar is that the received signal detected by one or more receiving antennas exhibits a high noise component or a low signal-to-noise ratio due to the low power of the reflected signal. Furthermore, reflections from structures of no interest, such as stones on the ground surface or in the soil, or the like, complicate signal analysis, as these unwanted reflections are superimposed on the reflections generated by the structures to be detected.

[0061] Fig. 1 shows a simplified flow diagram 1 of a method for evaluating data generated by a high-frequency sensor, e.g., ground-penetrating radar. In a first step 2, data on the object to be examined acquired by the high-frequency sensor is provided, the provided data comprising a three-dimensional volume data set of the object to be examined. In a subsequent step 3, voxels are generated based on the provided volume data set, each voxel being assigned a voxel value. In a subsequent step 4, one or more two-dimensional projections are generated from the voxels thus generated or the associated voxel values, each two-dimensional projection being assigned a predetermined or to-be-determined projection direction.Each two-dimensional projection is created by combining voxel values ​​that lie one after the other in the projection direction to form a pixel value of the two-dimensional projection. In a further step 5, the two-dimensional projections are evaluated using the pixel values ​​of the two-dimensional projections. The evaluation can be performed using known image processing methods. In particular, the contrast in the two-dimensional image of the projection in question can be evaluated. In a final step 6, information is provided about any detected structure embedded in the examined voxel volume. This information can, for example, contain information about the type of detected (recognized) embedded structure.For example, using appropriate known pattern recognition methods used for this purpose, it can be determined that the detected embedded structure is a pipe, a cable, an unexploded bomb or the like.

[0062] Two-dimensional projections (the term "projection" describes, on the one hand, the process of projecting the voxel volume onto a plane perpendicular to the projection direction and, on the other hand, a two-dimensional image, particularly a pixel image, which is created by the projection) offer the advantage of producing two-dimensional images that exhibit a significantly higher contrast than the three-dimensional voxel values ​​or the three-dimensional voxel image represented thereby. However, when the voxel volume is projected onto a two-dimensional image plane, the information in the third dimension (in the direction of projection) is naturally lost. This disadvantage, however, is offset by using multiple projections with different projection directions.

[0063] Since, as already mentioned above, known high-frequency sensors, such as ground-penetrating radar, often provide data that can be interpreted as voxels without any further processing, or at least without significant further processing, creating one or more two-dimensional projections requires only minimal computational effort. A voxel is characterized in a three-dimensional data field by its position within the data field, with the position in the data field corresponding to the position of the voxel in the respective voxel volume. The voxel value assigned to a voxel can be interpreted and displayed as a brightness value, for example, on a monitor or a printout.

[0064] Creating and using two-dimensional projections offers the advantage of reducing the influence of outliers in the voxel values ​​and the influence of other disturbances, such as noise, in the two-dimensional pixel image representing a two-dimensional projection.

[0065] To generate each two-dimensional projection, the voxel values ​​of those voxels that lie one behind the other in the projection direction along a straight line parallel to the voxel are combined; in other words, those whose projections overlap in a projection plane perpendicular to the relevant projection direction. The corresponding pixel values ​​can be generated such that the projection of a center point of a corresponding voxel in the projection plane is spaced by a maximum of half a pixel width from a straight line passing through the (geometric) center of a pixel in the projection direction. The pixel size can generally be freely defined. However, it can, in particular, correspond to the size of an area of ​​one side of a cube-shaped voxel, so that the pixels have a square area.

[0066] The relevant voxel values ​​can be combined into a single pixel value by summing the pixel values. To eliminate the influence of different numbers of summed voxel values, normalization to the number of summed voxel values ​​can be performed.

[0067] As already mentioned at the beginning, when voxel values ​​are combined to form a pixel value, the voxel values ​​can also be weighted. This weighting can be carried out taking into account the distance of the respective center of a voxel from the relevant straight line that is perpendicular to the center of the relevant pixel (i.e. runs in the relevant projection direction). A corresponding weighting factor can be defined which is greater the closer the center of the relevant voxel is to the straight line. If the center lies on the straight line, the value can be 1, for example, and if the center is at a distance of half a pixel width, the weighting value can be 0, for example. Between these values, a linear or quadratic (i.e. parabolic) curve of a dependence of the weighting factor on the distance from the straight line can be used.

[0068] According to the Fig. In the embodiment illustrated in Figure 1, step 5 of evaluating the generated pixel values ​​of each projection may comprise determining a contrast. To further increase the contrast in an image of a two-dimensional projection displayed, for example, on a monitor or printout, a threshold value may be defined, above which a pixel value exceeding the threshold value is assigned a maximum pixel value and otherwise a minimum pixel value. A maximum pixel value may represent maximum brightness, and a minimum pixel value may represent minimum brightness. This may, for example, generate a black-and-white image in which the embedded structure is displayed at maximum brightness.

[0069] As already mentioned above, the evaluation of the pixel image representing a two-dimensional projection can be done using the Fourier transform or Hough transform.

[0070] By applying a Fourier transformation or a Hough transformation, it is possible to search specifically for geometric structures in the corresponding image data, such as ellipses or circles.

[0071] Based on detected embedded structures or the corresponding information provided, corresponding controllable systems or robot systems, for example civil engineering machines, can then be controlled in such a way that damage to the detected embedded structure, for example a pipe, is avoided.

[0072] Fig. 2 shows a flowchart of a method 10 for detecting a structure embedded in an object to be examined according to a second embodiment of the invention.

[0073] Procedure 10 according to Fig. 2 differs from that in Fig. 1 in that step 4 of generating one or more two-dimensional projections is replaced by steps 15, 16, 17, 18, which elaborate step 4 in more detail.

[0074] In step 15, an initial projection direction is determined. This initial projection direction can, for example, be parallel to a horizontal plane, e.g., parallel to the surface of a subsurface if the object to be examined lies in the subsurface in question (or is part of it). Within this plane, the projection direction can be chosen arbitrarily.

[0075] In step 16, using this initial projection direction, a two-dimensional projection of the provided voxels is generated using the relevant voxel values.

[0076] In step 17, one or more additional projection directions are generated from the initial projection direction by rotation around a rotation axis perpendicular to it. The additional projection directions can be generated such that adjacent projection directions each have a fixed angular separation, for example, 10°, 5°, or 1°. Using these additional projection directions, additional two-dimensional projections are generated.

[0077] Of course, it is also possible to initially only subdivide the projection directions relatively roughly, for example at a fixed angular interval of 45°. If it is then determined in the further step 5, in which an embedded structure is detected, that this structure has a longitudinal extension, and the direction of the longitudinal extension is to be determined more precisely, then further projection directions with a correspondingly fine angular interval of, for example, 5° or 1° can be defined between the two adjacent projection directions in question at this rough angular interval, and corresponding projections can be created. From these projections, it can then be determined by evaluating the contrast in particular in which projection direction the longitudinal axis of the elongated embedded structure runs. In this case, the projection can be sought in which the highest contrast of a certain (i.e. detected) structure is greatest.If it is determined that the contrast initially increases and then decreases again with decreasing or increasing angle of the projection direction, it is also possible to interpolate between the two corresponding angles (e.g. linearly) in order to determine the maximum of the contrast and assign it to a corresponding projection angle, which is then interpreted as the angle of the longitudinal axis of the detected embedded structure.

[0078] If the projection angle is varied around two mutually perpendicular rotation axes, such a contrast maximum can also be determined by determining a maximum of the two-dimensional dependence of the contrast on the two rotation angles defining the projection direction using known mathematical methods.

[0079] According to another variant, only a single or a predefined number of projection directions can initially be defined, a two-dimensional projection created for each of these, and this two-dimensional projection evaluated with respect to an embedded structure (if necessary, taking into account previously created and evaluated projections). One or more additional projection directions can then be created and associated projections generated until the evaluation of the respective projection shows that the respective projection direction runs in the direction of the longitudinal axis of a detected elongated structure. In other words, the projection direction is varied until it runs parallel to the longitudinal axis of the elongated structure.This procedure also allows the angle of rotation to be determined during each step (creating and evaluating one or more projections).

[0080] The initial projection direction can also be determined or selected based on basic knowledge of the embedded structure in order to determine the orientation of the embedded structure as quickly as possible and with comparatively low resource consumption. For example, if it is known that the provided voxel volume characterizes a portion of the subsurface of a road, and the course of the road within the voxel volume is known, and it can further be assumed that a pipe is to be located in the longitudinal direction of the road, the initial projection direction can be selected in the longitudinal direction of the road.

[0081] Fig. 3 shows a schematic block diagram of a system 20 for detecting a structure embedded in an object to be examined according to embodiments of the invention.

[0082] The system 20 comprises a first provision unit 21 configured to provide data of the object to be examined acquired by a high-frequency sensor, wherein the provided data comprises a three-dimensional volume data set of the object to be examined. The provision unit 21 can be configured, for example, as a ground-penetrating radar. Furthermore, the system 20 comprises a generation unit 22 configured to generate voxels based on the provided volume data set, wherein voxel values ​​are assigned to each voxel.

[0083] An evaluation unit 23 of the system 20 is configured to perform the above-explained evaluation of the generated voxels representing the object to be examined, including the recognition or detection of any embedded structure present. A second provision unit 24 of the system 20 is configured to provide information about the recognized embedded structure. This provision unit 24 can, for example, further specify or classify a detected embedded structure. For example, the provision unit 24 can be configured to be able to recognize the type of the detected embedded structure, whether it is, for example, a pipe, a line, or another object.In addition, the provision unit 24 can provide further information, for example the outer diameter and / or the inner diameter and / or the wall thickness of a pipe.

[0084] The first provision unit 21 can also be a receiver configured to receive corresponding sensor data, for example, from a ground-penetrating radar. The second provision unit 24 can also comprise a transmitter capable of transmitting corresponding data wirelessly or via a wired connection to a higher-level unit. The first provision unit 21 and the second provision unit 24 can also be integrated into a common transceiver.

[0085] The generation unit 22 and the evaluation unit 23 can each be implemented as code stored in a memory and executable by a processor.

[0086] How Fig. 3 further shows, the evaluation unit 23 has a determination unit 25 which is designed to determine an initial projection direction, a summarizing unit 26 which is designed to generate pixel values ​​of a two-dimensional projection (for each specific projection direction) from the voxel values ​​of the voxels, and a rotation unit 27 which is designed to rotate the initial projection direction about a rotation axis running perpendicular to the initial projection direction.

[0087] As explained above, the rotation unit 27 can also be configured to generate any desired projection directions according to a predetermined procedure. In particular, the rotation unit 27 can generate further projection directions from an initial projection direction by rotating the initial projection direction by a predetermined angle θ about a first rotation axis and by a further predetermined angle φ about a second rotation axis perpendicular to the first rotation axis.

[0088] The determination unit 25 can also be implemented based on input units, via which corresponding basic or expert knowledge can be provided. For this purpose, the input units can be integrated into the determination unit 25 or connected to it wirelessly or via a cable. The summarization unit 26 and the rotation unit 27 can each be implemented based on a code stored in a memory and executable by a processor, i.e., as software.

[0089] Finally, the evaluation unit 23 has an image processing unit 28, which is designed to evaluate the two-dimensional projections for the presence of an embedded structure. The image processing unit 28 can apply a Fourier transformation or a Hough transformation for this purpose.

[0090] In the following, the method for detecting a structure embedded in an object to be examined and with reference to the Fig. 4 and Fig. 5 is explained in more detail.

[0091] Fig. 4a schematically shows a voxel volume 40 of 50 x 50 x 50 voxels in a Cartesian coordinate system, which represents the object to be examined. The voxel values ​​are represented as brightness values ​​on a standardized scale from 0 to 1. A black voxel means a voxel value of 0 and a white voxel means a voxel value of 0.1, whereby in this case the voxel values ​​can assume values ​​from a range of 0 to 1. In this respect, Fig. 4a represents an idealized case, since the voxel values ​​do not contain any noise or unwanted reflection components.

[0092] The bright voxels in Fig. 4a with a voxel value of 0.1 represents signal components of a ground-penetrating radar system originating from a pipe 42 located in a part of the subsurface represented by voxel volume 40. In the example shown, the longitudinal axis of the pipe runs parallel to the z-axis of the coordinate system. Within voxel volume 40, the bright voxels of pipe 50 are covered by the dark voxels, each with a voxel value of 0.

[0093] Fig. 4b shows the same simulated voxel volume as Fig. 4a, however, this simulation is much more realistic, since all voxel values, including the voxel values ​​containing the signal components generated by the tube 50, have (additional) noise components of a Gaussian noise with a value range from 0 to 1. Since the voxel volume in this case has the same geometric structure, but different voxel values ​​are assigned to the voxels, the voxel volume in Fig. 4b is provided with the reference number 41.

[0094] Fig. Figure 4b illustrates the poor signal-to-noise ratio of ground-penetrating radar data. The tube 50 is Fig. 4b is no longer visible to the human eye. Conventional signal analysis methods also fail here. Only computationally intensive correlation methods would be able to detect or visualize pipe 50 in such noisy data.

[0095] The procedure explained above, according to which the voxel values ​​are summarized in two-dimensional projections with specific projection directions, is based on the Fig. 5a and Fig. 5b illustrates this.

[0096] Fig. Figure 5a shows a matrix with 36 two-dimensional projections or corresponding pixel images, which are generated for the voxel volume 40 according to Fig. 4a. The projections were created with an initial projection direction of 0° (this runs in the z-direction, i.e. in the direction of the longitudinal axis of the tube 50) and a rotation of this initial projection direction around a rotation axis perpendicular to the longitudinal axis of the tube 50 with a constant angular distance of 10°. The projections are arranged row by row from top left to bottom right. The projection with a projection direction of 0° is thus top left in Fig. 5a and the projection with a projection direction of 350° is shown at the bottom right.

[0097] As from Fig. As can be seen in Figure 5a, the two-dimensional projections are, as expected, identical for projection directions with an angular separation of 180°. Thus, it is sufficient to cover an angular range from 0° to 180° to reliably detect an embedded structure.

[0098] The two-dimensional projections for projection directions of 0° and 180° show the circular cross-section of the tube 50. The projection with a projection direction of 90° (the fourth projection in the second row of Fig. 5a) images the pipe 50 from a direction perpendicular to its longitudinal axis. Since in this case, significantly fewer pixels within the pipe volume are consecutive when the voxel values ​​are combined into one pixel value, the contrast is significantly lower than in the case of projection for a projection direction of 0° or 180°.

[0099] Fig. 5b shows the two-dimensional projections created in an analogous manner for the voxel volume 41 according to Fig. 4b. This figure illustrates the advantage of the method according to the invention by forming two-dimensional projections. While for the idealized simulation data according to Fig. 4a in the projections according to Fig. 5a nor the pipe 50 is recognizable in practically all simulation directions, is using the noisy simulation data according to Fig. 4b the tube 50 is practically only visible in the two-dimensional projections with projection directions of 0° and 180°.

[0100] The method is therefore able to detect a structure embedded in the voxel volume 41 even with such noisy data.

[0101] As explained above, the method can also be implemented using certain prior knowledge or certain assumptions about embedded structures to be detected. For example, if it is assumed that only one pipe, if any, can be present in a voxel volume, then this knowledge can naturally also be used to determine the position of the pipe by identifying the projection direction in which the corresponding two-dimensional projection contains a circular structure. This procedure can of course also be used assuming the presence of other types of structures, for example, for the detection of metallic pipes or approximately cylindrical bodies, such as unexploded ordnance. This also applies to elongated structures with a rectangular cross-section, such as walls in the ground. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2016 207 973 A1

[0005]

Claims

[1] Method for detecting a structure embedded in an object to be examined, the method (1,10) comprising the following steps: (a) providing data of the object to be examined acquired by a high-frequency sensor, wherein the provided data comprises a three-dimensional volume data set of the object to be examined (2,11); (b) generating voxels based on the provided volume data set, each voxel being assigned a voxel value (3,12); (c) generating one or more two-dimensional projections corresponding to an assigned projection direction, wherein for this purpose the voxel values ​​of consecutive voxel values ​​in the assigned projection direction are combined to form a pixel value of the respective two-dimensional projection; (d) detecting the embedded structure by evaluating the pixel values ​​of the one or more two-dimensional projections (4, 13); and (e) Providing information about the embedded structure (5,14). [2] Method (1,10) according to claim 1, wherein for generating each two-dimensional projection, the voxel values ​​of those voxels are combined whose projections overlap in a projection plane perpendicular to the relevant projection direction. [3] Method (1,10) according to claim 1 or 2, wherein the combining of the voxel values ​​to form a pixel value comprises a normalization to the number of the respectively combined voxel values. [4] Method (1,10) according to one of claims 1 to 3, wherein summarizing the voxel values ​​comprises weighting the voxel values. [5] Method (1,10) according to one of claims 1 to 4, wherein the step of generating one or more two-dimensional projections (13) comprises the following steps: (a) determining an initial projection direction (15); (b) generating an initial two-dimensional projection by combining voxel values ​​of the voxels along the initial projection direction into pixel values ​​(16); (c) generating one or more further projection directions by rotating the initial projection direction by a predetermined angle about a rotation axis (19) perpendicular to the initial projection direction; and (d) generating a further two-dimensional projection for each further projection direction by combining voxel values ​​of the voxels along the respective further projection direction into pixel values ​​(17). [6] Method (1,10) according to one of claims 1 to 5; wherein the evaluation of the generated pixel values ​​for detecting the embedded structure is carried out using a Fourier transformation and / or a Hough transformation. [7] Method (1,10) according to one of claims 1 to 6, wherein the high-frequency sensor is a radar detection device, in particular a ground radar. [8] System for detecting a structure embedded in an object to be examined, the system (20) comprising: (a) a first provision unit (21) which is designed to provide data of the object to be examined which are acquired by a high-frequency sensor, wherein the provided data comprise a three-dimensional volume data set of the object to be examined; (b) a generation unit (22) which is designed to generate voxels based on the provided volume data set, wherein each voxel is assigned a voxel value; (c) an evaluation unit (23) which is designed to (i) to generate one or more two-dimensional projections corresponding to an assigned projection direction, whereby for this purpose the voxel values ​​of consecutive voxel values ​​in the assigned projection direction are combined to form a pixel value of the respective two-dimensional projection; and (ii) to detect the embedded structure by evaluating the pixel values ​​of the one or more two-dimensional projections (4,13); and (d) a second providing unit (24) which is designed to provide information about the recognized structure. [9] System (20) according to claim 8, wherein the evaluation unit (23) is designed such that, for generating each two-dimensional projection, the voxel values ​​of those voxels are combined whose projections at least partially overlap in a projection plane perpendicular to the relevant projection direction. [10] System (20) according to claim 8 or 9, wherein the evaluation unit (23) is designed to carry out a normalization to the number of respectively combined voxel values ​​when combining the voxel values ​​to form a pixel value. [11] System (20) according to one of claims 8 to 10, wherein the evaluation unit (23) is designed to weight the voxel values ​​when combining the voxel values ​​to form a pixel value. [12] System (20) according to one of claims 8 to 11, wherein the evaluation unit (23) comprises: (a) a setting unit (25) configured to set an initial projection direction; (b) a rotation unit (27) which is designed to generate one or more further projection directions by rotating the initial projection direction by a respectively predetermined angle about a rotation axis (19) running perpendicular to the initial projection direction. (c) a summarizing unit (26) which is designed to generate an initial two-dimensional projection of one or more further two-dimensional projections by summarizing voxel values ​​of the voxels along the respective projection direction to form pixel values. [13] System (20) according to one of claims 8 to 12, wherein the evaluation unit (23) is designed to apply a Fourier transformation and / or a Hough transformation to evaluate the pixel values ​​of a two-dimensional projection. [14] System (20) according to one of claims 8 to 13, wherein the high-frequency sensor is a radar detection device, in particular a ground radar. [15] A computer program product comprising instructions which, when executed by a computer, cause the computer to carry out a method for detecting a structure embedded in an object to be examined according to any one of claims 1 to 7.

Images