Rail transit tunnel apparent disease high-precision identification and real scene model construction device and method

Abstract

The rail transit tunnel apparent disease high-precision identification and real scene model construction device and method disclosed by the application comprises a mobile image acquisition device, a laser scanner, a disease identification station and the like, wherein the mobile image acquisition device is used for acquiring apparent images at different positions of a tunnel; the laser scanner is used for acquiring tunnel laser point cloud data; the disease identification station is used for identifying the apparent images and acquiring disease identification results; all the apparent images are spliced into a tunnel panorama and spherical projection, and tunnel three-dimensional real scene models are constructed by fusing the tunnel laser point cloud data; and diseases are calibrated through the tunnel three-dimensional real scene models. The accurate identification and calibration of tunnel diseases are realized.

Description

Technical Field

[0001] This invention relates to the field of quality inspection technology in tunnel maintenance, and in particular to a device and method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Subways and railway tunnels have become the main forms of rail transit, making the safety of tunnel structures paramount. However, due to complex factors such as construction conditions and operating environment, rail transit tunnels suffer from defects such as cracks, water leakage, structural deformation, and segment misalignment. These defects seriously affect the safety, stability, and durability of rail transit structures, posing significant safety hazards to tunnel operation.

[0004] To ensure tunnel structural safety and the normal operation of subways and railways, various methods of tunnel defect detection are employed. Traditional tunnel defect detection methods are primarily manual, relying mainly on human eyes or simple instruments, resulting in low efficiency and poor reliability. Currently, there are two main methods for automatically acquiring defect data: three-dimensional laser scanning and photogrammetry. Three-dimensional laser scanning uses laser scanning to acquire three-dimensional spatial point cloud data, and then analyzes the point cloud data to extract defect information. However, the inventors found that the acquired image resolution is low, and the efficiency in identifying major tunnel defects such as cracks is poor. Photogrammetry uses a line scan camera to acquire tunnel images through line scanning. The inventors found that it is difficult to simultaneously meet the requirements of high-speed inspection vehicle movement, high detection resolution, and continuous operation. Furthermore, a line scan CCD camera can only acquire one line of image information at a time, and a complete workpiece image is formed only through the physical accumulation of multiple lines of image stripes. In addition, line scan CCD cameras have relatively high requirements for movement speed and light source. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a device and method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels. By acquiring and analyzing apparent images of the tunnel, defects within the tunnel are identified and calibrated. All apparent images are stitched together to form a panoramic view of the tunnel and projected onto a spherical surface. Then, the tunnel laser point cloud data is fused to obtain a three-dimensional real-scene model of the tunnel. Finally, based on the three-dimensional real-scene model of the tunnel, the mileage and spatial location of defects are located, enabling rapid, unmanned, and intelligent daily tunnel inspections.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Firstly, a device for high-precision identification and real-scene modeling of apparent defects in rail transit tunnels was disclosed, including:

[0008] A mobile image acquisition device for acquiring apparent images of different locations within a tunnel;

[0009] Laser scanners are used to acquire laser point cloud data of tunnels;

[0010] The disease identification station is used to identify the appearance images and obtain the disease identification results; it stitches all the appearance images into a panoramic view of the tunnel and projects it onto a spherical surface, and integrates the tunnel laser point cloud data to construct a three-dimensional real-scene model of the tunnel; the disease is then calibrated using the three-dimensional real-scene model of the tunnel.

[0011] Secondly, a method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels is disclosed, including:

[0012] The appearance images of different locations in the tunnel are acquired using a mobile image acquisition device;

[0013] Acquire tunnel laser point cloud data using a laser scanner;

[0014] The disease identification station identifies the appearance images and obtains the disease identification results.

[0015] All the apparent images are stitched together to form a panoramic view of the tunnel and projected onto a spherical surface. The tunnel laser point cloud data is then integrated to construct a three-dimensional real-scene model of the tunnel.

[0016] The defects were identified using a 3D reality model of the tunnel.

[0017] Thirdly, an electronic device is proposed, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps described in the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels.

[0018] Fourthly, a computer-readable storage medium is proposed for storing computer instructions, which, when executed by a processor, complete the steps described in the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] 1. This invention acquires surface images of the tunnel by moving the camera and analyzes these images to identify and calibrate defects within the tunnel. By stitching all the surface images together to form a panoramic view of the tunnel and projecting it onto a spherical surface, and then fusing the tunnel's laser point cloud data, a three-dimensional real-scene model of the tunnel is obtained. This three-dimensional real-scene model of the tunnel locates the defects, enabling rapid, unmanned, and intelligent daily tunnel inspections.

[0021] 2. The mobile image acquisition device of this invention uses a black-and-white area array camera. Compared with most devices on the market equipped with non-zoomable lenses, this device's camera lens is zoomable, which can adapt to tunnels with different cross-sectional diameters. It has strong environmental adaptability and does not require designing a fixed lens support length based on the tunnel cross-sectional dimensions or adjusting the lens posture to achieve focusing. Moreover, the computer can simultaneously display the image during focusing, making focus adjustment more convenient and faster. The camera has high resolution and can identify cracks as small as 0.2mm and 4cm. 2 The device addresses issues such as water leakage and rockfall. Compared to existing mature equipment, the black-and-white area array camera used in this device features electrically adjustable zoom, focus, and aperture. For remote adjustment, it automatically adjusts the focus and sharpness, effectively ensuring accurate identification and location of defects while accurately acquiring surface images.

[0022] 3. When identifying diseases, this invention uses deep learning to analyze the appearance image and can calculate the length and width of crack diseases, resulting in high automatic detection accuracy.

[0023] 4. This invention fuses the acquired surface image with tunnel laser point cloud data to construct a three-dimensional real-scene model of the tunnel. This model can automatically mark the disease points and their location coordinates, reflecting the true location of the disease. Furthermore, through multiple comparative analyses, the development trend of the disease can be obtained.

[0024] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0026] Figure 1 This is an isometric view of the mobile image acquisition device disclosed in an embodiment of the present invention.

[0027] The components include: 1. Area scan camera, 2. Auxiliary light source, 3. Light source support frame, 4. Computer, 5. Fixed support frame, 6. Wheels, 7. Motor, and 8. Reducer. Detailed implementation method:

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0032] In this invention, terms such as "fixed connection," "connected," and "linked" should be interpreted broadly, indicating a fixed connection, an integral connection, or a detachable connection; a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can determine the specific meaning of these terms in this invention based on the specific circumstances, and they should not be construed as limitations on the invention.

[0033] Example 1

[0034] To achieve automatic identification and location of defects inside tunnels, this embodiment discloses a high-precision identification and real-scene model construction device for apparent defects in rail transit tunnels, including:

[0035] A mobile image acquisition device for acquiring apparent images of different locations within a tunnel;

[0036] Laser scanners are used to acquire laser point cloud data of tunnels;

[0037] The disease identification station is used to identify the appearance images and obtain the disease identification results; it stitches all the appearance images into a panoramic view of the tunnel and projects it onto a spherical surface, and integrates the tunnel laser point cloud data to construct a three-dimensional real-scene model of the tunnel; and it uses the three-dimensional real-scene model of the tunnel to calibrate the diseases.

[0038] Furthermore, the mobile image acquisition device includes multiple area array cameras and a walking support, with the multiple area array cameras arranged in a circular array on the walking support.

[0039] Furthermore, a computer platform is also installed on the walking support.

[0040] Furthermore, the mobile image acquisition device also includes a data acquisition unit and a synchronous encoder, and an auxiliary light source is set at each area scan camera. The area scan camera and the auxiliary light source are both connected to the synchronous encoder, and the data acquisition unit is connected to both the area scan camera and the synchronous encoder.

[0041] Furthermore, the shooting angles between adjacent area array cameras partially overlap.

[0042] Furthermore, the disease identification station identifies the appearance image, and the specific process for obtaining the disease identification results is as follows:

[0043] Preprocess the appearance image to obtain the preprocessed image;

[0044] Deep learning methods are used to identify the appearance images and obtain the disease identification results.

[0045] Furthermore, when the disease identification station identifies cracks as the disease, after crack identification is completed, the length and width of the crack are calculated to obtain the crack length and width calculation results.

[0046] Furthermore, when the disease identification station identifies the disease as water leakage or rock spalling, after completing the identification of water leakage or rock spalling, the area of ​​water leakage or rock spalling is calculated to obtain the area calculation result.

[0047] In practical implementation, the mobile image acquisition device includes multiple area array cameras 1 and a walking support. The multiple area array cameras 1 are arranged in a circular array on the walking support, so that the fixed positions of the area array cameras 1 on the walking support are located on the same semi-circular arc, and the shooting angles between adjacent area array cameras 1 partially overlap, reducing the possibility of missed detection and facilitating subsequent image stitching. The overall maximum shooting angle of the multiple area array cameras 1 can reach 270° in the circumferential direction.

[0048] The traveling support includes a fixed support and a traveling component. The traveling component is fixed to the bottom of the fixed support, and the area scan camera 1 is fixed to the fixed support. The traveling support moves by moving the traveling component. During the movement, the area scan camera 1 acquires appearance images at different locations inside the tunnel.

[0049] A computer platform is also installed on the walking support.

[0050] The fixed support includes a light source support frame 3 and a fixed support frame 5. The fixed support frame 5 includes a vertical support frame and a bottom support frame. One end of the vertical support frame is connected to the light source support frame 3, and the other end of the vertical support frame is connected to the bottom support frame. The walking component is fixed on the bottom support frame. In addition, a computer platform is also set on the vertical support frame. The computer platform is fixed to the vertical support frame by two horizontal supports. A computer 4 is placed on the computer platform to store the appearance images acquired by the area scan camera 1 in a timely manner.

[0051] In order to ensure the clarity of the images acquired by the area scan camera 1, and thus ensure the accuracy of disease identification, an auxiliary light source 2 is configured for each area scan camera 1.

[0052] Area scan cameras 1 and auxiliary light sources 2 are fixed sequentially and at intervals on the light source support frame 3. One light source provides exposure for both cameras, reducing exposure differences between camera shots and improving image processing efficiency. For example, 13 area scan cameras 1 and 14 auxiliary power sources 2 can be used.

[0053] To ensure the synchronous operation of the auxiliary light source 2 and the area scan camera 1, the mobile image acquisition device is also equipped with a synchronous encoder and a data acquisition device. The synchronous encoder is connected to the area scan camera 1 and the auxiliary light source 2 respectively, and is used to send synchronous pulse signals at set intervals according to the displacement of the mobile image acquisition device to trigger the synchronous operation of multiple area scan cameras 1 and auxiliary light sources 2. The data acquisition device is connected to the area scan camera 1 and the synchronous encoder respectively, and is used to acquire the apparent images inside the tunnel acquired by the area scan camera 1, and to number the acquired apparent images in conjunction with the synchronous encoder.

[0054] Among them, the area array camera 1 adopts a black and white wide-area shooting area array camera. The lenses of multiple black and white wide-area shooting area array cameras are zoomable, and the computer can display the image in real time. The focus adjustment is convenient and quick. There is no need to design a fixed lens bracket length according to the tunnel cross-section size or to achieve focus by adjusting the lens posture. When the tunnel cross-section size is not much different, it can adapt to tunnels with different cross-section diameters and has strong environmental adaptability.

[0055] The auxiliary light source is a strobe LED light source, and the fixed support frame 5 is made of aluminum.

[0056] The walking assembly includes a walking frame, wheels, a reducer 8, and a motor 7. The wheels are fixed on the walking frame, and the motor, reducer, and wheels are connected in sequence. The motor drives the wheels to rotate, thereby enabling the walking assembly to move.

[0057] The motor is a servo motor, and the reducer is a planetary reducer. The servo motor controls the speed and has very accurate positioning. The planetary reducer can reduce the motor speed while increasing the output torque and timely control the start, stop and speed change. The combination of motor and reducer can effectively maintain the function of the motor and save costs.

[0058] A 3D laser scanner is used to acquire laser point cloud data of tunnels.

[0059] The 3D laser scanner emits laser light to the tunnel surface and receives information reflected by the target objects. Based on the recorded angle between the incident and reflected light, it calculates the position of each point on the tunnel surface in the coordinate system according to the geometric relationship of triangles, thereby obtaining laser point cloud data of various locations in the tunnel.

[0060] Considering the operating speed of the mobile image acquisition device, an additional defect identification station was set up. The defect identification station is a ground workstation that analyzes and processes the tunnel appearance images acquired by the mobile image acquisition device and the laser point cloud data acquired by the 3D laser scanner to identify and calibrate tunnel defects.

[0061] The defect identification station analyzes and processes the surface images and laser point cloud data to achieve the identification and calibration process of tunnel defects as follows:

[0062] Preprocess the appearance image to obtain the preprocessed image;

[0063] The preprocessed images are then identified to obtain the identification results of tunnel defects;

[0064] All the apparent images are stitched together to form a panoramic view of the tunnel and projected onto a spherical surface. The tunnel laser point cloud data is then integrated to construct a three-dimensional real-scene model of the tunnel.

[0065] By integrating the ring number and mileage information recorded by the 3D real-scene model and the 3D laser scanner, the mileage and spatial location of the defects in the tunnel are determined.

[0066] The median filtering method is used to preprocess the apparent image to eliminate noise and make the image clearer and easier to identify. Then, deep learning is used to identify the preprocessed image and obtain the disease identification results.

[0067] The identified tunnel defects include water leakage, lining cracks, rail fissures, honeycomb cracks on sleepers, and loose bolt fasteners.

[0068] Taking tunnel fissures as an example, the identification process is explained. A Pascal Voc dataset is created from tunnel fissure images, which is then used to train a Faster R-CNN network. Finally, a fissure model is established, and the preprocessed images are identified using this model to obtain the fissure identification results.

[0069] When the identified tunnel defect is a crack, crack features are extracted from the preprocessed image, and the length and width of the crack are calculated based on the extracted crack features. Specifically:

[0070] Histogram calculation is performed on the preprocessed image, and contrast is calculated by gray-scale stretching. Then, feature points are calculated by finding pyramids through Gaussian difference. Next, Laplace transform is performed to calculate the second derivative edge of the image. Finally, crack edges are filtered out based on the mathematical features of the image, and edge calculation is performed to obtain crack features.

[0071] In terms of crack length and width calculation, the crack is linearly segmented according to the angle change, and the pixel length of the segmented cracks is accumulated. Then, the pixel width of each segment is calculated separately. Finally, a scale is used to realize the conversion from pixel space to actual space to obtain the calculated results of crack length and width.

[0072] All the apparent images are stitched together to form a panoramic view of the tunnel and then projected onto a spherical surface. The tunnel's laser point cloud data is then fused to construct a 3D reality model of the tunnel. The specific process is as follows:

[0073] A 3D view frustum is obtained from appearance image regions at different locations. Then, based on the segmented target point cloud, the position of the midpoint in each appearance image is obtained, and the alignment point of each appearance image is obtained. The alignment point is translated using the PointNet network to form a 3D detection result. Based on the VoteNet method, a deep Hough voting method is used in combination with the acquired appearance image to perform 3D target detection on the laser point cloud data through image voting. The 3D detection results obtained from the appearance image and the 3D detection results obtained from the laser point cloud data are fused using gradient blending. The 3D detection results obtained from the appearance image guide the 3D target detection of the laser point cloud data, realizing the fusion of point cloud data and appearance image, and establishing a 3D reality model.

[0074] The high-precision identification and real-scene model construction device for apparent defects in rail transit tunnels disclosed in this embodiment uses a mobile image acquisition device in conjunction with a 3D laser scanner to acquire high-definition apparent images and 3D point cloud data of the tunnel. The acquired apparent images are first preprocessed, and then image recognition software based on deep learning is used to identify defects, label them, and extract features such as length and width. Finally, the images are stitched together and the laser point cloud data is fused to construct a 3D real-scene model of the tunnel. The location of defects is calibrated using the 3D real-scene model of the tunnel, achieving the effects of high detection efficiency, high accuracy, high degree of automation, and understanding of the development trend of defects.

[0075] Example 2

[0076] This embodiment discloses a method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels, including:

[0077] The appearance images of different locations in the tunnel are acquired using a mobile image acquisition device;

[0078] Acquire tunnel laser point cloud data using a laser scanner;

[0079] The disease identification station identifies the appearance images and obtains the disease identification results.

[0080] All the apparent images are stitched together to form a panoramic view of the tunnel and projected onto a spherical surface. The tunnel laser point cloud data is then integrated to construct a three-dimensional real-scene model of the tunnel.

[0081] The defects were identified using a 3D reality model of the tunnel.

[0082] Example 3

[0083] In this embodiment, an electronic device is disclosed, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps described in the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels disclosed in Embodiment 2.

[0084] Example 4

[0085] In this embodiment, a computer-readable storage medium is disclosed for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps described in the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels disclosed in Embodiment 2.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A device for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels, characterized in that, include: A mobile image acquisition device for acquiring apparent images of different locations within a tunnel; Laser scanners are used to acquire laser point cloud data of tunnels; The disease identification station is used to identify the appearance images and obtain the disease identification results; it stitches all the appearance images into a panoramic view of the tunnel and projects it onto a spherical surface, and integrates the tunnel laser point cloud data to construct a three-dimensional real-scene model of the tunnel; and it uses the three-dimensional real-scene model of the tunnel to calibrate the diseases. The process of stitching together all the apparent images into a panoramic view of the tunnel and projecting it onto a spherical surface, then fusing the tunnel's laser point cloud data to construct a 3D reality model of the tunnel, is as follows: A 3D view frustum is obtained from appearance image regions at different locations. Then, the position of the point in each appearance image is obtained based on the segmented target point cloud. Alignment points of each appearance image are obtained. The alignment points are translated using the PointNet network to form a 3D detection result. Based on the VoteNet method, a deep Hough voting method is used in combination with the acquired appearance images to perform 3D target detection on the laser point cloud data through image voting. The 3D detection results obtained from the appearance images and the 3D detection results obtained from the laser point cloud data are fused using gradient mixing. The 3D detection results obtained from the appearance images guide the 3D target detection of the laser point cloud data, realizing the fusion of point cloud data and appearance images and establishing a 3D reality model. The tunnel 3D real-scene model can automatically mark the disease points and their location coordinates, reflecting the real location of the disease, and through multiple comparative analyses, it can obtain the disease development trend. The mobile image acquisition device includes multiple area scan cameras and a walking support. The multiple area scan cameras are arranged in a circular array on the walking support. A computer platform is also installed on the walking support. The shooting angles between adjacent area scan cameras partially overlap. The specific process by which the disease identification station identifies the appearance image and obtains the disease identification results is as follows: Preprocess the appearance image to obtain the preprocessed image; Deep learning methods are used to identify preprocessed images and obtain disease identification results. The mobile image acquisition device uses a black and white area array camera. The camera lens of this device is zoomable and can adapt to tunnels with different cross-sectional diameters. It is highly adaptable to the environment and does not require designing a fixed lens support length according to the tunnel cross-sectional size or adjusting the lens posture to achieve focus. Moreover, the computer can simultaneously display the image during focus adjustment, making focus adjustment more convenient and faster.

2. The high-precision identification and real-scene model construction device for apparent defects in rail transit tunnels as described in claim 1, characterized in that, The mobile image acquisition device also includes a data acquisition unit and a synchronous encoder, and an auxiliary light source is set at each area scan camera. The area scan camera and the auxiliary light source are both connected to the synchronous encoder, and the data acquisition unit is connected to both the area scan camera and the synchronous encoder.

3. The high-precision identification and real-scene model construction device for apparent defects in rail transit tunnels as described in claim 1, characterized in that, When the disease identification station identifies cracks as the disease, it extracts tunnel crack features from the preprocessed image and calculates the length and width of the cracks based on the extracted tunnel crack features.

4. A method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels, employing the device for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels as described in any one of claims 1-3, characterized in that, include: The appearance images of different locations in the tunnel are acquired using a mobile image acquisition device; Acquire tunnel laser point cloud data using a laser scanner; The disease identification station identifies the appearance images and obtains the disease identification results. All the apparent images are stitched together to form a panoramic view of the tunnel and projected onto a spherical surface. The tunnel laser point cloud data is then integrated to construct a three-dimensional real-scene model of the tunnel. The defects were identified using a 3D reality model of the tunnel. The mobile image acquisition device includes multiple area scan cameras and a walking support. The multiple area scan cameras are arranged in a circular array on the walking support. A computer platform is also installed on the walking support. The shooting angles between adjacent area scan cameras partially overlap. The specific process by which the disease identification station identifies the appearance image and obtains the disease identification results is as follows: Preprocess the appearance image to obtain the preprocessed image; Deep learning methods are used to identify preprocessed images and obtain disease identification results. The mobile image acquisition device uses a black and white area array camera. The camera lens of this device is zoomable and can adapt to tunnels with different cross-sectional diameters. It is highly adaptable to the environment and does not require designing a fixed lens support length according to the tunnel cross-sectional size or adjusting the lens posture to achieve focus. Moreover, the computer can simultaneously display the image during focus adjustment, making focus adjustment more convenient and faster.

5. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps of the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels as described in claim 4.

6. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps of the method for high-precision identification and real-scene model construction of apparent defects in rail transit tunnels as described in claim 4.

Images