JP2015106237A - Method for detecting installation object from concrete surface image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a detection omission and an excessive detection and to improve efficiency of processing in a method for detecting an installation object such as a distance face plate from a concrete surface image by image processing.SOLUTION: A method for detecting a distance face plate from a concrete surface image includes: a step (S11) of obtaining installation information such as an installation position and an installation interval of the distance face plate; a step (S12) of setting a primary search region on a concrete surface image based on the installation information; steps (S13 to S14) of detecting number "0" from an image of the primary search region by a pattern matching method; a step (S15) of setting a secondary search region by narrowing down the primary search region based on a detection result of the number "0"; and steps (S16 to S17) of detecting the distance face plate from the image of the secondary search region by the pattern matching method.

Description

  The present invention relates to a method for detecting an installation from a concrete surface image, and in particular, a long civil engineering structure such as a distance plate, a lower bundle, a fluorescent lamp, a viaduct, etc. installed on a concrete lining surface of a railway or road tunnel. The present invention relates to a method for detecting an image of an installation object having a known installation position from a concrete surface image by image processing.

Inspection of railway and road tunnels is carried out regularly by the manager of the tunnel, but inspection of deformations such as cracks in the lining and water leakage that cause concrete fragments to fall, etc., is particularly careful. There must be. However, inspection of deformation is not easy. The area of the lining surface of a railway or road tunnel, which is a long civil engineering structure, is very large, and the number of deformations is large. Therefore, the work requires a lot of labor and time. Therefore, efficient tunnel inspection work is required.

  Japanese Patent Laid-Open No. 2002-288180 (Patent Document 1) discloses a database system that centrally manages data related to a tunnel on a computer. In this tunnel database system, it is possible to transmit data to a server from a plurality of terminals (computers) installed in a tunnel inspection site office, etc., and the server centrally manages data input from each terminal. At the same time, it is possible to display or print out the tunnel data in a single screen for each tunnel data (for example, a development view inside the tunnel, a photograph of the lining surface, and inspection / measurement data). .

  The construction of this tunnel database makes it possible to instantly read the position of the corresponding kite on the monitor screen of the terminal installed at the office of the tunnel inspection site or the like based on joint information, lattice information, and kilometer. The joint information indicates the position of joints between the blocks, the lattice information is lattice data obtained by subdividing the tunnel inner surface into a size of 1 m square on the developed view, and the block is a lining area in which concrete is placed. It is. For example, in the tunnel shown in the tunnel structure inspection record book shown in FIG. 6, the minimum block length is 5 m and the maximum is 10 m. Also, the kilometer is about 10 m apart.

In JP-A-2005-105682 (Patent Document 2), a database server 10 including a memory unit that records a plurality of inspection data related to tunnel health diagnosis and position data in the tunnel in association with each other, and the memory unit records the data. A tunnel data management system is disclosed that includes a fixed terminal 30 that can be viewed through a recording medium that can carry a plurality of inspection data, and a portable terminal 20 that can be connected to the fixed terminal via a communication line. . This tunnel data management system is characterized in that position data can be input from the mobile terminal 20 and that tunnel inspection data can be viewed on the mobile terminal 20 at a work site. The tunnel data at the inspection location can be obtained at the tunnel inspection site, and the tunnel data can be viewed at the inspection site, so that it is easy to find the inspection location.

By the way, according to the description of paragraph 0036 of the specification of Patent Document 2, the position data input from the portable terminal 20 is position information in the tunnel, and the relative distance between the kilometer in the tunnel and the data measurement position. This is a test corresponding position that is derived from various positional relationships. The inspection relevant position is derived by an inspector at the work site, and the derived position data is input to the portable terminal 20 by the inspector. Therefore, the position data does not accurately specify the inspection location, but is a level that can specify the range of the hammering inspection.

The concrete wall image of the tunnel wall surface is indispensable for finding the inspection location of the tunnel wall surface. By the way, deformations such as cracks existing on the concrete wall surface of the inspection location have various shapes and sizes. For this reason, it is difficult for a skilled inspector to accurately recognize the deformation visually in the concrete wall surface image. Therefore, an image of an installed object whose installation position is known, such as a distance sign plate, about a kilometer, a lower bundle, and a fluorescent lamp, is an important mark for finding an inspection location.

However, it is not easy to search for an image of about a kilometer such as a distance sign board in a concrete wall image. This is because the kilometer such as the distance nameplate is not always clearly visible. Therefore, the present applicant has developed a method for detecting a wall surface attachment such as a lower bundle or a fluorescent lamp from a concrete wall image of a tunnel by image processing as described in Patent Document 3 (Japanese Patent Laid-Open No. 2012-202858). That is, this is a method of detecting a wall surface object to be detected by shape-based pattern matching corresponding to a change in scale using a wall surface object template that is created in advance and stored in a memory. If this shape-based pattern matching method is used, it is possible to detect a distance mark plate or the like from a concrete wall image by image processing.

For example, when the installation object is a fluorescent lamp, the flow of image processing is as shown in FIG. First, the inspector displays a part of the concrete wall surface image of the tunnel that is considered to have the fluorescent lamp as a search area on the monitor screen of the image processing apparatus (S41). Next, the inspector reads the fluorescent lamp template from the memory of the image processing apparatus, and moves the template in the image of the search area, and the similarity to the image of the portion overlapping the template is transferred to the image processing apparatus. Processing for calculation, that is, pattern matching based on the shape is performed (S42), and a matching score is determined to detect a fluorescent lamp (S43).

The detection by the above-described image processing of the installed object does not use the brightness gradation, but uses the contour data defining the characteristics of the object and the gradient data of the density value in the normal direction. As a result, extremely robust search is realized even for images with concealment or disturbance. However, in the state where the wall of the civil engineering structure has dirt and soot, it is necessary to clean the wall surface and remove the dirt before acquiring the wall image. It is difficult to obtain a wall image that can clearly grasp the object to be detected. In addition, depending on the shooting conditions, the appearance size (scale) of the object is different each time. Because of such a problem, it is difficult to accurately detect the target installation object by image processing. Furthermore, since the target installation object has dropped out and is not necessarily installed at a predetermined position, searching for a limited installation location may cause erroneous detection. Furthermore, since the search area displayed on the monitor screen of the image processing apparatus is set with a margin in order to prevent detection omission, there is a problem that it takes time to search.

JP 2002-288180 A JP 2005-105682 A JP 2012-202859 A

A first problem to be solved by the present invention is to improve detection efficiency while preventing detection omission and excessive detection in a method for detecting an image of an installation object whose installation position is known from a concrete surface image. That is.
The second problem to be solved by the present invention is to improve detection efficiency while preventing detection omission and excessive detection in a method of detecting an image of a distance sign board from a concrete surface image.
The third problem to be solved by the present invention is to improve detection efficiency while preventing detection omission and excessive detection in a method of detecting an image of a lower bundle from a concrete surface image.
A fourth problem to be solved by the present invention is to improve detection efficiency while preventing detection omission and excessive detection in a method of detecting an image of a fluorescent lamp from a concrete surface image.

A method for detecting an installation object from a concrete surface image that solves the above problems includes a step of acquiring installation information such as an installation position and an installation interval of a target installation object, and a search area on the concrete surface image based on the installation information. And a step of processing a wall surface image of the search region and detecting a target installation object by a pattern matching method.

According to the present invention, an installation object whose installation position is already known can be efficiently detected from a concrete surface image while preventing detection omission and excessive detection.
In addition, according to the present invention, it is possible to reliably detect, even in the case of an object that is soiled and unclear, or an object that is dropped or displaced from the concrete surface image.
Furthermore, according to the present invention, since the position of the image of the installation object whose installation position is already known is clearly identified in the concrete surface image, it becomes easier for the inspector to quickly identify the deformation position of the concrete surface of the tunnel, The efficiency of tunnel health inspection was improved.

It is an example of the flowchart of the process which detects a distance sign board. It is an example of the concrete wall surface image containing the distance nameplate of a detection target. It is a partial enlarged image of one place with the distance nameplate of FIG. It is a partial enlarged image of another one place with the distance nameplate of FIG. It is an example of a template image used for “0” detection processing. It is an example of an output file as a result of the “0” detection process. It is an example of the output file of the final matching result of a distance mark nameplate. It is an example of the flowchart of the process which detects a lower bundle. It is an example of the flowchart of the process which detects a fluorescent lamp. It is an example of the flowchart of the conventional process which detects a fluorescent lamp.

The present invention is a method for detecting an image of an installation object from a concrete surface image using an image processing apparatus, the step of acquiring installation information such as an installation position and an installation interval of a target installation object, based on the installation information And setting a search area on the concrete surface image, and processing a wall surface image of the search area to detect a target installation object by a pattern matching method.

(Detection of distance nameplate)
With reference to the flowchart of FIG. 1, the process which detects the image of a distance nameplate from the concrete wall surface image of a tunnel is demonstrated. Since a plurality of distance nameplates are installed on the wall surface of the tunnel, a plurality of distance nameplate images exist in the tunnel wall surface image. Since the distance nameplates are detected one by one, for example, the distance nameplates installed on the tunnel wall surface are detected in order from one tunnel entrance to the other tunnel entrance.

First, the inspector accesses the tunnel database, acquires the installation information of the distance nameplate on the concrete wall (S11), and records it in the memory of the image processing apparatus. The installation information of the distance nameplate is recorded in the tunnel construction ledger, inspection ledger, etc., and is a part of the tunnel database, and includes at least data of type, number of installations, installation interval, and installation height. It is. In order to avoid complicating the description, in this embodiment, the distance nameplates installed on the tunnel wall are rectangular plates of 10 cm × 20 cm and 10 cm × 40 cm. It is assumed that the height is 5 m and the installation interval is 10 m.

  Subsequently, in the concrete wall image of the tunnel displayed on the monitor screen, the primary search area is set for the first distance nameplate to be detected based on the installation position and installation interval information (S12). . The concrete wall image of the tunnel displayed on the monitor screen is as shown in FIG. This concrete wall image has a plurality of images of distance marking plates, but they are so small that they cannot be recognized visually. Thus, FIGS. 3 and 4 show the distance nameplate image with “390” and the distance nameplate image with “559K400M” displayed in an enlarged manner.

The primary search area set for the concrete wall image in FIG. 2 shows the concrete wall image displayed vertically based on the installation information that the installation position of the distance nameplate is 1.5 m from the tunnel floor. The width is about one-eighth on the lower side, and the side becomes a long and narrow image area having the same width as the displayed concrete wall image. Since the installation position of the distance marking plate is acquired from the tunnel database and stored in the memory of the image processing device, the primary search area can be set even in exceptional tunnels that are not about 1.5 m, which is the standard installation height. Can be done appropriately.

Subsequent to step S12, the image processing apparatus performs a process of detecting the number “0” for the set primary search area according to the program (S13). This is based on the knowledge that the distance nameplates are installed at intervals of 10 m, and that the last digit of the number displayed on the distance nameplate is always zero. This detection process is a shape-based pattern matching for the primary search region using a number “0” template image as shown in FIG.

In step S13, when the image processing apparatus detects the numeral “0” according to the program, the image processing apparatus attaches a detection number to the position coordinate (X, Y) where the “0” is detected and the detection score in the memory as a detection result. Remember. The detection score is a score representing the degree of matching with the number “0” image of the template, and the closer to 1.0, the higher the accuracy.

Subsequent to step S13, the number “0” detection process is verified (S14). According to the program, the image processing apparatus adapts to the installation information that the installation height is 1.5 m from the tunnel floor and the installation interval is 10 m, and the distance marking plate whose detection score is close to 1 is a candidate. Further, the image processing apparatus reads all the detection result data stored in the memory according to the program and creates an output file of the zero detection result. A printout of this output file is shown in FIG. In FIG. 6, detection number 1, detection number 3, detection number 5, detection number 7, detection number 9, detection number 11, detection number 13, detection number 15, detection number 17, detection number 19, detection number 21, and detection Number 23 is a distance nameplate candidate.

By the way, the starting kilometer of the tunnel can be obtained from the tunnel database. For example, if the starting point of the tunnel is 500 k375 m, the first kilometer with an installation interval of 10 m is 500 k380 m, and therefore a distance sign plate displaying the number “380” appears first. Therefore, the distance nameplate first found from the left side of the image of FIG. 2 in the “0” detection process is a distance nameplate displaying the number “380”, and the memory of the image processing apparatus stores the first distance nameplate. Is "500k380m", and thereafter, the installation information of the distance nameplate in increments of 10m is stored in the memory. Accordingly, the image processing apparatus determines that the distance nameplates detected thereafter are “390 m”, “500 k400 m”, “410 m”,.

The detection of the distance nameplates of the fences that were originally installed at intervals of 10 m was “detected at a short interval within 10 m from the previous distance nameplate.”, “More than 10 m away from the previous distance nameplate. ”,“ Found at a position higher than normal ”,“ found at a position lower than normal ”, and the like can be detected from the pattern matching result of the“ 0 ”detection process in step S13. . Therefore, in the verification of the “0” detection process in step S14, the data detected at a short interval of 10 m or less is determined to be excessively detected from the intervals before and after that, and the data is excluded. If it is detected 10 m or more away, the search in step S13 is performed again at an image position of 10 m that should be found. Further, data detected at a clearly high position or low position is excluded as abnormal data.

Subsequent to step S14, the image processing apparatus sets a secondary search area according to the program (S15). The setting of the secondary search area is a process for further narrowing down the primary search area based on the installation position of the distance nameplate and the size of the distance nameplate to be detected. The secondary search area is set as a minimum necessary rectangular area in consideration of the size of the distance nameplate candidate with the coordinate position of the distance nameplate candidate determined to be probable in step S14 as the center. The minimum necessary rectangular area is, for example, about 1 m in length and width.

Subsequent to step S15, a distance nameplate detection process by pattern matching is performed (S16). That is, according to the program, the image processing apparatus reads the template of the distance sign plate from the memory, moves the template in the image of the secondary search area of the minimum necessary rectangular area, A pattern matching score is calculated by the image processing apparatus, and the distance nameplate is detected. In step S16, the image processing apparatus determines that the pattern matching score is close to 1 as a distance nameplate, and thereby detects the distance nameplate (S17). FIG. 7 is an example of an output file of the pattern matching result of the distance nameplate in step S16.

The template used in step S16 is a graphic template in which two bolt holes are arranged in the rectangle of the outline of the distance sign plate. Since the two bolt holes are for attaching the distance nameplate to the concrete wall surface, they are always holes in the distance nameplate and are observed together with the mounting bolts. When a template incorporating this feature is used, it is easier to make a similar comparison with an image of a portion overlapping the template than a template having only a rectangular outline. Therefore, the distance nameplate can be detected more quickly and reliably.

As described above, according to the method of the embodiment for detecting the distance nameplate from the concrete wall image of the tunnel according to the present invention, the distance search plate of the distance nameplate is obtained by pattern matching using the template displaying the number “0” as the primary search area. The present invention is characterized in that the location name plate is detected by pattern matching using a template in which the features of the distance name plate are included in the narrowed-down secondary search area. Although the pattern matching is performed twice, the processing efficiency and the detection accuracy are higher than those of the conventional method of detecting the distance nameplate by one pattern matching. The reason is that the search area is narrowed down properly, and the template used for pattern matching is the above-mentioned feature of the distance nameplate, that is, the characteristic arrangement of the numbers displayed on the distance nameplate, and for attachment This is because it reflects the characteristic shape that the bolt holes are always open.

By the way, if the distance nameplate to be detected is dropped or characters are hidden by soot or dirt, the distance nameplate may not be detected in the process of step S16. In such a case, an image of the distance nameplate is created, pasted at the estimated position, and restored. By performing this restoration work on the entire developed image of the concrete wall surface of the tunnel, a concrete wall developed image with a clear distance marking plate can be obtained. Therefore, it becomes easy for the inspector to quickly specify the deformed position of the concrete wall surface, and the efficiency of the tunnel soundness inspection can be improved.

(Under bundle detection)
A process for detecting the lower bundle installed on the tunnel wall surface will be described below with reference to the flowchart of FIG. First, the installation information regarding the installation position and installation interval of the lower bundle is acquired (S21). The installation position and installation interval of the lower bundle are information recorded in the tunnel database. Therefore, it is downloaded, recorded in advance in the memory of the image processing apparatus, and read out from the memory for acquisition. Alternatively, it may be acquired by accessing a tunnel database.

Subsequently, the inspector performs search area setting processing (S22). Common lower bundles are installed at intervals of 40 to 50 m, and are observed as rectangular regions elongated in the vertical direction around the crown of the tunnel. The inspector sets an image region for searching for a lower bundle of detection targets on the concrete wall image of the tunnel displayed on the monitor screen of the image processing device based on information such as the installation position and the installation interval as described above. .

Subsequently, a lower bundle detection process by pattern matching is performed (S23). That is, the image processing apparatus reads a lower bundle template according to a program, and performs matching with an image of a portion overlapping the template while moving the template in the image of the search area. The lower bundle template is an image of a typical lower bundle model, and is registered in the memory in advance. If the score of the matching result is close to 1, the image processing apparatus determines that the candidate is a lower bundle candidate, and stores the position coordinates (X, Y) and the detected score at which the lower bundle candidate is detected in the memory as a detection result. When the search of the set search area is completed, the process proceeds to verification of search results (S24).

In step S24, the image processing apparatus reads the data of the lower bundle candidate from the memory according to the program, and verifies whether the installation information regarding the installation position and the installation interval is satisfied. If the verification is passed, the lower bundle is detected (S25).

In the second embodiment, since the search area is set based on the installation information regarding the installation position and the installation interval of the lower bundle, the lower bundle candidates can be efficiently detected by pattern matching. Also, based on the installation information related to the installation position and installation interval of the lower bundle, the lower bundle candidates detected by pattern matching are verified to detect the lower bundle, thus preventing detection of the lower bundle and overdetection. I can do it now.

(Fluorescent light detection)
Processing for detecting a fluorescent lamp installed on the tunnel wall surface will be described below with reference to the flowchart of FIG. First, the installation information regarding the installation position and installation interval of the fluorescent lamp is acquired (S31). The installation position and the installation interval of the fluorescent lamp are information recorded in the tunnel database. Therefore, it is downloaded, recorded in advance in the memory of the image processing apparatus, and read out from the memory for acquisition. Alternatively, it may be acquired by accessing a tunnel database.

Subsequently, the inspector performs search area setting processing (S32). Common fluorescent lamps are installed at intervals of 10 m, and are observed as elongated regions in the horizontal direction around the lower part of the side wall of the tunnel. The inspector sets an image area for searching for the fluorescent lamp to be detected on the concrete wall image of the tunnel displayed on the monitor screen of the image processing apparatus based on the knowledge such as the installation position and the installation interval as described above.

Subsequently, a fluorescent lamp detection process is performed by pattern matching (S33). That is, the image processing apparatus reads a fluorescent lamp template according to a program, and performs matching with an image of a portion overlapping the template while moving the template in the image of the search area. The fluorescent lamp template is an image of a typical fluorescent lamp model, and is registered in the memory in advance. When the score of the matching result is close to 1, the image processing apparatus determines that the candidate is a fluorescent lamp candidate, and stores the position coordinates (X, Y) where the fluorescent lamp candidate is detected and the detection score in the memory as a detection result. When the search of the set search area is completed, the process proceeds to verification of search results (S34).

In step S <b> 34, the image processing apparatus reads fluorescent lamp candidate data from the memory according to the program, and verifies whether the information matches the installation information regarding the installation position and the installation interval. If the verification is passed, the fluorescent lamp is detected (S35).

In the third embodiment, since the search area is set based on the installation information regarding the installation position and the installation interval of the fluorescent lamp, the fluorescent lamp candidate can be efficiently detected by pattern matching. In addition, because fluorescent lamp candidates were detected by verifying fluorescent lamp candidates detected by pattern matching based on installation information related to fluorescent lamp installation positions and intervals, fluorescent lamp detection omissions and over-detection were prevented. I can do it now.

Claims (11)

A method for detecting an image of an installation object from a concrete surface image using an image processing apparatus, the step of obtaining installation information such as an installation position and an installation interval of the target installation object, a concrete surface image based on the installation information A method for detecting an installation object from a concrete surface image comprising: setting a search area on the upper surface; and processing a wall surface image of the search area to detect a target installation object. The method for detecting an installation object from a concrete surface image according to claim 1, wherein the installation object is a distance nameplate, and the detecting step is shape-based pattern matching. 3. The concrete surface image according to claim 2, wherein the installation object is a distance nameplate, and the template used for the shape-based pattern matching is a rectangular template in which two bolt holes are arranged. How to detect the installation. The installation object is a distance nameplate, and the setting step of the search area is based on a primary search area setting step performed by narrowing down a concrete surface image based on installation information, and an examination of the search result of the primary search area. The method for detecting an installation object from a concrete surface image according to claim 1, further comprising: a secondary search area setting step which is performed by further narrowing down the primary search area. 5. The installation from the concrete surface image according to claim 4, wherein the installation object is a distance nameplate, and the search of the primary search area is shape-based pattern matching using a template of the number “0”. How to detect things. The installed object is a distance nameplate, and the verification of the search result of the primary search area is based on the distance information plate candidate detected by shape-based pattern matching using the number “0” template according to the installation information. The method for detecting an installation object from the concrete surface image according to claim 4, wherein the detection is performed. 7. The method for detecting an installation object from a concrete surface image according to claim 6, wherein the installation object is a distance nameplate, and the installation information is an installation height and an installation interval of the distance nameplate. The method according to claim 1, wherein the installation object is a lower bundle, and the detection step is shape-based pattern matching. The method according to claim 1, wherein the installation object is a lower bundle, and the installation information includes an installation height and an installation interval of the lower bundle. The method for detecting an installation from a concrete surface image according to claim 1, wherein the installation is a fluorescent lamp, and the detection step is shape-based pattern matching. The method according to claim 1, wherein the installation object is a fluorescent lamp, and the installation information includes an installation height and an installation interval of the fluorescent lamp.