Device for detecting the wetness of tunnel lining, method for detecting the wetness, and program

Abstract

The system detects the moisture level of the tunnel lining surface with high precision while a moving object is in motion, and outputs the results in a format that is easy to use for inspection. [Solution] The wetness detection device 4 includes an image acquisition unit 5 that acquires a first reflection image 19 and a second reflection image 20 using short-wavelength infrared light in a first wavelength band 32 and a second wavelength band 33, a driving information acquisition unit 11 that includes a DMI 12, an IMU 13, a LiDAR 14 and a rangefinder 15, and a processing unit 16 and an output unit that associate the images with the lining surface coordinate system 21 using a pseudo-simultaneous mapping unit and a reprojection correction unit, calculate a wetness index 22, a wetness candidate area and priority confirmation order for each ring based on the dryness reference area, and output them as a lining development diagram 24, a three-dimensional point cloud 25 or an inspection result list 31.

Description

Technical Field

[0001] The present invention relates to an inspection technique used for the maintenance and management of tunnels.

[0002] More specifically, the present invention irradiates a tunnel lining surface with illumination light in a plurality of wavelength bands in the short-wavelength infrared region, obtains the reflected light from the lining surface as a two-dimensional image, and detects the wet state of the tunnel lining surface. The present invention relates to a wet state detection device, a wet state detection method, and a program for a tunnel lining.

[0003] Furthermore, the present invention associates reflection images corresponding to at least two wavelength bands, one with a relatively high water absorption wavelength band and the other with a relatively low water absorption wavelength band, with at least one of position information, attitude (orientation and inclination of a measuring device, camera, or moving body) information, and distance information obtained while moving, and outputs the reflection images to at least one of a lining development drawing or a three-dimensional point cloud, and relates to a technique for assisting the extraction of wet candidate locations and inspection determination.

Background Art

[0004] In road tunnels, railway tunnels, and other tunnel structures, it is required to continuously grasp cracks, water leakage, peeling, floating, efflorescence, and other abnormalities occurring on the lining surface and reflect them in maintenance and management. In particular, since water leakage or wetness may be related to deterioration of the lining surface, occurrence of efflorescence, frost damage, and other abnormalities, an inspection technique capable of efficiently grasping the wet state of the lining surface is demanded.

[0005] Conventionally, in the inspection of tunnel lining surfaces, visual inspection by inspectors, photography, impact sound inspection, etc. have been widely performed. However, these methods have a large human burden, it is difficult to homogenize inspection results, and there is a problem that work time and cost tend to increase when the tunnel extension is long.

[0006] For this reason, in recent years, technologies have been proposed that involve mounting imaging and lighting devices on vehicles to image the tunnel lining while driving through the tunnel, or using laser devices in combination to acquire three-dimensional shape or three-dimensional point cloud data of the tunnel lining. For example, technologies are known that involve measuring the three-dimensional shape of the surface of the tunnel lining to investigate the lining, and technologies that involve associating image data with three-dimensional point cloud data and using them to understand the condition of the lining concrete.

[0007] Furthermore, technologies have been proposed to generate unfolded images of the tunnel's inner surface based on acquired imaging images, to extract necessary images using the correspondence between the coordinates of a three-dimensional point cloud and the coordinates of an unfolded image, and to detect cracks, water leakage, etc., from images of the tunnel's inner surface for use in determining structural integrity or supporting inspections.

[0008] On the other hand, optical measurement techniques are also known that involve irradiating an object with light and acquiring the reflected or transmitted light in multiple wavelength bands to estimate the presence, location, or amount of specific components in the object. For example, regarding the detection of moisture, techniques are known that estimate the moisture content based on the difference or ratio of the reflection intensity between wavelength bands where water absorption is relatively high and wavelength bands where water absorption is relatively low, as well as techniques that estimate the location and amount of an object by spectral imaging using multiple spectral filters.

[0009] However, in the conventional technologies described above, technologies that primarily focus on acquiring visible images or three-dimensional shapes still have room for improvement in terms of reliably extracting the actual moisture state of the tunnel lining surface. Furthermore, general moisture detection or spectral imaging technologies may still be insufficient when imaging a curved tunnel lining surface while moving inside the tunnel, due to positional shifts, distance variations, and incident angle variations, as well as their application to inspection practices where the acquired results are mapped to a tunnel lining diagram or three-dimensional point cloud.

[0010] Therefore, there is a need for a technology that can irradiate the tunnel lining surface with illumination light in multiple short-wavelength infrared wavelength bands while moving inside the tunnel, associate the reflected image from the lining surface with the same or nearby lining surface areas, detect the wetness, and output the results to at least one of the lining development diagram and a three-dimensional point cloud for use in inspection and decision-making. [Prior art documents] [Patent Documents]

[0011] Prior art documents include, for example, the following patent documents. [Patent Document 1] Japanese Patent Publication No. 2016-31249 [Patent Document 2] WO2018 / 066086 publication [Patent Document 3] Patent No. 6584735 [Patent Document 4] Japanese Patent Publication No. 2023-047631 [Patent Document 5] U.S. Patent No. 4,463,261 [Patent Document 6] U.S. Patent No. 9551616 [Patent Document 7] WO2021 / 068746 publication [Patent Document 8] Chinese Patent Publication No. 105548205 [Patent Document 9] Chinese Patent Publication No. 109087291 [Patent Document 10] WO2021 / 049109 publication [Overview of the project] [Problems that the invention aims to solve]

[0012] Patent Document 1 discloses a tunnel lining surface inspection system and vehicle, which involve arranging multiple imaging means and multiple slit laser light projection means in a staggered pattern along the circumferential direction of an arc-shaped or substantially arc-shaped installation surface in order to acquire the three-dimensional shape of the tunnel lining surface. However, Patent Document 1 mainly deals with acquiring three-dimensional shapes using visible images and light sectioning, and does not explicitly mention, at least from its published content, the extraction of the wetting state of the lining surface based on short-wavelength infrared reflection images of two wavelengths: one with relatively high water absorption and the other with relatively low water absorption.

[0013] Patent Document 2 discloses a technology that involves mounting an imaging device, an illumination device, and a laser device on a vehicle, linking image data and point cloud data based on predetermined identification information, and further assigning an imaging count value as a count pulse, thereby improving the accuracy of a three-dimensional composite model and reducing the amount of work involved in associating the two sets of data. It also discloses a configuration for synchronously controlling the exposure timing of the imaging device and the lighting timing of the illumination device. However, Patent Document 2 focuses primarily on linking and combining image data and point cloud data, and does not explicitly state, at least from its published content, that it can acquire reflection images in multiple short-wavelength infrared wavelength bands while moving and calculate a wetness index by pseudo-simultaneously associating these with the same cross-section or nearby cross-sections.

[0014] Patent Document 3 discloses an image generation device that determines the coordinates on an unfolded image corresponding to the coordinates of selected three-dimensional points from three-dimensional point cloud data, and outputs an image of the range containing those coordinates from the unfolded image. In other words, Patent Document 3 is an excellent technology for associating and displaying three-dimensional point clouds with unfolded images. However, Patent Document 3 mainly focuses on the correspondence display and extraction process between unfolded images and point cloud data, and does not explicitly mention, at least from its published content, detecting the wetness of the lining surface itself based on two-wavelength short-wavelength infrared reflection images, and evaluating the detection results in relation to position information, attitude information, and distance information obtained during driving.

[0015] Patent Document 4 discloses an inspection support system that detects cracks, crack patterns, water leakage conditions, and reinforcement exposure conditions based on images of the inner surface of a tunnel, performs primary and secondary judgments based on these results to determine the degree of structural integrity, applies the results to divided regions of the inner surface, and projects the judgment results onto the inner surface of the tunnel. However, Patent Document 4 mainly focuses on visible image-based deformation detection and structural integrity determination, and does not explicitly state, at least from its published content, that it directly extracts the wet state from two-wavelength short-wavelength infrared reflection images utilizing the water absorption characteristics, or that it relatively evaluates the extracted wetness index based on distance correction, angle correction, and a dry reference region.

[0016] Patent Document 5 discloses an analytical device that utilizes the fact that, in the infrared reflectance spectrum, the change in reflectance due to increased moisture is small around 1300 nm, while the change in reflectance due to moisture is significant around 1450 nm. Therefore, Patent Document 5 presents the basic idea of ​​evaluating the moisture content by combining wavelength bands in which water absorption is large and wavelength bands in which the effect is relatively small. However, Patent Document 5 relates to a general analytical device, and it is not explicitly stated, at least from its published content, that it captures a curved lining surface as a two-dimensional image while moving inside a tunnel, associates the obtained images of multiple wavelength bands based on position and orientation, or outputs them to a lining development diagram or a three-dimensional point cloud.

[0017] Patent Document 6 discloses a spectral imaging system that remotely and non-invasively detects the presence, position, and amount of a target substance using a spectral filter array and an image acquisition array, and it is shown that freeze motion images can be easily obtained for a moving target by simultaneous image acquisition. Therefore, Patent Document 6 shows useful findings regarding the detection of a target substance using multiple wavelength images and the imaging of a moving target. However, Patent Document 6 relates to a general spectral imaging system, and regarding evaluating the wet state in association with a lining development drawing or a three-dimensional point cloud while considering the position variation, posture variation, and distance variation during travel for a curved inner surface such as a tunnel lining surface, it is not explicit at least from its disclosed content.

[0018] Patent Document 7 discloses an image acquisition device, system, and method for tunnel inspection that arranges a plurality of CCD cameras and a plurality of auxiliary light sources on the same annular ring, widely photographs a tunnel lining surface in a positional relationship corresponding to the tunnel center, and realizes a synchronous encoder, synchronous imaging of images, damage identification, and real-time splicing. However, Patent Document 7 mainly relates to the wide-range imaging and positioning of the inner peripheral surface of a tunnel by a visible imaging system, and regarding obtaining two-wavelength short-wavelength infrared reflection images suitable for extracting the wet state and correcting the positional deviation therebetween to obtain a wet index, it is not explicit at least from its disclosed content.

[0019] Patent Document 8 discloses a positioning method and system that provide a plurality of mileage markers in an image acquisition area of a tunnel surface and determine the position of a tunnel surface defect based on the mileage markers included in continuously acquired images. Therefore, Patent Document 8 is useful in suppressing the defect positioning error caused by the error or drift of the travel distance. However, Patent Document 8 is an image-based position correction technology premised on mileage markers on a tunnel wall surface, and regarding re-projecting multiple wavelength images of short-wavelength infrared onto the same cross-section in combination with travel information such as DMI, IMU, or LiDAR and evaluating the wet state, it is not explicit at least from its disclosed content.

[0020] Patent Document 9 discloses a technique for recognizing position marking points using a marked image obtained from a tunnel design drawing and a three-dimensional panoramic image inside the tunnel acquired by an imaging device, constructing a tunnel position coordinate system and a position information library, and using the same for subsequent defect positioning. Therefore, Patent Document 9 is useful in that it organizes position information for each area inside the tunnel and provides a position information infrastructure that contributes to defect inspection and maintenance. However, Patent Document 9 mainly focuses on the position information library and defect positioning, and at least from its disclosed content, it is not explicit about two-wavelength short-wavelength infrared reflection imaging for extracting the wet state, distance correction and angle correction for the imaging result, and calculation of the wet index.

[0021] Patent Document 10 discloses a moisture detection device that uses a reference wavelength, absorption wavelength 1, and absorption wavelength 2, drives these light sources in a time-division manner, and determines the presence or absence or state of water, ice, or snow based on the detection signal ratio of the absorption wavelength with respect to the reference wavelength. It also discloses a configuration for suppressing a decrease in light reception efficiency by setting an oblique incidence arrangement and a polarization direction. However, Patent Document 10 is a moisture detection device for deposits on a road surface, and for obtaining a two-dimensional reflection image while traveling, re-projecting the obtained two-wavelength image onto a lining surface coordinate system to calculate a wet index, and further outputting it to a lining development drawing or a three-dimensional point cloud for a curved inner surface such as a tunnel lining surface, at least from its disclosed content, it is not explicit.

[0022] Thus, Patent Documents 1-4 and 7-9 primarily concern imaging of tunnel lining surfaces, acquisition of three-dimensional shapes or point clouds, positioning, correspondence with unfolded images, or visible image-based deformation determination and inspection support. On the other hand, Patent Documents 5, 6, and 10 primarily concern the detection of moisture or target substances using two-wavelength or multi-wavelength optical information. Therefore, although useful knowledge is disclosed in each of these documents, there is still room for improvement in the technology of acquiring reflection images in at least two wavelength bands of short-wavelength infrared as two-dimensional images from a curved lining surface while moving inside a tunnel, corresponding these to the same or nearby cross-sections, stably detecting the wet state, and outputting the results to at least one of the lining unfolded diagram and three-dimensional point cloud for use in inspection and judgment.

[0023] In particular, when acquiring reflection images in the first and second wavelength bands while moving, using time-division or wavelength switching, positional misalignment is likely to occur between the two images due to vibrations associated with driving, changes in vehicle posture, and changes in distance to the tunnel lining. Furthermore, because the tunnel lining is curved, the incident angle and observation angle tend to vary from section to section even under the same driving conditions, making it difficult to perform stable wetness assessment using only simple difference or ratio calculations. Therefore, there is a need for a technology that can pseudo-simultaneously associate reflection images in multiple wavelength bands in the lining surface coordinate system based on at least one of the moving object's position information, posture information, and distance information.

[0024] Furthermore, in actual tunnel inspections, it is desirable to evaluate the relative degree of wetness on a ring or segment basis while suppressing the effects of dirt, efflorescence, differences in repair material properties, uneven lighting, distance dependence, angle dependence, etc., and to output the results in a development drawing or three-dimensional point cloud associated with positional attributes such as distance, ring number, top, shoulder, and side wall, which can then be used to prioritize key inspection points. However, none of the above-mentioned patent documents contain a technology that satisfies all of these requirements together. Therefore, the present invention has been made in view of the above problems, and aims to provide an apparatus, method, and program that can detect the wetness state of the tunnel lining surface with high accuracy while moving, and output the results in a format that is easy to use in inspection work. [Means for solving the problem]

[0025] To solve the above problems, according to one aspect of the present invention, a tunnel lining moisture state detection device is provided, which is mounted on a mobile body that moves inside a tunnel and detects the moisture state of the tunnel lining surface. This device includes an image acquisition unit that irradiates the lining surface with light in the short-wavelength infrared region and acquires a first reflection image corresponding to a first wavelength band in which water absorption is relatively high and a second reflection image corresponding to a second wavelength band in which water absorption is relatively lower than that of the first wavelength band, as two-dimensional images based on the reflected light from the lining surface.

[0026] In this invention, relative movement is not limited to the movement of a mobile body within a tunnel, but includes various measurement modes in which the relative position between the image acquisition unit and the tunnel lining surface changes. For example, it may include modes in which a mobile body equipped with an image acquisition unit travels, modes in which the image acquisition unit is moved or its posture is changed by a support mechanism, and modes in which the target position or imaging area changes relative to a fixed or semi-fixed image acquisition unit.

[0027] Furthermore, the image acquisition unit is not limited to a configuration in which it emits illumination light itself and acquires reflected light, but may also include a configuration in which it receives reflected light in the short-wavelength infrared region originating from external illumination, existing illumination, or other light sources. Moreover, the first wavelength band and the second wavelength band may be at least two wavelength bands in which the absorption characteristics of water are relatively different, and are not limited to a configuration in which one is a wavelength band in which water absorption is relatively large and the other is a reference wavelength band.

[0028] The image acquisition unit preferably includes a two-dimensional array type short-wavelength infrared image sensor, and acquires images of the first wavelength band and the second wavelength band by time-division switching of illumination, switching of filters provided in the imaging optical path, or wavelength selection means equivalent thereto. The first wavelength band and the second wavelength band may be, for example, near 1450 nm and near 1320 nm, respectively, but are not limited to these, and may be other short-wavelength infrared wavelength bands that can utilize the absorption difference due to moisture.

[0029] The device further includes a travel information acquisition unit that acquires information regarding at least one of the position, orientation, and distance to the lining surface of the moving body. The travel information acquisition unit may include, for example, a distance meter, encoder, DMI, IMU, laser rangefinder, LiDAR, three-dimensional point cloud acquisition device, or any combination thereof.

[0030] The apparatus further includes a processing unit that associates the first and second reflected images with the same or nearby cross-sections on the lining surface based on the travel information, and performs at least one of reprojection onto the lining surface coordinate system and pixel correspondence correction. The processing unit may associate the first and second reflected images in a pseudo-simultaneous manner based on the travel direction, imaging period, wavelength switching period, and the travel information.

[0031] The processing unit performs at least one of the following on the associated first and second reflection images: dark correction, white reference correction, and correction using a correction coefficient Cλ(d,θ) set for each wavelength according to the distance d to the target surface and the angle of incidence θ. After that, it calculates a wetness index based on the difference, ratio, normalized difference, logarithmic transformed value, or other calculated value of the two images.

[0032] The processing unit may further estimate the local normal of the lining surface based on a three-dimensional point cloud or shape information when calculating the wetness index, and perform angle correction based on the local normal. It may also calculate the reliability based on alignment error, local normal estimation error, insufficient reflectivity, saturation, and other factors, and exclude or reduce the weight of pixels or regions whose reliability does not meet a predetermined standard.

[0033] The processing unit may further automatically extract a dry reference area for each ring or segment, normalize the wetness index as a relative value to the dry reference area, and extract wetness candidates based on the normalized wetness index. Furthermore, in addition to the wetness index, the processing unit may classify the target area as a wetness candidate, suspected of being soiled or efflorescent, or no abnormality based on at least one of image features, shape features, positional attributes, or history information.

[0034] The processing unit may output the wetness index and the classification result superimposed on the lining development drawing, and may also assign them as attribute values ​​to the three-dimensional point cloud. Furthermore, the processing unit may assign a priority verification order based on at least one of the distance, ring number, position within the cross-section, continuity of the region, size of the wetness index, reliability, and historical change. Furthermore, according to another aspect of the present invention, a method and program for detecting the wet state of a tunnel lining that realize each of the above functions are provided. [Effects of the Invention]

[0035] According to the present invention, in contrast to Patent Documents 1-4 and 7-9 which primarily focus on acquiring visible images or three-dimensional shapes, the wetting state of the lining surface can be evaluated based on reflection images in at least two short-wavelength infrared wavelength bands that reflect the absorption characteristics of water. Therefore, wetness candidates can be extracted more directly while distinguishing them from mere dirt, variations in density, or changes in shape.

[0036] Furthermore, according to the present invention, the first and second reflection images acquired during travel can be associated with the same or nearby cross-sections in a pseudo-simultaneous manner based on at least one of positional information, attitude information, and distance information, thereby reducing the effects of positional shifts associated with time-resolved imaging or wavelength switching. Therefore, compared to Patent Documents 2 and 7, which disclose image-point cloud linking or synchronized imaging, and Patent Documents 6 and 10, which disclose general spectral imaging or moisture detection, a configuration more suitable for moisture evaluation under actual operating conditions targeting a moving and curved tunnel lining surface can be realized.

[0037] Furthermore, according to the present invention, dark correction, white reference correction, distance correction, and angle correction can be combined, and correction based on local normals and exclusion of unreliable pixels can be performed as needed, thereby suppressing the effects caused by curvature of the tunnel lining surface, variations in measurement position, and fluctuations in lighting conditions. In addition, since the wetness index can be relative to the dryness reference area for each ring or segment, practical wetness evaluation can be performed that is less affected by material differences, dirt, efflorescence, etc.

[0038] Furthermore, according to the present invention, since the wetness index can be output to at least one of the lining development drawing and the three-dimensional point cloud and associated with distance, ring number, position within the cross-section, and priority inspection order, it is possible to present the inspector with a new layer of information, namely the wetness state, while utilizing the advantages of the development image mapping or inspection support described in Patent Documents 3 and 4. This makes it possible to improve the efficiency of reconfirmation at the inspection site, clarify prioritization, and support maintenance management decisions.

[0039] Therefore, according to the present invention, it is possible to provide an apparatus, method, and program that can detect the wetness of a tunnel lining surface with high accuracy and efficiency while moving, and present the results in a format that is easy to use in practice.

[0040] Furthermore, according to the present invention, the extraction, recording, and reconfirmation of potential wet areas can be streamlined in periodic inspections, detailed surveys, comparisons before and after repairs, history management, and other maintenance operations of tunnels. [Brief explanation of the drawing]

[0041] [Figure 1] This is a schematic diagram showing the overall configuration of a tunnel lining wetness detection system according to one embodiment of the present invention. [Figure 2] This is a schematic cross-sectional diagram showing the arrangement of a dual-wavelength illuminator and a short-wavelength infrared imaging unit in a tunnel cross-section according to one embodiment of the present invention (a: example of symmetrical arrangement, b: example of irradiation range). [Figure 3] This is a functional block diagram of a wetness detection device according to one embodiment of the present invention. [Figure 4] This is an explanatory diagram showing the acquisition timing of a dual-wavelength image according to one embodiment of the present invention (a: time-division illumination method, b: bandpass filter switching method, c: simultaneous measurement method of two optical paths). [Figure 5] This flowchart shows the processing flow from image acquisition to wet candidate extraction and output according to one embodiment of the present invention. [Figure 6]This is an explanatory diagram illustrating a process for pseudo-simultaneously associating a first reflected image and a second reflected image according to one embodiment of the present invention with the same cross-section or a nearby cross-section. [Figure 7] This is an explanatory diagram showing reprojection to a lining surface coordinate system and pixel correspondence correction according to one embodiment of the present invention (a: before reprojection, b: after reprojection, c: example of subpixel correction). [Figure 8] This is an explanatory diagram showing dark correction, white reference correction, distance correction, angle correction, and wetness index calculation related to one embodiment of the present invention. [Figure 9] This is an explanatory diagram (a: local normal estimation, b: confidence map generation) showing local normal estimation and confidence evaluation based on a three-dimensional point cloud according to one embodiment of the present invention. [Figure 10] This is an explanatory diagram (a: extraction of dry reference areas, b: relative ranking, c: classification result) showing the extraction of dry reference areas and relative wet ranking of ring or segment units according to one embodiment of the present invention. [Figure 11] This is an explanatory diagram showing an example of output where a wetness candidate map according to one embodiment of the present invention is superimposed on a lining development drawing. [Figure 12] This is an explanatory diagram showing an example output in which a wetness index according to one embodiment of the present invention is assigned as an attribute value to a three-dimensional point cloud. [Figure 13] This is an explanatory diagram showing an example of an output screen or list with priority verification order related to one embodiment of the present invention. [Modes for carrying out the invention]

[0042] <Overview of Embodiments> Embodiments of the present invention will be described below with reference to the drawings. The embodiments described below are examples of the present invention, and the present invention is not limited to these embodiments. Furthermore, the same or corresponding parts are denoted by the same reference numerals in each figure, and redundant explanations are omitted as appropriate.

[0043] As shown in Figure 1, the wetness detection system according to this embodiment comprises a mobile body 3 that travels inside the tunnel 1 and a wetness detection device 4 mounted on the mobile body 3. The wetness detection device 4 irradiates the lining surface 2 of the tunnel 1 with short-wavelength infrared light, acquires the reflected light from the lining surface 2 as a two-dimensional image, and detects the wetness of the lining surface 2 based on the two-dimensional image. The mobile body 3 can be an inspection vehicle, maintenance vehicle, trolley, or any other movable mounting body, and is configured to continuously measure a wide area of ​​the lining surface 2 while traveling inside the tunnel 1 at a predetermined speed.

[0044] In this embodiment, in order to understand the wetness of the lining surface 2, at least two wavelength bands are used: a first wavelength band 32 with relatively high water absorption and a second wavelength band 33 with relatively low water absorption. The wetness detection device 4 acquires a first reflection image 19 corresponding to the first wavelength band 32 and a second reflection image 20 corresponding to the second wavelength band 33, and calculates a wetness index 22 based on the difference, ratio, normalized difference, and other calculations of these reflection images. This makes it possible to obtain information that reflects the degree of wetness on the lining surface 2, distinct from apparent differences caused by mere changes in density, dirt, or efflorescence in the visible image.

[0045] The objective of this embodiment is not limited to precisely quantifying the absolute moisture content on the lining surface 2, but rather to obtain relative moisture information that is useful for identifying areas that should be checked intensively during inspection work. In other words, this embodiment emphasizes efficiently extracting candidate moist areas 28 by relative comparison of each ring 26 or segment, comparison with a dry reference area 27, and comparison between parts within the same tunnel. For this reason, this embodiment does not simply acquire images, but employs a configuration that associates two-wavelength reflected images with the same or nearby cross-sections on the lining surface 2 using at least one of the position information, attitude information (direction and tilt of the measuring device, camera, and moving body) and distance information obtained during travel.

[0046] More specifically, the wetness detection device 4 includes an image acquisition unit 5 including a dual-wavelength illuminator and a short-wavelength infrared imaging unit 8, a travel information acquisition unit 11 that acquires information on the position, attitude and distance associated with the movement of the mobile body 3, a pseudo-simultaneous correspondence unit 34 that pseudo-simultaneously associates the first reflection image 19 and the second reflection image 20 based on the acquired information, a reprojection correction unit 35 that performs reprojection onto the lining surface coordinate system 21 and pixel correspondence correction, a determination unit 36 ​​that calculates and determines a wetness index 22, and a processing unit 37 that outputs the results. As a result, even with images of multiple wavelength bands acquired while moving, the effects of positional shifts, attitude changes, and distance changes associated with movement can be suppressed, and wetness evaluation can be performed on corresponding areas on the lining surface 2.

[0047] Furthermore, in this embodiment, the calculated wetness index 22 and judgment results are superimposed on the lining development drawing 24 and output, and can be retained as attribute values ​​of the three-dimensional point cloud 25 as needed. This allows inspectors to understand the candidate wet areas 28 in association with positional attributes such as the distance of the tunnel 1, ring number, top, shoulder, and side walls, and to efficiently extract areas that should be checked intensively on-site based on the reliability 23 or priority confirmation ranking 30.

[0048] For the first wavelength band 32 and the second wavelength band 33, for example, wavelengths near 1450 nm and 1320 nm can be used, but the invention is not limited to these, and other short-wavelength infrared wavelength bands that can utilize the absorption difference due to moisture may also be used. Furthermore, for the short-wavelength infrared imaging unit 8, for example, an area-scan type camera using a two-dimensional array type InGaAs image sensor can be employed, but the invention is not limited to this. In the following, based on the outline of the embodiment described above, the apparatus configuration, image acquisition method, correction processing, moisture index calculation processing, relative determination processing, and output processing will be described in detail in order.

[0049] <Overall System Configuration> The overall configuration of the wetness detection system according to this embodiment will be described with reference to Figures 1 and 3. The wetness detection system according to this embodiment targets the lining surface 2 of a tunnel 1 and detects the wetness of the lining surface 2 using a wetness detection device 4 mounted on a mobile body 3. The wetness detection device 4 comprises at least an image acquisition unit 5, a driving information acquisition unit 11, a processing unit 16, a storage unit 17, and a display unit 18.

[0050] The mobile body 3 is not particularly limited as long as it is a mountable vehicle capable of traveling inside the tunnel 1, and may be, for example, an inspection vehicle for road tunnels, a maintenance vehicle for railway tunnels, a trolley, a rail-road vehicle, a transport vehicle, or other mobile vehicle. The mobile body 3 may be configured to hold the wetness detection device 4 on the front of the vehicle body, the center of the vehicle body, the rear of the vehicle body, the roof, a liftable support base, or a support structure similar thereto. This makes it possible to image a wide area of ​​the tunnel lining surface 2 along the inner circumference of the tunnel 1.

[0051] Furthermore, the wetness detection device 4 may be integrated into a newly designed, dedicated inspection vehicle, or it may be retrofitted to an existing tunnel photography vehicle, tunnel inspection vehicle, tunnel shape measurement vehicle, or other existing vehicle. For example, the image acquisition unit 5 and processing unit 16 according to the present invention may be added while utilizing the visible camera, laser measuring instrument, odometer, position measuring instrument, or on-board power supply already installed in the existing vehicle. Moreover, part or all of the wetness detection device 4 may be configured as a detachable unit so that it can be moved and used between multiple vehicles as needed.

[0052] The image acquisition unit 5 irradiates the lining surface 2 with light in the short-wavelength infrared region and acquires a first reflection image 19 and a second reflection image 20 based on the reflected light from the lining surface 2. In the example shown in Figure 3, the image acquisition unit 5 includes a first illuminator 6, a second illuminator 7, a short-wavelength infrared imaging unit 8, a filter switching unit 9, and an illumination control unit 10. Note that the filter switching unit 9 is not essential and may be omitted depending on the acquisition method described later.

[0053] The driving information acquisition unit 11 acquires information regarding the position, attitude, distance, or shape that changes as the moving body 3 moves, and includes, for example, at least one of a DMI 12, an IMU 13, a LiDAR 14, and a rangefinder 15. The DMI 12 acquires information regarding the distance traveled, distance, or amount of movement of the moving body 3. The IMU 13 acquires information regarding the acceleration, angular velocity, attitude angle, and other attitude change information of the vehicle body. The LiDAR 14 acquires distance information to the lining surface 2 and a three-dimensional point cloud 25. The rangefinder 15 acquires information regarding the distance from at least the short-wavelength infrared imaging unit 8 to the lining surface 2. This information is used for matching two-wavelength images, reprojection correction, distance correction, angle correction, and reliability evaluation.

[0054] The processing unit 16 includes, for example, a CPU, GPU, FPGA, DSP, or other computing device, or a combination thereof, and performs wetness estimation processing based on information input from the image acquisition unit 5 and the driving information acquisition unit 11. As shown in Figure 3, the processing unit 16 may functionally include a pseudo-simultaneous mapping unit 34, a reprojection correction unit 35, a determination unit 36, and an output unit 37. The pseudo-simultaneous mapping unit 34 maps the first reflection image 19 and the second reflection image 20 to the same or nearby cross-sections on the lining surface 2. The reprojection correction unit 35 reprojects the two reflection images onto the lining surface coordinate system 21 as needed and corrects the pixel correspondence. The determination unit 36 ​​generates a wetness index 22, confidence level 23, wetness candidate area 28, and other determination results based on the corrected image. The output unit 37 outputs the results as a lining development diagram 24, a three-dimensional point cloud 25, or an inspection result list 31.

[0055] The storage unit 17 includes, for example, a semiconductor memory, magnetic disk, optical disk, or other storage medium, and stores the first reflection image 19, the second reflection image 20, driving information, a three-dimensional point cloud 25, a lining development diagram 24, a wetness index 22, a reliability score 23, a priority inspection order 30, and other information. The storage unit 17 may be a storage device mounted on the mobile body 3, or it may be an external storage device or server device connected via wireless or wired communication. The display unit 18 is, for example, an in-vehicle monitor, a tablet terminal, a mobile terminal, a head-mounted display, or other display device, and visually presents the lining development diagram 24, the wetness candidate area 28, the reliability score 23, and the priority inspection order 30 to the inspector.

[0056] In this embodiment, the arrangement of each component of the wetness detection device 4 can be changed as appropriate. For example, the image acquisition unit 5 and the driving information acquisition unit 11 may be placed on the mobile body 3, and all or part of the processing unit 16, storage unit 17, and display unit 18 may be placed on an analysis device outside the vehicle. Furthermore, at least a part of the processing unit 16 may be configured to operate in real time simultaneously with image acquisition, or it may be configured to perform offline analysis using recorded data after measurement.

[0057] <Configuration of the optical and imaging unit> Next, with reference to Figures 2 to 4, the optical and imaging units constituting the image acquisition unit 5 according to this embodiment will be described. The image acquisition unit 5 is configured to irradiate the lining surface 2 with light in the short-wavelength infrared region and acquire a first reflection image 19 corresponding to a first wavelength band 32 where water absorption is relatively high, and a second reflection image 20 corresponding to a second wavelength band 33 where water absorption is relatively low, based on the reflected light from the lining surface 2.

[0058] The first wavelength band 32 and the second wavelength band 33 can be any wavelength band in the short-wavelength infrared region that can extract differences in reflection characteristics due to the wet state of the lining surface 2, and are not limited to specific wavelengths. For example, the first wavelength band 32 can be a wavelength band around 1450 nm, and the second wavelength band 33 can be a wavelength band around 1320 nm. With such a combination, it is easy to obtain information that reflects the degree of wetness on the lining surface 2 by utilizing the difference in moisture absorption characteristics. However, the present invention is not limited to the above combination of wavelength bands, and other short-wavelength infrared wavelength bands may be used as long as they are a combination of a wavelength band with relatively large moisture absorption and a wavelength band with relatively small moisture absorption.

[0059] In the example shown in Figure 2, the image acquisition unit 5 includes a first illuminator 6, a second illuminator 7, and a short-wavelength infrared imaging unit 8. The first illuminator 6 is an illuminator that illuminates the lining surface 2 with light in the first wavelength band 32, and the second illuminator 7 is an illuminator that illuminates the lining surface 2 with light in the second wavelength band 33. The first illuminator 6 and the second illuminator 7 may each be composed of independent light sources, or they may be composed of array light sources with multiple light-emitting elements arranged in a row, or they may be configured to include a switchable single light source.

[0060] For example, light-emitting diodes, laser diodes, SLDs, and other semiconductor light sources can be used as light sources for the first illuminator 6 and the second illuminator 7. The illuminators may be configured to perform narrowband illumination with a predetermined bandwidth, or they may be configured to emit relatively broadband light while defining the effective wavelength band with a subsequent filter. Furthermore, collimators, diffusers, reflectors, lenses, light guide members, and other optical components may be appropriately combined in the illuminators.

[0061] The short-wavelength infrared imaging unit 8 receives reflected light from the lining surface 2 and acquires a first reflection image 19 and a second reflection image 20 as two-dimensional images. A short-wavelength infrared imaging unit 8 preferably uses an imaging device equipped with a two-dimensional array type short-wavelength infrared image sensor. For example, an area-scan type camera equipped with an InGaAs image sensor can be used. Such a configuration makes it easier to acquire a two-dimensional image of the curved lining surface 2 of the tunnel 1 in a single step, which is advantageous for subsequent reprojection correction, unfolding diagram generation, and correspondence with a three-dimensional point cloud.

[0062] However, the short-wavelength infrared imaging unit 8 is not limited to the area-scan type InGaAs camera described above. That is, other imaging devices capable of acquiring two-dimensional reflection images in the short-wavelength infrared region may be used, and a configuration combining multiple imaging units may be used as needed. Therefore, the short-wavelength infrared imaging unit 8 in the present invention should be broadly interpreted as an imaging unit capable of acquiring two-dimensional reflection images corresponding to at least two wavelength bands.

[0063] The image acquisition unit 5 may employ multiple acquisition methods to acquire the first reflected image 19 and the second reflected image 20. For example, a time-division illumination method may be employed in which the illumination control unit 10 alternately lights up the first illuminator 6 and the second illuminator 7 in a time-division manner, and the short-wavelength infrared imaging unit 8 acquires images in synchronization with each lighting timing. Alternatively, a method may be employed in which a filter switching unit 9 located in the imaging optical path switches the bandpass filter to acquire images corresponding to the first wavelength band 32 and the second wavelength band 33. Furthermore, a configuration may be adopted in which the optical path for the first wavelength band 32 and the optical path for the second wavelength band 33 are separated, and images of both wavelength bands are acquired simultaneously.

[0064] In this embodiment, a configuration in which a first reflection image 19 and a second reflection image 20 are acquired using a single short-wavelength infrared imaging unit 8 is particularly preferred. This makes it easier to suppress positional shifts, magnification differences, or distortion differences between images corresponding to two wavelength bands that are caused by differences in the imaging optical system. However, the present invention is not limited to a single imaging unit, and a configuration in which images for each wavelength band are acquired using multiple imaging units may also be used. In this case, the correspondence between images may be corrected based on the positional relationship, optical characteristics, or calibration results of each imaging unit.

[0065] Thus, the optical and imaging unit according to this embodiment is equipped with an illuminator and a short-wavelength infrared imaging unit 8 corresponding to a first wavelength band 32 in which water absorption is relatively high and a second wavelength band 33 in which water absorption is relatively low, thereby enabling the acquisition of two-wavelength reflection images that reflect the wet state of the lining surface 2. While wavelength bands near 1450 nm and near 1320 nm, and area-scan type InGaAs cameras are cited as preferred examples, the present invention is not limited to these examples and includes other wavelength bands and other short-wavelength infrared imaging configurations that achieve similar effects.

[0066] <Lighting layout configuration> Next, with reference to Figure 2, the configuration of the illumination arrangement according to this embodiment will be described. In this embodiment, the first illuminator 6 and the second illuminator 7, and the short-wavelength infrared imaging unit 8 are arranged symmetrically or quasi-symmetrically with respect to the tunnel cross-section, according to the cross-sectional shape of the tunnel 1 and the extent of the lining surface 2. This makes it possible to illuminate and image at least a portion of the top, shoulders, and side walls of the lining surface 2 relatively uniformly.

[0067] Here, "symmetrical arrangement" refers to an arrangement in which the first illuminator 6 and the second illuminator 7 are arranged symmetrically or circumferentially with respect to the center line of the tunnel cross-section, the center line of the moving body 3, or the optical axis of the short-wavelength infrared imaging unit 8. Furthermore, "quasi-symmetrical arrangement" refers to an arrangement that is not perfectly symmetrical, but is arranged substantially symmetrically to obtain a desired illumination distribution for multiple parts of the lining surface 2, depending on the tunnel cross-sectional shape, vehicle structure, existing equipment, attached objects, and other constraints. Therefore, the present invention is not limited to cases where the positional relationship between the illuminators and the camera is geometrically strictly symmetrical, but broadly includes arrangements that can image a substantially wide area of ​​the lining surface 2.

[0068] For example, as shown in Figure 2(a), the short-wavelength infrared imaging unit 8 may be positioned in the center or approximately center in the width direction of the moving body 3, with the first illuminator 6 and the second illuminator 7 positioned to its left and right. Alternatively, the short-wavelength infrared imaging unit 8 may be slightly offset downward or upward, and the first illuminator 6 and the second illuminator 7 may be positioned at different heights or angles. Furthermore, multiple first illuminators 6 and multiple second illuminators 7 may be dispersed along the tunnel cross-sectional direction, so that the illumination areas of each illuminator partially overlap. Such a configuration makes it easier to achieve illumination with less unevenness in illumination over a wide area from the top to the shoulder and from the shoulder to the side wall.

[0069] The illumination angle, illumination width, and light distribution characteristics of the illuminator are appropriately set according to the range of the lining surface 2 to be imaged, the distance from the moving body 3 to the lining surface 2, the field of view of the short-wavelength infrared imaging unit 8, and the required spatial resolution. For example, when focusing on imaging the top surface, an illuminator with a relatively upward illumination angle may be used, and when covering a wide area of ​​the shoulder and side walls, an illuminator with a relatively wide illumination width may be used. Alternatively, different illumination angles or widths may be set for each illuminator to create an optimized illumination distribution for each part within the cross-section.

[0070] The height and position of the short-wavelength infrared imaging unit 8 can also be modified as appropriate. For example, the short-wavelength infrared imaging unit 8 may be placed on the roof or elevated support of the mobile body 3 to ensure a clear line of sight to the top and shoulders, or it may be placed in the center or lateral support of the vehicle body to improve imaging conditions for the side walls. Furthermore, the short-wavelength infrared imaging unit 8 may be supported by a lifting mechanism, telescopic mechanism, rotating mechanism, or tilt mechanism so that its height, orientation, or line of sight can be changed according to the tunnel cross-sectional shape or driving conditions.

[0071] Furthermore, the first illuminator 6 and the second illuminator 7 may consist of a single illuminator for each wavelength band, or they may be composed of a combination of multiple illuminators. For example, multiple first illuminators 6 corresponding to the first wavelength band 32 may be arranged separately for the top, shoulder, and side walls, and multiple second illuminators 7 corresponding to the second wavelength band 33 may be arranged accordingly. In this case, by individually controlling the emission intensity, irradiation angle, or lighting timing of multiple illuminators belonging to the same wavelength band, illumination control can be performed according to the reflection conditions of each part of the lining surface 2.

[0072] Furthermore, the lighting arrangement may be modified depending on whether the cross-sectional shape of tunnel 1 is circular, horseshoe-shaped, rectangular, or otherwise. For example, in a horseshoe-shaped cross-section, the angle of incidence tends to change near the shoulder, so directional illuminators may be added towards the shoulder, and in a rectangular cross-section, auxiliary lighting may be provided from an oblique direction to the connection between the side wall and the ceiling. In addition, if shadows are likely to occur due to attached objects, wiring, signs, or other obstacles, the effect of shadows may be reduced by providing lighting from multiple directions.

[0073] Thus, the lighting arrangement configuration in this embodiment is based on arranging the illuminators and cameras symmetrically or semi-symmetrically with respect to the tunnel cross-section, enabling imaging of at least a portion of the top, shoulder, and side walls. However, the illumination angle, illumination width, arrangement height, arrangement position, number and combination of illuminators, and other design elements can be appropriately selected according to the conditions of the target tunnel. Therefore, the present invention is not limited to the specific arrangement examples shown, but includes various lighting arrangements suitable for wide-area imaging of the lining surface 2 and detection of wet conditions.

[0074] <Method for acquiring dual-wavelength images> Next, with reference to Figure 4, the dual-wavelength image acquisition method according to this embodiment will be described. In this embodiment, in order to obtain information reflecting the wet state of the lining surface 2, a first reflection image 19 corresponding to the first wavelength band 32 and a second reflection image 20 corresponding to the second wavelength band 33 are acquired. As acquisition methods for these two reflection images, for example, a bandpass filter switching method, a time-division illumination method, and a dual-path simultaneous measurement method can be employed. These methods may be used individually or selectively switched depending on the device configuration or operating conditions.

[0075] The bandpass filter switching method, as shown in Figure 4(b), involves providing a filter switching unit 9 on the imaging optical path of the short-wavelength infrared imaging unit 8. By switching between a first filter that transmits the first wavelength band 32 and a second filter that transmits the second wavelength band 33 using the filter switching unit 9, a first reflection image 19 and a second reflection image 20 are acquired. This method allows for the selection of a desired wavelength band on the receiving side while keeping the illumination side configuration relatively simple. Furthermore, since the short-wavelength infrared imaging unit 8 is common and the target wavelength band can be sequentially switched, it is easy to miniaturize the device or apply it to retrofitting existing vehicles.

[0076] In the bandpass filter switching method described above, the filter switching unit 9 can be a mechanical filter wheel, a sliding switching mechanism, a liquid crystal variable filter, an acousto-optic filter, an interference filter switching mechanism, or other wavelength selection means. The illuminator side may be broadband illumination including both the first wavelength band 32 and the second wavelength band 33, or multiple illuminators suitable for each wavelength band may be used in combination. The timing of the filter switching is preferably controlled in synchronization with the exposure start, exposure end, or frame update timing of the short-wavelength infrared imaging unit 8.

[0077] The time-division illumination method, as shown in Figure 4(a), involves the illumination control unit 10 alternately illuminating the first illuminator 6 and the second illuminator 7 in a time-divided manner, and the short-wavelength infrared imaging unit 8 acquiring the first reflected image 19 and the second reflected image 20 in accordance with the respective illumination timings. For example, the first illuminator 6 may be illuminated during a certain frame period to acquire the first reflected image 19, and the second illuminator 7 may be illuminated during the next frame period or at a different timing within the same cycle to acquire the second reflected image 20. This method allows for the acquisition of images corresponding to two wavelength bands using the same short-wavelength infrared imaging unit 8 and under substantially the same line-of-sight conditions.

[0078] In particular, in the time-division illumination method described above, the first reflected image 19 and the second reflected image 20 can be acquired using the same camera, the same or substantially the same imaging optical system, and the same or substantially the same line of sight direction. Therefore, compared to cases where different cameras or different optical systems are used for each wavelength band, it is easier to suppress pixel shift, magnification difference, distortion difference, or parallax between images. For this reason, it is easier to reduce the amount of correction when matching the first reflected image 19 and the second reflected image 20 in a later stage, which is advantageous in improving the accuracy of calculating the wetness index 22. However, since positional or attitude changes may occur between time-division acquisitions while the moving body 3 is in motion, pseudo-simultaneous matching or reprojection correction, as described later, can be used in combination as needed.

[0079] The dual-path simultaneous measurement method, as shown in Figure 4(c), is a method in which an optical path corresponding to the first wavelength band 32 and an optical path corresponding to the second wavelength band 33 are branched, and the corresponding reflected light is simultaneously received in each optical path to acquire the first reflected image 19 and the second reflected image 20. For example, the incident light may be branched using a beam splitter, dichroic mirror, or other spectroscopic optical element and guided to the image sensor or light-receiving surface corresponding to each optical path. With this method, images corresponding to the two wavelength bands can be acquired at substantially the same time, so that positional or attitude differences caused by time differences during travel can be suppressed to a smaller extent.

[0080] However, in the aforementioned dual-path simultaneous measurement method, differences in magnification, distortion, or parallax may occur between the optical paths due to the optical characteristics of each optical path, the characteristics of the image sensor, or assembly errors. For this reason, the geometric correspondence between each optical path may be corrected based on previously acquired calibration data, if necessary. Furthermore, while the dual-path simultaneous measurement method may result in a more complex optical system compared to the time-division illumination method or the bandpass filter switching method, it is effective when the travel speed of the moving object 3 is high or when it is desirable to minimize the time difference.

[0081] As described above, in this embodiment, a bandpass filter switching method, a time-division illumination method, and a dual-path simultaneous measurement method can be employed as acquisition methods for the first reflection image 19 and the second reflection image 20. The appropriate method can be selected according to the simplification of the device configuration, suppression of pixel misalignment, ensuring simultaneity, acquisition speed, and other design requirements. Therefore, the present invention is not limited to the specific examples shown in Figure 4, but includes various acquisition methods capable of acquiring reflection images corresponding to at least two short-wavelength infrared wavelength bands.

[0082] <Configuration of position and distance synchronization system> Next, the configuration of the position-distance synchronization system according to this embodiment will be described with reference to Figures 1, 3, and 5. In this embodiment, since the moving body 3 images the lining surface 2 while traveling inside the tunnel 1, it is important to acquire information regarding the position, orientation, distance, or shape of the moving body 3 in order to associate the first reflected image 19 and the second reflected image 20 with the same or nearby cross-sections on the lining surface 2. For this reason, the wetness detection device 4 is equipped with a travel information acquisition unit 11, and is configured to acquire various information that changes as the moving body 3 travels.

[0083] The driving information acquisition unit 11 includes, for example, at least one of a DMI 12, an IMU 13, a LiDAR 14, and a distance meter 15. The DMI 12 acquires information regarding the driving distance, distance, amount of movement, or speed of the moving body 3, and may consist of, for example, a distance meter based on wheel rotation speed, a rotary encoder, a pulse generator, or other distance measuring means. The IMU 13 acquires information regarding acceleration, angular velocity, tilt angle, yaw angle, roll angle, pitch angle, and other attitude changes, thereby enabling the detection of vibrations, swaying, or changes in vehicle attitude that occur during driving.

[0084] Furthermore, the LiDAR 14 acquires distance information to the lining surface 2, three-dimensional position information, and a three-dimensional point cloud 25, which are used as shape information representing the inner circumferential shape or local surface shape of the tunnel 1. As a result, the processing unit 16 can reproject the acquired first reflection image 19 and second reflection image 20 onto the lining surface coordinate system 21, or obtain basic information for estimating the local normal 29. The rangefinder 15 is composed of, for example, a laser rangefinder, an ultrasonic rangefinder, an optical range sensor, or other distance measuring means, and can acquire information regarding the distance from at least the short-wavelength infrared imaging unit 8 or each illuminator to the lining surface 2.

[0085] The driving information acquisition unit 11 is not limited to the above-mentioned sensors, and may include various sensors or measuring means capable of acquiring information regarding the position, attitude, distance, or shape of the moving body 3. For example, it may include a speedometer that outputs vehicle speed information of the moving body 3, a vibration sensor that detects vertical movement of the vehicle body, a stereo camera, a line laser measuring instrument, an image sensor that detects an external reference marker, an inertial navigation system, or a sensor equivalent thereto. Furthermore, the system may be configured to combine position information that can be acquired outside the tunnel with distance information or attitude information acquired inside the tunnel.

[0086] In this embodiment, it is preferable that the various information acquired by the driving information acquisition unit 11 is acquired in synchronization with the image acquisition timing by the short-wavelength infrared imaging unit 8. For example, common time information, frame number, trigger number, or count value may be assigned to the pulse of the DMI 12, the output time of the IMU 13, the scan time of the LiDAR 14, the measurement time of the rangefinder 15, and the exposure time of the short-wavelength infrared imaging unit 8. This allows the processing unit 16 to manage the correspondence between each reflected image and the driving information with high accuracy, and to appropriately perform pseudo-simultaneous correspondence and reprojection correction of the first reflected image 19 and the second reflected image 20.

[0087] Furthermore, the processing unit 16 may use information acquired from multiple sensors individually, or it may combine and integrate this information. For example, by using distance information acquired by DMI 12 as a reference, correcting attitude change information acquired by IMU 13, and further combining this with distance information or shape information acquired by LiDAR 14 or rangefinder 15, the driving state of the mobile body 3 and the geometric conditions of the lining surface 2 can be grasped with higher accuracy. Such sensor fusion makes it easier to suppress the cumulative error in driving distance, image position shift due to vibration, and the effects of distance fluctuations to the lining surface 2.

[0088] Furthermore, in this embodiment, if the wetness detection device 4 is retrofitted to an existing tunnel photography vehicle, tunnel inspection vehicle, or shape measurement vehicle, it may be configured to utilize the driving information held by the existing vehicle. For example, necessary information may be acquired from an odometer, wheel pulse signal, attitude sensor, laser rangefinder, three-dimensional point cloud acquisition device, distance information output device, or time synchronization device already installed in the existing vehicle and input to the driving information acquisition unit 11. In this case, by additionally installing the image acquisition unit 5 and processing unit 16 according to the present invention, a wetness detection function can be added to an existing tunnel photography system or inspection system.

[0089] Furthermore, if the format, accuracy, or update cycle of the driving information acquired from existing vehicles does not directly conform to the processing of the present invention, the processing unit 16 may interpolate, resample, correct, or transform the information before using it. For example, interpolation processing may be performed on distance information obtained from existing vehicles in accordance with the image acquisition timing, or the lining surface coordinate system 21 may be constructed using existing three-dimensional shape data or point cloud data. This makes it possible to apply the effects of the present invention to existing equipment while minimizing the addition of new sensors.

[0090] As described above, the position-distance synchronization system in this embodiment is configured to acquire position, orientation, distance, or shape information of the moving object 3 using DMI12, IMU13, LiDAR14, rangefinder15, and other sensors, and to supply this information to the processing unit 16 in synchronization with the image acquisition timing. This makes it possible to appropriately perform same-section correspondence, reprojection correction, distance correction, angle correction, and reliability evaluation on the first reflected image 19 and second reflected image 20 acquired while moving in a subsequent stage.

[0091] <Data acquisition process> Next, with reference to Figure 5, the data acquisition process according to this embodiment will be described. In this embodiment, while the moving body 3 is traveling inside the tunnel 1, the image acquisition unit 5 acquires a first reflection image 19 and a second reflection image 20, and the travel information acquisition unit 11 acquires position information, attitude information, distance information, or point cloud information. These acquired data are then recorded in a state that allows them to be correlated with each other and supplied to the subsequent pseudo-simultaneous correlation process, reprojection correction process, wetness index calculation process, and output process.

[0092] Specifically, the image acquisition unit 5 sequentially or simultaneously acquires a first reflection image 19 based on the reflected light of the first wavelength band 32 from the lining surface 2 and a second reflection image 20 based on the reflected light of the second wavelength band 33, according to the acquisition methods described above. At this time, it is preferable that each reflection image acquired by the short-wavelength infrared imaging unit 8 is assigned at least a frame number, acquisition time, wavelength band identification information, and exposure condition information as needed. Here, the wavelength band identification information is information indicating whether the image corresponds to the first wavelength band 32 or the second wavelength band 33.

[0093] Meanwhile, the driving information acquisition unit 11 acquires position information, attitude information, distance information, or point cloud information of the moving object 3 in parallel with image acquisition. For example, the distance, travel distance, or amount of movement may be acquired by the DMI 12, the attitude angle, angular velocity, or acceleration may be acquired by the IMU 13, the distance information to the lining surface 2 or a three-dimensional point cloud 25 may be acquired by the LiDAR 14, and the representative distance from the imaging position to the lining surface 2 may be acquired by the rangefinder 15. It is preferable that each of these pieces of information is also associated with the image data, with acquisition time, trigger number, count value, or distance information.

[0094] In this embodiment, it is preferable to use a common synchronization reference in order to synchronously acquire the first reflected image 19, the second reflected image 20, and various driving information. For example, the driving timing of the illumination control unit 10 and the short-wavelength infrared imaging unit 8 may be used as a reference, and a trigger signal corresponding to that timing may be supplied to the driving information acquisition unit 11, or conversely, the distance pulse of the DMI 12 or an external synchronization signal may be used as a reference to synchronize the operation of the image acquisition unit 5 and the driving information acquisition unit 11. Alternatively, common time information may be added to each data so that they correspond to each other on the time axis in subsequent stages.

[0095] When employing a time-division illumination method or a bandpass filter switching method, the acquisition times of the first reflection image 19 and the second reflection image 20 may differ. Therefore, additional information indicating the acquisition order, position within the acquisition cycle, illumination conditions, or filter status may be added to each image. This makes it easier for the processing unit 16 to determine which first reflection image 19 should be associated with which second reflection image 20. For example, a common pair number may be assigned to two images acquired consecutively within the same cycle, and the first reflection image 19 and the second reflection image 20 may be treated as a pair of image data based on this pair number.

[0096] Furthermore, when acquiring a three-dimensional point cloud 25, it is preferable to assign a frame number, acquisition time, distance, or position identifier that can be associated with the image to each point cloud data or point cloud frame. This allows the processing unit 16 to refer to the three-dimensional point cloud 25 acquired at the corresponding time or distance for a given first reflection image 19 and second reflection image 20, and use it for constructing the lining surface coordinate system 21, reprojection correction, estimation of local normals 29, or evaluation of confidence 23.

[0097] The acquired first reflection image 19, second reflection image 20, position information, orientation information, distance information, and point cloud information are input to the processing unit 16 or the storage unit 17. At this time, the processing unit 16 may temporarily buffer each data, organize it based on the frame number, time information, distance, trigger number, or set number, and then pass it on to the subsequent pseudo-simultaneous mapping unit 34 and reprojection correction unit 35. When recording to the storage unit 17, each data may be recorded as an individual file or record, or it may be recorded as an integrated data structure including mutual correspondence information.

[0098] Thus, according to the data acquisition process in this embodiment, the first reflection image 19, the second reflection image 20, position information, attitude information, distance information, or point cloud information acquired during driving can be acquired and recorded in a state in which they can be correlated with each other based on the frame number, time information, distance, etc. Therefore, in subsequent processing, combination determination of two-wavelength images, correlation to the same or nearby cross-sections, reprojection correction to the lining surface coordinate system 21, and calculation of the wetness index 22 can be performed efficiently and with high accuracy.

[0099] <Simulated simultaneous measurement and correspondence of the same cross-section> Next, with reference to Figures 6 and 7, the pseudo-simultaneous measurement and same-section correspondence according to this embodiment will be described. In this embodiment, when the first reflection image 19 and the second reflection image 20 are acquired by a time-division illumination method or a bandpass filter switching method, the acquisition times of both images may not perfectly coincide. Therefore, due to position changes, distance progression, attitude changes, or distance fluctuations to the lining surface 2 that occur during the movement of the mobile body 3, the first reflection image 19 and the second reflection image 20 may represent slightly different regions on the lining surface 2. In this embodiment, the acquired two-wavelength images are processed to correspond to the same or nearby cross-sections on the lining surface 2 in a pseudo-simultaneous manner.

[0100] Here, "same cross section" refers to the same positional range in a cross section substantially perpendicular to the extension direction of the tunnel 1, and "neighboring cross section" refers to a cross section within a predetermined distance from the same cross section. In this embodiment, an allowable range to be treated as the same cross section or an allowable range to be treated as a neighboring cross section may be set according to the travel speed, image acquisition cycle, lighting switching cycle, and required accuracy. For example, the processing unit 16 may select the first reflected image 19 and the second reflected image 20, whose distance difference, acquisition time difference, or attitude difference is less than or equal to a predetermined threshold, as a candidate image pair representing the same cross section or a neighboring cross section.

[0101] The pseudo-simultaneous matching unit 34 uses at least one of the following when selecting candidate image pairs: travel direction, acquisition time, distance, and attitude change. For example, based on distance information from DMI 12, a second reflected image 20 acquired at the closest distance to the acquisition position of a certain first reflected image 19 may be extracted. Alternatively, based on the acquisition time difference, images within a predetermined time window may be considered as candidates, and the image with the smallest distance difference or attitude difference may be selected from among them. Furthermore, considering the amount of pitch, roll, or yaw change based on IMU 13, image pairs with large attitude changes may be excluded, or the image pair with the smallest correction amount may be given priority.

[0102] The pseudo-simultaneous matching unit 34 estimates the relative positional relationship between the first reflection image 19 and the second reflection image 20 based on the displacement amount in the direction of travel of the moving body 3, the change in attitude, and the change in distance to the lining surface 2, as needed. For example, the amount of movement of the moving body 3 between the acquisition time of the first reflection image 19 and the acquisition time of the second reflection image 20 may be determined by the DMI 12, and the amount of displacement in the tunnel extension direction may be estimated based on this amount of movement. Alternatively, the rotation component in the optical axis direction or in the cross-sectional direction may be estimated based on the output of the IMU 13, and the parallax component or scaling component may be corrected using distance information obtained by the LiDAR 14 or rangefinder 15, as needed. This makes it possible to match two-wavelength images acquired at different times as if they were acquired at the same time.

[0103] Furthermore, the pseudo-simultaneous correspondence unit 34 may apply translation, rotation, scaling, affine transformation, projection transformation, or similar geometric correction to at least one of the first reflection image 19 or the second reflection image 20 based on the estimation results of the relative positional relationship. This allows for adjustment so that the corresponding areas on the lining surface 2 are more aligned in the image. In addition, if a cross-sectional model of the tunnel 1, a lining surface coordinate system 21, or a three-dimensional point cloud 25 is available, the correspondence may be performed after projecting each image onto the same cross-sectional reference or reference coordinate system by referring to these.

[0104] In this embodiment, subpixel alignment may be performed after or in combination with the geometric correction, if necessary. For example, the processing unit 16 may estimate a positional displacement amount finer than pixel-level based on luminance correlation, edge distribution, feature points, template match, phase correlation, and other image matching indicators between the first reflection image 19 and the second reflection image 20. Based on the estimated positional displacement amount, the local correspondence accuracy between the two wavelength images may be improved by rearranging one of the images at the subpixel level or by applying interpolation processing. This makes it easier to reduce errors caused by slight mismatches between images when calculating the wetness index 22.

[0105] Furthermore, the pseudo-simultaneous matching unit 34 may perform a quality evaluation of the matching results. For example, it may evaluate the consistency quality of the image pair based on the acquisition time difference, distance difference, estimated displacement, posture difference, image correlation value, residual amount, and other indicators, and exclude image pairs whose quality does not meet predetermined standards from the calculation of the wetness index, or set a lower confidence level 23. This makes it possible to suppress misjudgments based on inappropriate image pairs, even when there are large vibrations, sudden changes in posture, or excessive positional displacement during driving.

[0106] Thus, with the pseudo-simultaneous measurement and same-section correspondence in this embodiment, the first reflection image 19 and the second reflection image 20, acquired by time-division or wavelength switching, can be pseudo-simultaneously associated with the same or nearby cross-sections on the lining surface 2 using the direction of travel, acquisition time, distance, change in attitude, etc. Furthermore, by performing sub-pixel alignment as needed, the correspondence accuracy between the two wavelength images can be improved, contributing to improved accuracy in subsequent reprojection correction, wetness index calculation, and judgment processing.

[0107] <Reprojection correction to the lining surface coordinate system> Next, with reference to Figure 7, the reprojection correction to the lining surface coordinate system 21 according to this embodiment will be explained. Due to the pseudo-simultaneous correspondence described above, the first reflection image 19 and the second reflection image 20 are arranged to correspond to the same cross-section or nearby cross-section on the lining surface 2. However, since the lining surface 2 of the tunnel 1 is curved, and the position, orientation, and distance to the lining surface 2 of the moving body 3 fluctuate during travel, areas that appear to be at the same pixel position in the image do not necessarily represent the same physical position on the lining surface 2. Therefore, in this embodiment, the first reflection image 19 and the second reflection image 20 are reprojected to the lining surface coordinate system 21 and corrected to correspond to the same area on the lining surface 2.

[0108] The lining surface coordinate system 21 is a coordinate system for representing the position of the tunnel 1 on the lining surface 2, and may include, for example, distance coordinates along the tunnel extension direction and cross-sectional coordinates along the circumferential, arc length, height, or width directions within the cross-section. Alternatively, it may be configured as a coordinate representation based on ring 26 number, segment number, cross-sectional angle, distance from the top, distance from the side wall, or other positional attributes. Therefore, the lining surface coordinate system 21 in the present invention only needs to be a coordinate system that can uniquely or substantially uniquely identify each region on the lining surface 2, and its specific definition is appropriately set according to the shape of the tunnel 1, the maintenance method, or the output format.

[0109] The processing unit 16 uses at least one of the following to construct the lining surface coordinate system 21: model information relating to the cross-sectional shape of the tunnel 1, a three-dimensional point cloud 25 acquired by the LiDAR 14, distance information acquired by the rangefinder 15, and position or orientation information acquired by the IMU 13 or DMI 12. For example, circular, horseshoe, rectangular, or other cross-sectional models may be defined in advance based on design drawings, known tunnel cross-sectional specifications, or prior measurement results. If the three-dimensional point cloud 25 is available, the actual shape of the lining surface 2 may be estimated from the three-dimensional point cloud 25, the cross-sectional model may be corrected, or the lining surface coordinate system 21 based on the measured shape may be constructed.

[0110] Specifically, the processing unit 16 calculates the line of sight corresponding to each pixel of the first reflected image 19 or the second reflected image 20 based on the internal and external parameters of the short-wavelength infrared imaging unit 8, as well as the position and orientation of the moving object 3 at the time of acquisition. Then, it finds the intersection point between this line of sight and the lining surface 2 represented by the cross-sectional model or three-dimensional point cloud 25, and converts this intersection point to coordinates on the lining surface coordinate system 21. This allows each pixel on the image to be associated with a real-space position on the lining surface 2. The intersection point between the line of sight and the lining surface 2 may be determined analytically, or by numerical calculation, nearest neighbor search, or surface approximation calculation.

[0111] For a tunnel 1 with a circular cross-section, the intersection point may be determined using the relationship between the radial and circumferential directions with respect to the center of the cross-section. For a horseshoe-shaped cross-section, the top, shoulder, and side walls may be modeled as a combination of multiple curved or curved surface elements. For a rectangular cross-section, each planar wall or corner may be modeled individually, and an appropriate intersection point can be selected from among the candidates between the line of sight and each wall. Furthermore, even in cases of other deformed cross-sections, flattened cross-sections, or localized repair shapes, the reprojection correction of the present invention can be applied by reflecting the actual lining surface shape using a three-dimensional point cloud 25 or distance information. Therefore, the present invention is not limited to a specific cross-sectional shape and is applicable to circular, horseshoe, rectangular, and various other tunnel cross-sectional models.

[0112] Furthermore, if a three-dimensional point cloud 25 is available, the processing unit 16 may estimate the nearest neighbor point, local surface, or approximate surface on the lining surface 2 from the relationship between the line of sight corresponding to each pixel and the three-dimensional point cloud 25, and use this to calculate the reprojection position. For example, local plane approximation or local surface approximation may be performed on multiple point cloud points that exist near the line of sight, and the intersection point with the approximate surface may be found. Alternatively, the intersection point may be calculated after applying radius correction, wall position correction, or scale correction to the cross-sectional model using the representative distance obtained from the distance meter 15. This makes it possible to suppress the discrepancy between the design cross-section and the actual lining surface 2, or the effects of distance fluctuations during driving.

[0113] After converting each pixel of the first reflection image 19 and the second reflection image 20 to the lining surface coordinate system 21 as described above, the processing unit 16 rearranges the pixel values ​​of each image on a predetermined lining surface coordinate grid, unfolded image grid, or cross-sectional mesh. For example, by associating the pixel values ​​of the first reflection image 19 and the pixel values ​​of the second reflection image 20 corresponding to the same coordinate or the same mesh on the lining surface coordinate system 21, it may be possible to compare reflection information of two wavelengths for the same area on the lining surface 2. At this time, if necessary, nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, or other interpolation processes may be used to obtain the pixel values ​​after reprojection.

[0114] Furthermore, the processing unit 16 may evaluate the matching status of the first reflected image 19 and the second reflected image 20 obtained as a result of reprojection correction, and may further adjust the correction amount as necessary. For example, the reprojection parameters may be updated to improve the matching accuracy based on local correlation values, edge consistency, degree of agreement with point cloud shape, or residual amount between the reprojected images. In addition, blind spots, shaded areas, areas with insufficient reflection, or areas with insufficient point cloud density may be masked as invalid areas and excluded from the calculation of the wetness index 22.

[0115] Thus, by reprojection correction to the lining surface coordinate system 21 in this embodiment, the first reflection image 19 and the second reflection image 20 can be converted to the lining surface coordinate system 21 using the tunnel cross-sectional shape, the three-dimensional point cloud 25 obtained from LiDAR 14, distance information, and other geometric information, and corrected to correspond to the same area on the lining surface 2. As a result, even when targeting a curved lining surface 2, the accuracy of calculating the wetness index 22 based on the two-wavelength image can be improved, and the reliability of subsequent relative determination and output processing can be increased.

[0116] <Correction process> Next, the correction process according to this embodiment will be described with reference to Figure 8. In this embodiment, after or in parallel with the reprojection correction to the lining surface coordinate system 21, at least one of the following is performed on the first reflection image 19 and the second reflection image 20: dark correction, white reference correction, illumination unevenness correction, distance correction, and angle correction. This reduces the effects caused by the dark current of the image sensor, variations in illumination conditions, fluctuations in the distance to the lining surface 2, and fluctuations in the angle of incidence, and makes it possible to obtain reflection information suitable for calculating the wetness index 22.

[0117] Dark correction is a correction method to reduce dark current, offset components, thermal noise, and other background components originating from the short-wavelength infrared imaging unit 8 or its image sensor. For example, a dark image acquired in advance with the lights off or the imaging optical path shielded may be used, and the pixel values ​​of the dark image may be subtracted from the corresponding pixels of the first reflection image 19 and the second reflection image 20. The dark image may also be updated at regular intervals, with predetermined temperature changes, or at startup. This makes it easier to suppress the influence of dark current components that differ for each wavelength band or imaging condition.

[0118] White reference correction is a correction that compensates for sensitivity differences in the imaging system or illumination system and normalizes the reflectance intensity in each wavelength band. For example, a reference plate 38 having known reflectance characteristics may be placed within or near the field of view of the short-wavelength infrared imaging unit 8, and the pixel values ​​of the first reflected image 19 and the second reflected image 20 may be corrected based on the reflectance intensity of the reference plate 38. Alternatively, a known reflective surface in the tunnel 1, such as a reference surface of a known material, a known dry area, or a pre-calibrated reflective reference section, may be used to determine sensitivity correction coefficients for each wavelength band. This makes it possible to correct wavelength-specific output differences of the illuminator, wavelength sensitivity differences of the image sensor, or transmittance differences of the optical system.

[0119] Illumination unevenness correction is a correction to reduce non-uniformity caused by differences in illumination intensity or light reception intensity depending on the position on the lining surface 2. For example, the illuminance distribution or sensitivity distribution within the field of view may be measured in advance using a reference plate 38 or a known reflective surface, and the pixel values ​​of the first reflection image 19 and the second reflection image 20 may be corrected in a position-dependent manner based on that distribution. Illumination unevenness correction may also be performed using a correction map corresponding to the position in the tunnel cross-sectional direction, the screen coordinate direction, or the lining surface coordinate system 21. Furthermore, when using multiple illuminators in combination, the correction amount may be set considering the contribution distribution of each illuminator.

[0120] Distance correction is a process for correcting the reflection intensity, which changes depending on the distance from the short-wavelength infrared imaging unit 8 or each illuminator to the lining surface 2. In this embodiment, the distance d corresponding to each pixel or region on the lining surface 2 is determined based on the LiDAR 14, the rangefinder 15, or geometric information obtained during reprojection correction. Then, the pixel values ​​of the first reflection image 19 and the second reflection image 20 may be corrected using a correction coefficient 39 set according to the distance d. For example, the reflection intensity may be increased or decreased according to the distance d and normalized to pixel values ​​that can be compared between close-range imaging and long-range imaging.

[0121] Angle correction is a process for correcting changes in reflection intensity caused by the angle between the incident or received direction of illumination light and the local normal 29 of the lining surface 2. In this embodiment, the incident angle θ corresponding to each pixel or region on the lining surface 2 may be determined based on the local normal 29, the position of the illuminator, and the position of the short-wavelength infrared imaging unit 8. Here, the incident angle θ may be defined, for example, as the angle between the incident direction of illumination light and the local normal 29, or, if necessary, as the angle between the observation direction and the local normal 29, or a combination thereof. Then, correction may be performed to compensate for the angle dependence of the reflection intensity according to the incident angle θ.

[0122] In this embodiment, a correction coefficient Cλ(d,θ) may be used for each wavelength as the correction coefficient 39 that takes into account both the distance d and the incident angle θ. Here, λ represents the first wavelength band 32 or the second wavelength band 33. For example, the processing unit 16 may correct the reflected intensity Iλ after dark correction and white reference correction so that the corrected reflected intensity I'λ = Cλ(d,θ) × Iλ, or it may calculate the distance-dependent component and the angle-dependent component individually and apply them sequentially. The correction coefficient Cλ(d,θ) may be stored in the storage unit 17 as a table, function, approximation formula, interpolation map, or learning model based on calibration data obtained experimentally in advance. This makes it possible to appropriately compensate for the distance dependence or angle dependence that may differ between the first wavelength band 32 and the second wavelength band 33, respectively.

[0123] When setting the correction coefficient Cλ(d,θ), calibration may be performed using a reference plate 38 or a known reflective surface. For example, a reference plate 38 having a known reflectivity may be imaged at multiple distances and angles, and the response characteristics for each wavelength band may be obtained to derive the correction coefficient Cλ(d,θ). Alternatively, a reflective surface of known material and condition within the tunnel 1 may be used as a reference surface, and the correction coefficient may be updated to match the site conditions. Furthermore, the correction coefficient Cλ(d,θ) may be recalibrated at predetermined intervals to reflect the effects of changes in the device over time, changes in the output of the illuminator, fluctuations in the sensitivity of the image sensor, or the accumulation of dirt.

[0124] Furthermore, the processing unit 16 may evaluate the reliability of the correction process and identify pixels or regions where the correction accuracy is insufficient. For example, if the visibility of the reference plate 38 is insufficient, the accuracy of acquiring the distance d is low, the estimation error of the local normal 29 is large, or the incident angle θ is outside a predetermined range, the confidence level 23 of the corrected data for that pixel or region may be set low, and it may be weighted down or excluded in the subsequent wet index calculation or classification process. This can suppress misjudgments caused by correction errors.

[0125] As described above, the correction process in this embodiment can suppress various fluctuation factors included in the two-wavelength reflection image by appropriately combining dark correction, white reference correction, illumination unevenness correction, distance correction, and angle correction. In particular, by applying a correction coefficient Cλ(d,θ) corresponding to the distance d and incident angle θ for each wavelength, it is possible to obtain reflection information that more stably reflects the wet state even when there is curvature of the lining surface 2, position fluctuations during driving, and fluctuations in illumination conditions.

[0126] <Local normal estimation and confidence evaluation> Next, with reference to Figure 9, the local normal estimation and reliability evaluation according to this embodiment will be described. In order to perform the angle correction described above with higher accuracy, it is desirable to appropriately understand the orientation of the surface at each position of the lining surface 2. Therefore, in this embodiment, the local normal 29 is estimated based on the three-dimensional point cloud 25 or the shape information of the lining surface 2, and the incident angle θ or observation angle is determined using the local normal 29 to refine the angle correction.

[0127] The local normal 29 can be estimated, for example, using a three-dimensional point cloud 25 acquired by LiDAR 14. Specifically, the processing unit 16 may extract neighboring point clouds corresponding to the target pixel or target region on the lining surface coordinate system 21, and obtain the local normal 29 at that location by applying local plane approximation, least squares plane fitting, principal component analysis, or other approximation methods to the neighboring point clouds. Furthermore, the selection range of the neighboring point clouds may be set variably according to the spatial resolution, point cloud density, or curvature of the lining surface 2.

[0128] If a sufficient three-dimensional point cloud 25 is not obtained, or if the point cloud density is low, the local normal 29 may be estimated based on a cross-sectional model of the tunnel 1, a distance image, geometric information obtained during reprojection, or a known lining surface shape. For example, theoretical normals at each location may be determined based on a circular, horseshoe, or rectangular cross-sectional model and used as the local normal 29, or the local normal 29 may be calculated based on the gradient of the distance distribution obtained from the rangefinder 15 or LiDAR 14. Furthermore, the normals derived from the point cloud and the normals derived from the model may be integrated and the more reliable one may be used first.

[0129] The processing unit 16 determines the geometric relationship between the incident direction from the first illuminator 6 or the second illuminator 7 toward the target position and the observation direction from the short-wavelength infrared imaging unit 8 toward the target position, based on the estimated local normal 29, and performs angle correction based on these. For example, the incident angle θ in the correction coefficient Cλ(d,θ) may be the angle between the local normal 29 and the incident direction, or a correction coefficient that takes into account the angle between the local normal 29 and the observation direction may be used as necessary. This makes it possible to perform more precise angle correction that reflects local irregularities, repair shapes, or construction errors compared to correction based only on a simple cross-sectional model.

[0130] In this embodiment, the confidence level 23 for each pixel or region is calculated based on the estimation result of the local normal 29 and the image matching result. The confidence level 23 may be determined based on, for example, alignment error, reprojection residual, local normal estimation error, neighboring point cloud variance, point cloud density, acquisition time difference, distance difference, orientation difference, and other geometric consistency indicators. Furthermore, the confidence level 23 may be calculated based on indicators based on the image data itself, for example, the presence or absence of saturated pixels, the proportion of dark area pixels, insufficient reflectivity, signal-to-noise ratio, edge loss, occlusion, shading, or the degree of illumination unevenness.

[0131] For example, the processing unit 16 may determine that saturation is highly likely if the brightness of a corresponding pixel in both the first reflection image 19 and the second reflection image 20 is above a predetermined upper limit, and may reduce the reliability 23 of that pixel. Also, if the brightness in both or one of the reflection images is below a predetermined lower limit, or if sufficient reflection intensity cannot be obtained, it may determine that it is a dark area or has insufficient reflection, and similarly reduce the reliability 23. Furthermore, the reliability 23 of the pixel or region may be set low if the neighboring point cloud used to estimate the local normal 29 is sparse, if the residual of the approximation plane is large, or if the alignment residual between images is large.

[0132] The confidence score 23 may be calculated on a pixel-by-pixel, mesh-by-mesh, region-by-region, or ring-by-ring basis. The processing unit 16 may exclude pixels or regions with a confidence score 23 below a predetermined threshold from the calculation of the wetness index 22, or it may reduce the contribution of the wetness index 22 to such pixels or regions. For example, when calculating the wetness index 22, the confidence score 23 may be used as a weighting coefficient, so that pixels with a lower confidence score 23 have less influence on the judgment result. Furthermore, when outputting to the unfolded diagram 24 or the three-dimensional point cloud 25, the confidence score 23 may be included, or low-confidence regions may be displayed separately.

[0133] Furthermore, the processing unit 16 may control the subsequent classification process or the calculation of the priority confirmation order 30 based on the confidence level 23. For example, even if the wetness index 22 is high, areas with a low confidence level 23 may be classified as "requires reconfirmation" or "decision pending," or they may be extracted as targets to prompt additional imaging or detailed confirmation during on-site inspection. On the other hand, areas with a high wetness index 22 and a high confidence level 23 may be given a higher priority confirmation order 30.

[0134] Thus, according to the local normal estimation and reliability evaluation in this embodiment, local normals 29 can be estimated based on the three-dimensional point cloud 25 or shape information, and angle correction can be refined using the local normals 29. Furthermore, the reliability 23 can be calculated based on alignment error, local normal estimation error, saturation, dark areas, insufficient reflection, etc., and the reliability of the wetness index 22 and subsequent judgment results can be improved by excluding or reducing the weight of pixels or regions with low reliability 23.

[0135] <Moisture Index Calculation Process> Next, the wetness index calculation process according to this embodiment will be described with reference to Figures 8 to 10. In this embodiment, a wetness index 22 representing the wetness state of the lining surface 2 is calculated based on the first reflection image 19 and the second reflection image 20, which have undergone the reprojection correction and various correction processes described above. Here, the first reflection image 19 is a reflection image corresponding to the first wavelength band 32, in which water absorption is relatively high, and the second reflection image 20 is a reflection image corresponding to the second wavelength band 33, in which water absorption is relatively low. Therefore, by utilizing the difference in reflection intensity between the two images, an index that reflects the degree of wetness on the lining surface 2 can be obtained.

[0136] The determination unit 36 ​​acquires the reflection intensity for each wavelength band for each corresponding pixel or region in the corrected first reflection image 19 and second reflection image 20. For the sake of explanation, the reflection intensity of the corrected first reflection image 19 will be referred to as I1, and the reflection intensity of the corrected second reflection image 20 will be referred to as I2. However, these reflection intensities are not limited to the pixel values ​​themselves, but may be the average value, median value, representative value, or smoothed value within a predetermined region. The determination unit 36 ​​may also use reflection intensities weighted according to the confidence level 23.

[0137] As an example of a wetness index 22, the determination unit 36 ​​may calculate an index based on a difference. For example, if the reflectance intensity of the first wavelength band 32 tends to decrease relatively when wet, the difference index may be defined as D = I2 - I1. Alternatively, D = I1 - I2 may be used depending on the application. In other words, in the present invention, the signs or order can be appropriately set so that the index value increases with increasing wetness, and the specific definition of the difference index is not limited thereto. The difference index has the advantage of being easy to calculate and easily emphasizing local wetness changes.

[0138] Furthermore, as another example of a humidity index 22, the determination unit 36 ​​may calculate an index based on a ratio. For example, the ratio index may be defined as R = I2 / I1 or R = I1 / I2. The ratio index makes it easier to evaluate the relative relationship between two wavelengths while relatively reducing the influence of overall fluctuations in illumination intensity or uniform fluctuations in reflectance. In addition, to avoid numerical instability when the denominator is zero or extremely small, the determination unit 36 ​​may use R = (I2 + ε) / (I1 + ε) or a similar formula, which adds a small positive number ε to the denominator.

[0139] Furthermore, the determination unit 36 ​​may use an index based on normalized differences. For example, a normalized difference index may be defined as N = (I2 - I1) / (I2 + I1 + ε). With such a normalized difference index, the difference component can be normalized by the total reflectance intensity, making it easier to extract relative changes originating from the humid state while suppressing the influence of local brightness differences. Alternatively, the determination unit 36 ​​may use an index based on logarithmic ratios, for example, L = log((I2 + ε) / (I1 + ε)). By using logarithmic ratios, the dynamic range of the ratio can be compressed, making it easier to suppress the influence of extreme values.

[0140] In this embodiment, only a single calculation result may be used as the wetness index 22, but a primary index and a secondary index may be used in combination. For example, a normalized difference index or a ratio index may be adopted as the primary index, and a difference index, a log-ratio index, absolute reflectance intensity for each wavelength band, confidence 23, local contrast, continuity of the region, consistency with surrounding pixels, and other features may be used as secondary indexes. The determination unit 36 ​​may then integrate these primary and secondary indexes by linear combination, nonlinear combination, weighted averaging, rule-based integration, or other methods to generate the final wetness index 22 or wetness score.

[0141] Furthermore, the determination unit 36 ​​may perform spatial smoothing, noise reduction, or region consolidation processing when calculating the wetness index 22. For example, the wetness index 22 calculated for each corresponding pixel may be subjected to neighbor averaging, median processing, edge-preserving smoothing, or other spatial processing to reduce the influence of isolated noise. In addition, a representative value of the wetness index 22 may be obtained for each connected region consisting of multiple pixels, and a wetness candidate region 28 may be extracted based on this representative value. This suppresses false detections caused by variations at the single-pixel level and makes it possible to obtain candidate regions that are easy to handle in practice.

[0142] The determination unit 36 ​​extracts candidate wet areas 28 based on the wetness index 22. As an example of the extraction method, the determination unit 36 ​​may determine pixels or areas where the wetness index 22 exceeds a predetermined threshold as wet candidates. The threshold may be a fixed value, or it may be set variably based on imaging conditions, cross-sectional position, confidence level 23, conditions of the target tunnel, or dry reference area 27. Furthermore, the determination is not limited to binary determination using a single threshold, but may be performed using multiple thresholds to classify areas into strongly wet candidates, weakly wet candidates, areas requiring reconfirmation, etc.

[0143] Furthermore, the determination unit 36 ​​may extract wet candidate regions 28 using a trained classifier instead of, or in combination with, threshold determination. The trained classifier may be configured to output whether a target pixel or target region is a wet candidate, or to which category the degree of wetness belongs, using, for example, a principal index, a secondary index, a confidence score 23, a local normal 29, a positional attribute, a region shape feature, a time-series change feature, or other input features. The trained classifier can be, but is not limited to, a decision tree, a support vector machine, a neural network, a probabilistic model, or various other determination models.

[0144] Furthermore, the determination unit 36 ​​may calculate a wetness score as a continuous value based on the primary and auxiliary indicators, and extract wetness candidate regions 28 based on the wetness score. In this case, the wetness score can also be used for the relative evaluation of the ring 26 units in the subsequent stage, the calculation of the priority confirmation order 30, or the superimposed display on the unfolded diagram 24. In addition, pixels or regions with a low confidence level 23 may be excluded from the calculation of the wetness score or their weight may be reduced before candidate extraction.

[0145] As described above, according to the wetness index calculation process in this embodiment, a wetness index 22 can be calculated from the corrected first reflection image 19 and second reflection image 20 by calculations such as difference, ratio, normalized difference, logarithmic ratio, etc. Furthermore, by using the primary index and auxiliary index in combination and extracting wetness candidate regions 28 using a threshold or a trained classifier, flexible and practical wetness determination can be achieved according to the conditions of the target tunnel or the purpose of inspection.

[0146] <Relative determination by ring or segment unit> Next, with reference to Figure 10, the relative determination of rings or segments according to this embodiment will be explained. In this embodiment, not only are candidate wet areas 28 of the lining surface 2 extracted using the wetness index 22 described above, but the relative degree of wetness is also evaluated for each ring 26 or segment of the tunnel 1. This makes it possible to extract areas with relatively high wetness within the ring 26 or segment as priority areas for confirmation, while suppressing the influence of differences in material, surface condition, and lighting conditions for each target tunnel.

[0147] Here, the ring 26 or segment is a structural or management unit of the tunnel 1, and may be, for example, a ring and segment in a shield tunnel, a lining block unit, a management section at fixed distances, or a similar division unit. The processing unit 16 may divide the lining surface 2 into ring 26 or segment units based on design drawing information, distance information, joint position information, three-dimensional point cloud 25, development drawing 24, or image features. This makes it possible to individually evaluate the distribution of the wetness index 22 for each unit.

[0148] In this embodiment, a dry reference region 27 is automatically extracted for each ring 26 or each segment. The dry reference region 27 refers to a region within the ring 26 or segment that is relatively dry and suitable for use as a reference. For example, the processing unit 16 may extract regions with a low humidity index 22, a high reliability 23, and stable reflectance as candidates, and determine the dry reference region 27 from among them. Furthermore, the dry reference region 27 is not limited to a single continuous region, but may be defined as a set of multiple discrete regions.

[0149] When automatically extracting the drying standard region 27, exclusion processing may be performed to avoid selecting regions that are unsuitable as standards due to material differences, efflorescence, dirt, repair marks, or other factors. For example, the processing unit 16 may select drying standard candidates after excluding regions suspected of efflorescence, dirt, repair material, attached material, or low-reliability regions based on the visible image, shape features, reflectance intensity distribution, local texture, region continuity, or existing deformation judgment results. In addition, extremely dark regions, saturated regions, shaded regions, or regions with insufficient point cloud density may also be excluded from the candidates for the drying standard region 27.

[0150] Once the dry reference region 27 is determined, the processing unit 16 calculates a representative value of the moisture index 22 in the dry reference region 27 as the reference value. The representative value can be the mean, median, mode, lower predetermined percentile, or a statistic after outlier removal. For example, the processing unit 16 may use the median of the moisture index 22 of the pixels belonging to the dry reference region 27 as the reference value, or it may use the lower 20th percentile as the reference value. This makes it possible to obtain a stable reference value while suppressing the influence of minute outliers or local abnormalities remaining within the dry reference region 27.

[0151] Next, the processing unit 16 normalizes the moisture index 22 for each pixel, each region, or each candidate region as a relative value with respect to the reference value. For example, the normalized relative moisture index may be defined as the difference, ratio, standard score, percentage, rank value, or a quantity equivalent thereto between the moisture index 22 and the reference value. Furthermore, the normalization amount may be corrected using the distribution width, standard deviation, interquartile range, or other variability indicators for each ring 26 or segment. This makes it possible to relatively absorb differences in material, reflectivity, lighting conditions, or aging conditions that may differ for each ring 26 even within the same tunnel.

[0152] By performing such relative normalization, it becomes easier to extract regions with relatively high humidity within the ring 26 or segment, even if there are differences in reflectance due to repair marks, apparent brightness differences due to efflorescence, localized density differences due to dirt, or absolute value variations due to material differences. In other words, in this embodiment, instead of directly comparing only the humidity index 22 as an absolute value, a relative evaluation with respect to the dry reference region 27 can be performed, thereby increasing robustness against disturbance factors other than deformation.

[0153] Furthermore, the processing unit 16 may generate a relative humidity ranking for each region within the ring 26 or segment based on the relative humidity index. For example, it may assign ranks in descending order of relative humidity index and extract regions belonging to the top predetermined number or top predetermined percentage as priority verification candidates. The relative humidity ranking may also be weighted in combination with the extent, continuity, reliability 23, or historical change information of the candidate humidity regions 28. This makes it possible to prioritize regions that are practically meaningful, rather than based on outliers of single pixels.

[0154] Furthermore, in this embodiment, the relative judgment is not limited to the ring 26 units, but the same normalization and ranking processes may be applied to segment units, lining block units, cross-sectional zone units, or management units at fixed distances. For example, separate dry reference areas 27 may be extracted for the top surface, shoulders, and side walls, and relative evaluation may be performed for each. This allows for a more detailed consideration of the differences in reflection conditions for each position within the cross-section.

[0155] Thus, by using relative determination on a ring or segment basis in this embodiment, a dryness reference area 27 can be automatically extracted for each ring 26 or segment, and the moisture index 22 can be normalized as a relative value to that reference. This makes it possible to generate a relative moisture ranking that is less affected by material differences, efflorescence, dirt, repair marks, etc., and enables efficient extraction of candidate moisture areas 28 that inspectors should focus on checking.

[0156] <Classification Processing> Next, the classification process according to this embodiment will be described with reference to Figures 10 to 13. In this embodiment, not only are wet candidate areas 28 extracted based on the wetness index 22 and relative wetness ranking described above, but the wet candidate areas 28 or target areas on the lining surface 2 are also classified according to their properties or inspection priority. This allows inspectors to not only identify areas that are potentially wet, but also to efficiently extract areas that should be inspected in detail, while distinguishing them from apparent changes caused by dirt, efflorescence, repair marks, and other factors.

[0157] In this embodiment, the "target area" to be classified may be set on a pixel basis, a mesh basis, a connected area basis, a ring 26 basis, a segment basis, or a similar basis. For example, the determination unit 36 ​​may group together pixel groups whose wetness index 22 satisfies predetermined conditions as connected components and classify these connected components as a single target area. Alternatively, the wetness index 22, reliability 23, and positional attributes may be aggregated for each management mesh on the lining development drawing 24 and classified on a mesh basis.

[0158] The determination unit 36 ​​uses at least a moisture index 22 when classifying. Furthermore, if necessary, at least one of shape features, positional attributes, history information, and confidence level 23 may be used in combination. As shape features, for example, the area of ​​the target region, major axis, minor axis, aspect ratio, perimeter, continuity, slenderness, presence or absence of branching, contour roughness, texture, density distribution, contrast with the surroundings, and other feature quantities can be used. As positional attributes, for example, distance, ring number, segment number, whether it is the top, shoulder, or side wall, whether it is near a joint, whether it is near a water guide member, whether it is close to a location where deformation has been recorded in the past, etc.

[0159] Furthermore, historical information that can be used includes the moisture index 22 obtained during past inspections, past classification results, repair history, deformation history, trend of expansion or contraction over time, frequency of occurrence, and other information. For example, if a consistently high moisture index 22 is observed at the same or nearby location, it may be given a higher importance than a one-off noise or temporary concentration change. On the other hand, if a temporarily high moisture index 22 is observed only this time and the confidence level 23 is low, it may be treated as requiring "reconfirmation".

[0160] In this embodiment, the determination unit 36 ​​may classify the target area into at least one of the following categories: wetness candidate, suspected staining or efflorescence, no abnormality, or requires reconfirmation. Here, "wetness candidate" is an area where the wetness index 22 is relatively high and it is highly likely to be caused by actual wetness or water leakage based on shape features, positional attributes, or historical information. "Suspected staining or efflorescence" is an area where apparent density changes or reflection changes exist, but it is highly likely to be caused by factors other than wetness based on the relative wetness ranking, shape features, or historical information. "No abnormality" is an area where the wetness index 22 is low and the abnormality is low based on auxiliary features. "Requires reconfirmation" is an area where either the wetness index 22 or an auxiliary feature is near the determination boundary, or where additional confirmation is desirable because the confidence level 23 is low.

[0161] As for specific methods of classification processing, the determination unit 36 ​​may perform rule-based classification based on predetermined determination rules. For example, if the relative wetness index is above the first threshold, the confidence level 23 is above the second threshold, and the target area has a certain degree of continuity, it may be determined to be a "wetness candidate". Also, if the wetness index 22 is moderate, but the shape is powdery, patchy, or accompanied by localized high-reflectance areas, it may be determined to be "suspected efflorescence". Furthermore, if the wetness index 22 is low, and there is little abnormality in any of the shape characteristics, location attributes, and history information, it may be determined to be "no abnormality".

[0162] Furthermore, the determination unit 36 ​​may perform classification using a trained classifier instead of, or in combination with, the rule-based classification. The trained classifier may be configured to take the wet index 22, principal index and auxiliary index, shape features, positional attributes, history information and confidence 23 as input and output which classification the target region belongs to. As the trained classifier, for example, a decision tree, random forest, support vector machine, neural network, probabilistic model and various other classification models can be used. This makes it possible to perform more flexible classification by utilizing combinations of complex features that are difficult to represent with simple threshold judgments.

[0163] Furthermore, in this embodiment, the classification results are not limited to outputting them as discrete classes, but may also be multi-class classification or score output. For example, the determination unit 36 ​​may output multiple scores for each target area, such as the likelihood of wetness, the likelihood of suspected efflorescence, the likelihood of staining, and the degree to which re-inspection is required. Alternatively, the wetness score, non-wetness factor score, and inspection priority score may be calculated individually and included in the lining development diagram 24, the three-dimensional point cloud 25, or the inspection result list 31. This allows the inspector to make a judgment based on the basis or likelihood of the determination, without relying on a single classification label.

[0164] Furthermore, the determination unit 36 ​​may combine the classification result and the confidence score 23 and use them to calculate the priority confirmation order 30 in a later stage. For example, target areas classified as "wet candidate" and with a high confidence score 23 may be assigned a high priority confirmation order 30, while target areas classified as "requires reconfirmation" may be extracted as candidates for on-site re-imaging or detailed visual confirmation. On the other hand, target areas classified as "stained or suspected efflorescence" may be output with a display mode or priority that is distinct from the wet candidate.

[0165] Thus, according to the classification process in this embodiment, the target area can be classified into categories such as wetness candidate, suspected dirt or efflorescence, no abnormality, or requires re-inspection, based on the wetness index 22, shape characteristics, positional attributes, and historical information. Furthermore, by using multi-level classification or score output, it is possible to provide inspectors with more practical and easily interpretable judgment results.

[0166] <Output Processing> Next, the output processing according to this embodiment will be described with reference to Figures 11 to 13. In this embodiment, the aforementioned wetness index 22, relative wetness ranking, classification result, reliability 23, and priority confirmation order 30, if necessary, are output in a format that is easy for inspectors to use. Specifically, the output unit 37 performs a two-layer output, first displaying this information superimposed on the lining development drawing 24, and then, if necessary, retaining it as attribute values ​​of the three-dimensional point cloud 25.

[0167] The lining unfolded diagram 24 is a diagram showing the unfolded lining surface 2 of the tunnel 1 on a two-dimensional plane. For example, it may be configured as an image with the distance direction and the cross-sectional direction as axes, a mesh image that allows for the identification of rings 26 or segments, or a similar two-dimensional representation. The output unit 37 may assign a color, shade, hatching, symbol, or label to each position on the lining unfolded diagram 24 according to the wetness index 22, classification result, or confidence level 23, and display the candidate wet areas 28 in a visually recognizable manner. This allows inspectors to grasp the wetness distribution throughout the entire tunnel 1 at a glance.

[0168] For example, the output unit 37 may assign a continuous color tone according to the size of the moisture index 22 and provide different display modes depending on the classification result, such as a candidate for moisture, suspected dirt or efflorescence, no abnormality, or requires reconfirmation. In addition, for areas with a low confidence level 23, the low accuracy of the judgment may be indicated by semi-transparent display, dashed line display, warning marking, or other methods. Furthermore, for areas with a high priority confirmation rank 30, a rank number, border emphasis, or a corresponding symbol with a list may be displayed.

[0169] Overlaying the data onto the lining development diagram 24 offers excellent advantages for time-series comparison and explanation. Specifically, by displaying multiple sets of wetness indicators 22 or classification results obtained at different times in parallel, as differences, as overlays, or as a time-series animation on the same or corresponding lining development diagram 24, it is easy to grasp the trend of expansion, contraction, movement, or recurrence of the wetness candidate area 28. Furthermore, since the correspondence between the lining development diagram 24 and distance, ring number, and position within the cross-section is clear, it is easy to use as an inspection report, maintenance management document, or on-site explanation document.

[0170] On the other hand, the output unit 37 may, if necessary, retain the wetness index 22, classification result, confidence level 23, or priority verification rank 30 as attribute values ​​of the three-dimensional point cloud 25. For example, a wetness index 22 or classification label corresponding to the location may be assigned to each point, each point cloud block, or each mesh of the three-dimensional point cloud 25, so that it can be displayed in color in three-dimensional space. In addition, if there is a visible image, thermal image, crack information, displacement information, repair history, or other attribute information corresponding to the same location, these may also be retained together.

[0171] The output on the three-dimensional point cloud 25 is excellent for spatial comparison with other deformations. For example, the spatial relationship between the candidate wet area 28 and cracks, efflorescence, cross-sectional defects, repair marks, the vicinity of attached structures, or joint locations can be directly compared in three-dimensional space. Furthermore, since the candidate wet area 28 can be confirmed while taking into account the cross-sectional shape of the tunnel 1, local irregularities, positional relationships, or interference with surrounding equipment, it is useful for estimating the cause, planning on-site inspections, or examining the scope of repairs.

[0172] In the dual-layer output of this embodiment, the lining diagram 24 and the three-dimensional point cloud 25 may be associated with each other. For example, if an inspector selects a specific wet candidate area 28 on the lining diagram 24, the corresponding position on the three-dimensional point cloud 25 may be highlighted, or conversely, the area on the lining diagram 24 corresponding to the position selected on the three-dimensional point cloud 25 may be displayed. This allows for complementary use of the overview of the two-dimensional display and the spatial understanding of the three-dimensional display.

[0173] Furthermore, the output unit 37 may output the inspection results list 31 in a table format, containing the distance, ring number, position within the cross-section, classification result, wetness index 22, reliability 23, and priority inspection ranking 30 for each of the wet candidate areas 28. In addition, the lining development drawing 24, the three-dimensional point cloud 25, and the inspection results list 31 may be linked to each other, so that information about an area selected in one is reflected and displayed in the other. This allows the inspector to simultaneously obtain overview, positional understanding, and ease of confirming the basis for judgment.

[0174] Thus, the output processing in this embodiment enables a two-layer output that first overlays the wetness index 22 and classification results onto the lining development drawing 24, and then, if necessary, retains them as attribute values ​​of the three-dimensional point cloud 25. Therefore, the lining development drawing 24 can display information that is excellent for time-series comparisons and explanations, while the three-dimensional point cloud 25 can display information that is excellent for spatial comparisons with other deformations, thereby comprehensively supporting inspection, recording, explanation, and maintenance management decisions.

[0175] <Priority Check Ranking> Next, with reference to Figure 13, the priority verification order according to this embodiment will be described. In this embodiment, a priority verification order 30 is assigned to each of the wet candidate areas 28 or areas that have been subjected to classification processing, indicating the order in which they should be checked preferentially during on-site inspection. This allows inspectors to sequentially perform on-site inspections starting with the areas with the highest verification priority, under limited time and working conditions.

[0176] The processing unit 16 or the determination unit 36 ​​uses at least one of the following when calculating the priority confirmation order 30: distance, ring number, position within the cross-section, extent of the area, size of the wetness index 22, confidence level 23, and time-series change. Distance is information indicating which extension of the tunnel 1 the target area is located at, and ring number is information indicating which ring 26 or which management section the target area belongs to. Position within the cross-section may include the top, shoulder, side wall, near joints, near water guide members, and other positional attributes. Extent of the area may be expressed as area, length, perimeter, continuity, or diffusion range.

[0177] For example, the processing unit 16 may assign higher priority to areas with a larger wetness index 22, and higher priority to areas with a larger area spread. It may also assign higher priority to areas with a higher reliability 23, lower priority to areas with a lower reliability 23, or manage them separately as "requires reconfirmation." Furthermore, as a time-series change, higher priority may be assigned to areas with a large increase in the wetness index 22, an expanding trend, a high recurrence frequency, or a long duration compared to past inspection results. This allows for ranking based not only on the current wetness state, but also on the potential for future deterioration or its importance from a management perspective.

[0178] Furthermore, priority may be adjusted according to the position within the cross-section. For example, a candidate wet area 28 near the top or shoulder may be given a higher priority than an area with a similar wetness index 22 at the bottom of the side wall, as it may require focused inspection in relation to leakage paths, member deterioration, or secondary deformation. In addition, weights may be assigned to areas located at joint locations, known repair locations, locations where abnormalities have been detected in the past, near equipment mounting parts, or near drainage-related members, according to their location attributes.

[0179] The method for calculating the priority confirmation ranking 30 is not particularly limited, but for example, the processing unit 16 may quantify each of the elements and calculate an overall priority score by integrating them with predetermined weights. For example, the overall score may be obtained using the wetness index 22, confidence level 23, area, time-series change amount, and location attribute weights, and rankings may be assigned in descending order of score. Alternatively, instead of continuous ranking by overall score, the areas may be divided into multiple stages such as emergency confirmation, priority confirmation, normal confirmation, and observation. Furthermore, areas that meet predetermined conditions may be automatically classified into the highest priority group using a rule-based system.

[0180] The processing unit 16 may output the calculated priority confirmation order 30 to the lining development drawing 24, the three-dimensional point cloud 25, or the inspection results list 31. For example, the higher-ranking wet candidate areas 28 may be assigned a rank number on the lining development drawing 24, and they may be displayed in a color or symbol corresponding to their priority on the three-dimensional point cloud 25. In addition, the inspection results list 31 may list the distance, ring number, position within the cross-section, wetness index 22, reliability 23, classification result, and priority confirmation order 30 in association with each other. This makes it easier for inspectors to understand the location and priority of the items to be checked before going to the site.

[0181] Furthermore, the priority verification order 30 may be used for planning inspections or guiding during on-site inspections. For example, a verification route may be generated in order from the highest-priority areas, and the verification order may be presented to the inspector, or only high-priority areas may be extracted and displayed according to the range that can be verified within a certain time. In addition, if the on-site inspection confirms that the area is not actually wet, or if additional abnormalities are confirmed, the results may be fed back to update the ranking conditions, weights, or trained classifiers for subsequent inspections.

[0182] Thus, according to the priority verification ranking in this embodiment, a priority verification ranking 30 can be assigned based on distance, ring number, position within the cross-section, area extent, size of the wetness index 22, reliability 23, and time-series changes. Therefore, it is possible to clearly indicate the areas that should be checked as a priority during on-site inspections, contributing to increased efficiency in inspection work, reduced oversights, and faster maintenance decisions.

[0183] <Variation> Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and various modifications, substitutions, additions, or omissions are possible without departing from the spirit of the invention. Representative modifications of the present invention will be described below.

[0184] As a first modification, the specific wavelength settings for the first wavelength band 32 and the second wavelength band 33 are not limited to the vicinity of 1450 nm and the vicinity of 1320 nm. That is, in the present invention, it is sufficient to extract the difference in reflection characteristics caused by the wet state of the lining surface 2 by using at least two short-wavelength infrared wavelength bands, one wavelength band in which water absorption is relatively high and another wavelength band in which water absorption is relatively lower. Therefore, the center wavelength, bandwidth, or combination of each of the above wavelength bands may be appropriately changed depending on the material of the target tunnel, lighting conditions, sensitivity characteristics of the image sensor, required accuracy, or design conditions. Furthermore, it is not limited to two wavelengths, but three or more wavelength bands may be used, and the wetness index 22 may be calculated based on a combination of at least two of these wavelength bands.

[0185] As a second modification, the configuration of the short-wavelength infrared imaging unit 8 is not limited to a configuration using a single camera. For example, a multi-camera configuration may be used, in which a first imaging unit that acquires an image corresponding to the first wavelength band 32 and a second imaging unit that acquires an image corresponding to the second wavelength band 33 are separately provided. Alternatively, multiple imaging units for the same wavelength band may be distributed in the cross-sectional direction or the travel direction to achieve wide-field imaging, blind spot reduction, or redundant measurement. In this case, the correspondence between images can be corrected based on the internal parameters, external parameters, and relative positional relationships of each imaging unit, and the aforementioned reprojection correction or pseudo-simultaneous correspondence can be applied. Therefore, the present invention is not limited to a single-camera configuration, but is also applicable to a multi-camera configuration or a composite imaging system.

[0186] As a third variation, the output format of the output unit 37 is not limited to a two-layer output using both the lining development drawing 24 and the three-dimensional point cloud 25. For example, if the main purpose is to provide inspection reports, time-series comparisons, or link with management ledgers, the wetness index 22, classification results, and priority confirmation order 30 may be superimposed on the lining development drawing 24 only. On the other hand, if the main purpose is to understand spatial relationships, make three-dimensional comparisons with other deformations, or support on-site inspections, the data may be retained only as attribute values ​​of the three-dimensional point cloud 25 and displayed in three dimensions. Furthermore, only the inspection results list 31 may be output, and it may be possible to reconstruct it later into the lining development drawing 24 or the three-dimensional point cloud 25 as needed.

[0187] As a fourth modification, the wetness detection device 4 of the present invention is not limited to a configuration that uses only short-wavelength infrared images, but may be used in combination with visible images, thermal infrared images, or other types of images. For example, visible images may be used to extract candidates for dirt, efflorescence, cracks, repair marks, or deposits, and the classification accuracy may be improved by combining them with a wetness index 22 based on short-wavelength infrared images. Alternatively, thermal infrared images may be used to grasp temperature anomalies or thermal behavior originating from moisture, and this may be compared with the wetness determination based on short-wavelength infrared images. Furthermore, visible images, short-wavelength infrared images, and thermal infrared images may be integrated onto a common lining surface coordinate system 21 or a three-dimensional point cloud 25, and wetness candidate regions 28, classification results, or priority confirmation order 30 may be determined based on information from multiple modalities.

[0188] Thus, the present invention is not limited to the above-described embodiments in terms of the specific setting of the two wavelengths, the number and arrangement of imaging units, the output format, and the type of sensor or image used in combination, but can be appropriately modified according to the conditions of the target tunnel, existing equipment, inspection purpose, or operating mode. Therefore, the present invention is applicable to a wide range of embodiments, including the above-described modifications.

[0189] <Examples of implementation or evaluation> The following describes examples or evaluations of the present invention. Note that the following examples or evaluations are illustrative for explaining the effects and benefits of the present invention, and do not limit the present invention to these examples.

[0190] First, an example using a simulated tunnel will be described. In this example, a test specimen or simulated lining surface simulating the tunnel cross-section was prepared, and multiple evaluation target areas, including dry areas, wet areas, efflorescence areas, soiled areas, and repaired areas, were set on the simulated lining surface. The aforementioned wetness detection device 4 was mounted on the mobile body 3, and while traveling through the simulated tunnel at a predetermined speed, a first reflection image 19, a second reflection image 20, position information, attitude information, distance information, and, if necessary, a three-dimensional point cloud 25 were acquired.

[0191] In this embodiment, a wavelength band near 1450 nm was used as the first wavelength band 32, and a wavelength band near 1320 nm was used as the second wavelength band 33. An area-scan type short-wavelength infrared camera was used as the short-wavelength infrared imaging unit 8. A first reflection image 19 and a second reflection image 20 were acquired using a time-division illumination method or a similar acquisition method. Each image was assigned frame number, acquisition time, distance, and wavelength band identification information, and used for subsequent pseudo-simultaneous mapping and reprojection correction.

[0192] Upon examining the acquired uncorrected first reflection image 19 and second reflection image 20, it was found that in addition to the reflection difference corresponding to the wet areas, the images included illumination unevenness due to differences in position within the cross-section, intensity fluctuations due to the difference in distance between the moving body 3 and the simulated lining surface, and apparent density changes due to differences in the angle of incidence. Therefore, the processing unit 16 performed the aforementioned dark correction, white reference correction, illumination unevenness correction, distance correction, and angle correction, and further performed reprojection correction to the lining surface coordinate system 21 as necessary. As a result, in the corrected images, density variations due to non-wet factors were reduced compared to the uncorrected images, and the reflection difference corresponding to the wet areas became relatively clearer.

[0193] Next, the processing unit 16 calculated a wetness index 22 from the corrected first reflection image 19 and second reflection image 20. As the wetness index 22, an index based on difference, ratio, normalized difference, or logarithmic ratio can be used, but in this embodiment, a wetness score was generated by combining at least one primary index and, if necessary, an auxiliary index. As a result, a relatively high wetness index 22 was obtained in the wet areas compared to the dry areas, and different index distributions or auxiliary features were observed in the efflorescence areas, soiled areas, or repaired areas compared to the wet areas.

[0194] Furthermore, the simulated lining surface was divided into regions corresponding to rings 26 or control units, and a dry reference region 27 was automatically extracted within each region. The wetness index 22 was then normalized as a relative value to the dry reference region 27. As a result, it was confirmed that even when there are differences in absolute reflectance due to material differences, efflorescence, dirt, or repair marks, areas with relatively high wetness can be stably extracted within each region. This made it possible to generate a relative wetness ranking for each ring 26 or control unit.

[0195] Furthermore, the processing unit 16 classified each target area into a wetness candidate, suspected staining or efflorescence, no abnormality, or requiring reconfirmation, based on the wetness index 22, shape characteristics, positional attributes, history information, and confidence level 23. As a result, areas that were actually wetted were extracted as wetness candidates, and areas with a low probability of being wet among the efflorescence or staining areas could be distinguished and displayed as a separate classification from the wetness candidates. In addition, areas with low judgment accuracy due to insufficient reflection, shading, or poor alignment could be extracted as requiring reconfirmation.

[0196] Next, the obtained wetness index 22 and classification results were superimposed onto the lining development diagram 24 as shown in Figure 11, and further assigned as attribute values ​​to the three-dimensional point cloud 25 as shown in Figure 12. On the lining development diagram 24, it was easy to grasp the distribution of wetness candidate areas 28 along the distance direction and the cross-sectional direction at a glance, and it was easy to compare over time or confirm reproducibility when the same simulated tunnel was measured repeatedly. On the other hand, it was confirmed that it was easy to grasp the spatial relationship between the wetness candidate areas 28 and the cross-sectional shape, joint location, unevenness, or other deformation candidates on the three-dimensional point cloud 25.

[0197] Furthermore, as shown in Figure 13, each wet candidate area 28 was assigned a priority confirmation ranking 30 based on distance, ring number, position within the cross-section, extent of the area, size of the wetness index 22, reliability 23, and, if necessary, time-series changes. As a result, it was confirmed that wet candidate areas 28 near the top surface, near joints, or with a large extent were assigned a high priority, clearly indicating the areas that inspectors should prioritize checking on-site.

[0198] Next, an example of evaluation regarding pseudo-simultaneous measurement and alignment accuracy will be described. In this evaluation example, a reprojection correction based on the travel direction, acquisition time, distance, attitude change, and, if necessary, a three-dimensional point cloud 25 was applied to the first reflected image 19 and the second reflected image 20 acquired by the time-division illumination method, and the positional shift between the images was compared before and after the correction. As a result, it was confirmed that by performing pseudo-simultaneous correspondence and reprojection correction, the correspondence shift between images was reduced compared to before the correction, and the spatial consistency of the wetness index 22 was improved. If necessary, the alignment error may be evaluated by feature point correspondence, phase correlation, residual evaluation, or comparison with known markers.

[0199] Furthermore, as an example of evaluating reproducibility, measurements were taken by driving through the same simulated tunnel or the same target section multiple times, and the wet candidate areas 28, relative wetness ranking, and priority confirmation rank 30 obtained in each measurement were compared. As a result, it was confirmed that by applying the correction processing, relative judgment processing, and classification processing according to the present invention, variations between measurement runs were suppressed, and the reproducibility of the position and rank of the wet candidate areas 28 was improved. Therefore, the present invention is easily applicable not only to single measurements but also to time-series monitoring or periodic inspections.

[0200] Next, an evaluation example targeting an actual tunnel will be described. In this evaluation example, the lining surface 2, which contains dry areas, wet areas including water leakage marks, efflorescence, and repair marks in an actual tunnel section, was targeted, and a wetness detection device 4 was mounted on an existing tunnel photography vehicle or inspection vehicle, and measurements were taken while the vehicle was in motion. The acquired first reflection image 19 and second reflection image 20 were sequentially subjected to the aforementioned pseudo-simultaneous mapping, reprojection correction to the lining surface coordinate system 21, various correction processes, wetness index calculation process, relative determination process, and classification process.

[0201] As a result, it was confirmed that even in actual tunnels, it was possible to extract wet candidate areas 28 that are difficult to distinguish using only visible images, and that efflorescence, dirt, or some repair marks could be displayed separately from the wet candidate areas. Furthermore, by superimposing the wet candidate areas 28 onto the lining development drawing 24, explanations corresponding to distances and ring numbers became easier, and by assigning attributes to the three-dimensional point cloud 25, spatial comparisons with information such as cracks, joints, shape abnormalities, and other information became easier. In addition, it was confirmed that by including the priority inspection order 30, it was possible to efficiently extract areas that should be prioritized for inspection during on-site inspections.

[0202] From the above examples and evaluations, it has been confirmed that, according to the present invention, the moisture state of the lining surface 2 can be grasped in a practically useful way through image comparison before and after correction, calculation of moisture index 22, superimposition on lining development diagram 24, assignment of attributes to three-dimensional point cloud 25, and assignment of priority confirmation order 30, in both simulated tunnels and actual tunnels. In particular, the present invention is not limited to the precise measurement of absolute moisture content, but is a technology suitable for the practical application of tunnel maintenance management because it provides relative moisture information useful for extracting key confirmation points. [Explanation of Symbols]

[0203] 1 Tunnel 2 Lining surface 3 Mobile Units 4. Wetness detection device 5 Image acquisition unit 6. First Illuminator 7. Second Illuminator 8. Short-wavelength infrared imaging unit 9. Filter switching section 10 Lighting Control Unit 11. Driving Information Acquisition Unit 12 DMI 13 IMU 14 LiDAR 15 Rangefinder 16 Processing Unit 17 Memory section 18 Display 19. First reflection image 20 Second reflection image 21. Covering surface coordinate system 22. Moisture index 23. Confidence 24. Lining Development Diagram 25 3D point cloud 26 rings 27 Dry reference area 28 Wet Candidate Regions 29 Local Normal 30 Priority Confirmation Ranking 31 Inspection Results List 32. First wavelength band 33. Second wavelength band 34 Simultaneous correspondence unit 35 Reprojection correction section 36 Judgment section 37 Output section 38 Reference plate 39 Correction Factor

Claims

1. A wetness detection device for detecting the wetness of a tunnel lining surface while moving relative to the tunnel lining surface, An image acquisition unit that irradiates the tunnel lining surface with light in the short-wavelength infrared region, or receives reflected light in the short-wavelength infrared region from the tunnel lining surface, and acquires a first reflection image and a second reflection image as a two-dimensional image, corresponding to at least a first wavelength band and a second wavelength band, which have relatively different water absorption characteristics. A driving information acquisition unit acquires driving information relating to at least one of the relative position and orientation between the image acquisition unit and the tunnel lining surface, the distance to the tunnel lining surface, and the shape of the tunnel lining surface. Processing unit and Equipped with, The aforementioned processing unit, Based on the aforementioned driving information, the first and second reflection images are pseudo-simultaneously associated with the same or nearby cross-sections of the tunnel lining surface. The corresponding first and second reflection images are reprojected onto the lining surface coordinate system to correct the pixel correspondence. Correction is performed using a correction coefficient Cλ(d,θ) set for each wavelength according to the distance d to the tunnel lining surface and the angle of incidence θ. The moisture index is normalized as a relative value to the dry reference area extracted for each ring or segment. The wet candidate regions based on the normalized wetness index are output to at least one of the lining development drawing and the three-dimensional point cloud. Wetness detection device.

2. A wet state detection device according to Claim 1, The first wavelength band is a wavelength band or a wavelength band near thereto in which water absorption is relatively large, and the second wavelength band is a reference wavelength band in which water absorption is relatively smaller than that of the first wavelength band. Wetness detection device.

3. A wet state detection device according to Claim 1, The image acquisition unit includes an image sensor capable of acquiring a two-dimensional image in the short-wavelength infrared region, and the image sensor is an InGaAs image sensor, a short-wavelength infrared image sensor, or an image sensor capable of acquiring reflection images in multiple wavelength bands. Wetness detection device.

4. A wet state detection device according to Claim 1, The image acquisition unit includes a plurality of illuminators or light distribution members, and the plurality of illuminators or light distribution members supply illumination light toward at least a portion or multiple portions of the top, shoulders and side walls of the tunnel lining surface. Wetness detection device.

5. A wet state detection device according to Claim 1, The image acquisition unit acquires images corresponding to the first wavelength band and the second wavelength band by time-division switching of illumination, switching of bandpass filters provided in the imaging optical path, or simultaneous measurement by branched optical paths. Wetness detection device.

6. A method for detecting the wetness of a tunnel lining surface while moving relative to the tunnel lining surface, An image acquisition step to acquire a first reflection image and a second reflection image corresponding to at least a first wavelength band and a second wavelength band, where the absorption characteristics of water are relatively different; A driving information acquisition step that acquires driving information including at least one of position information, attitude information, distance information, and shape information acquired in conjunction with the relative movement, A mapping step is performed to associate the first reflection image and the second reflection image with the same or nearby cross-sections of the tunnel lining surface based on the aforementioned driving information, A reprojection correction step is performed to correct pixel correspondence by reprojecting the associated first reflection image and second reflection image onto the lining surface coordinate system, A correction step is performed by correcting using a correction coefficient Cλ(d,θ) set for each wavelength according to the distance d to the tunnel lining surface and the incident angle θ, A normalization step that normalizes the wetness index as a relative value to the dryness reference area extracted for each ring or segment, Output process: Outputting candidate wet areas based on the normalized wetness index to at least one of the lining development drawing and the three-dimensional point cloud, A method for detecting a wet state, including the above.

7. A method for detecting a wet state according to claim 6, The correction step involves performing at least one of the following on the first and second reflection images: dark correction, white reference correction, illumination unevenness correction, distance correction according to the distance d to the tunnel lining surface, angle correction according to the incident angle or observation angle, and correction using a correction coefficient set for each wavelength. The reliability is calculated based on at least one of the local normal estimation error, alignment error, reflectance, saturation, dark areas, or point cloud density, and pixels or regions whose reliability does not meet a predetermined standard are excluded or reduced in weight. Method for detecting wetness.

8. A method for detecting a wet state according to claim 6, The normalization step extracts the dry reference area for each ring, segment, lining block, cross-sectional zone, or control section, normalizes the wetness index as a relative value to the dry reference area, and classifies the target area as a wetness candidate, suspected soiling or efflorescence, no abnormality, or requires reconfirmation based on at least one of the normalized wetness index, image features, shape features, positional attributes, historical information, and confidence level. Method for detecting wetness.

9. A method for detecting a wet state according to claim 8, The output step assigns the wetness index, classification result, or reliability to at least one of the lining development drawing and the three-dimensional point cloud, and assigns a priority verification order or verification priority based on at least one of the distance, ring number, position within the cross-section, continuity of the region, size of the wetness index, reliability, and time-series change. Method for detecting wetness.

10. A computer receives as input a first reflection image and a second reflection image, which are acquired while moving relative to the tunnel lining surface and which correspond to at least a first wavelength band and a second wavelength band with relatively different water absorption characteristics, and driving information which includes at least one of position information, attitude information, distance information and shape information, Based on the aforementioned driving information, a function is provided to associate the first and second reflection images with the same or nearby cross-sections of the tunnel lining surface in a pseudo-simultaneous manner. A function to correct pixel correspondence by reprojecting the associated first and second reflection images onto the lining surface coordinate system, The function includes correcting the distance d to the tunnel lining surface and the incident angle θ using a correction coefficient Cλ(d,θ) set for each wavelength, A function that normalizes the moisture index as a relative value to the dry reference area extracted for each ring or segment, A function to output candidate wet areas based on the normalized wetness index to at least one of the lining development drawing and the three-dimensional point cloud, A program that makes this possible.

Images