Large-span suspension bridge main cable shape automatic calculation method based on airborne laser scanning

Abstract

The invention relates to the field of bridge line shape detection in civil engineering, in particular to a large-span suspension bridge main cable line shape automatic calculation method based on airborne laser scanning. The method comprises the following steps: fusing high-precision inertial navigation, a differential GNSS (Global Navigation Satellite System), airborne laser scanning equipment and an RTK (Real-Time Kinematic) technology, and acquiring a large-scale fine point cloud model by utilizing a two-level route planning method: on the basis of an SCF-Net network, providing an SCF-Bridge-Net based on a sling attention mechanism in combination with the structural characteristics of a suspension bridge; inclination angle, elevation and mileage information of all cable clamps are realized through a geometric information calculation method, and linear measurement of a bridge is realized. According to the method for automatically calculating the line shape of the main cable of the large-span suspension bridge based on airborne laser scanning, aiming at obtaining point cloud of a large suspension bridge structure in the construction process, a high-precision inertial navigation system and a differential GPS fusion method are combined with airborne laser scanning equipment, and a route planning method with a cable clamp as a core path is established through a coarse scanning model; and fine scanning of the large-span suspension bridge is realized.

Description

Technical Field

[0001] The present invention relates to the field of bridge alignment detection in civil engineering, and particularly to an automatic calculation method for the alignment of the main cable of a long-span suspension bridge based on airborne laser scanning. Background Art

[0002] The accuracy of geometric control during the construction stage of the main cable and cable clamp of a long-span suspension bridge directly affects the safety performance of the completed bridge state. Currently, the construction quality control of suspension bridges mainly relies on traditional measuring tools such as total stations, levels, and sensors. They perform discrete measurements on some characteristic points in the structure and compare the measured discrete point data with the corresponding positions in the theoretical design model point by point. The above traditional methods cannot comprehensively reflect whether the overall construction quality of the suspension bridge meets the design requirements, and the operation of total stations, levels, and other traditional measuring instruments is relatively cumbersome, which is not conducive to quickly, accurately, and comprehensively evaluating the construction quality at each construction stage of the suspension bridge. Therefore, there is an urgent need to discover a new method to quickly and comprehensively evaluate the three-dimensional continuous space state of the suspension bridge.

[0003] One of the important means of geometric perception of bridge structures based on vision. With the development of emerging technologies of consumer drones, three-dimensional reconstruction based on images has received more attention. It can perform 3D modeling of bridges through numerous overlapping aerial photos at low cost, providing a potential solution for bridge geometric perception. Compared with terrestrial laser scanning, although its accuracy is relatively low, three-dimensional reconstruction based on images is an economical and efficient three-dimensional mapping method. Generally speaking, photogrammetry or video measurement algorithms find distinct features from overlapping images or video frames, estimate the parameters and poses of cameras, and finally generate three-dimensional point cloud data. However, due to high requirements for the acquisition overlap rate, long reconstruction time, and low accuracy, three-dimensional reconstruction has less application in the measurement of large civil engineering structures. Compared with the method of obtaining point clouds through three-dimensional reconstruction, the acquisition method of laser scanning is more direct and can obtain the true three-dimensional information of the target more efficiently and accurately. It is important to ensure that the obtained point cloud data meets the requirements of specific construction applications. In practice, point cloud data is usually obtained based on the experience of engineers, which may lead to low data quality or redundancy. In addition, the types, orientations, shapes, sizes, and surrounding scenes of infrastructure such as bridges are more diverse and complex, which poses great challenges to data acquisition. Therefore, formulating a scanning strategy plan to minimize the time on site and maximize the quality and quantity of data requires careful consideration. In practice, engineering surveyors usually plan the scanning strategy after determining the purpose of data acquisition and on-site measurement, roughly determining the scanning station network and scanning parameters. The data quality setting has a great impact on the data acquisition rate (measured pixels / second), that is, the scanning time, which in turn affects the amount of redundant data and the noise range. Although various scholars have conducted extensive research on the quality of acquired data, there is little research on the airborne laser scanning route planning for long-span suspension bridges.

[0004] Accurate segmentation of large-scale point clouds is a key link in the process of point cloud processing. Segmentation is a key step in defining a logical partition for the acquired data points so that they can be interpreted as geometric shapes presenting the surfaces of detected objects. Since the number of acquired data points is usually very large, proposing an accurate automatic segmentation method has become a challenge in recent years. Methods designed for large-scale point clouds have been proposed, such as SPG, PCT, and RandLA-Net. However, how to accurately learn more effective features from the point clouds of long-span suspension bridges for semantic segmentation remains a difficult problem.

[0005] Therefore, limited by problems such as the complexity of large bridge span data processing, the research on the automatic segmentation method of point clouds and construction quality assessment for large suspension bridges during construction has become a key difficult problem that current engineering personnel are concerned about. Summary of the Invention

[0006] The object of the present invention is to provide an automatic calculation method for the alignment of the main cable of a long-span suspension bridge based on airborne laser scanning to overcome the complex construction environment of long-span suspension bridges and the processing of large-scale point cloud data, aiming to fully combine intelligent scanning equipment with efficient processing algorithms to improve the detection efficiency of bridge alignment.

[0007] The object of the present invention can be achieved by the following technical solutions:

[0008] According to one aspect of the present invention, there is provided an automatic calculation method for the alignment of the main cable of a long-span suspension bridge based on airborne laser scanning, the method comprising the following steps:

[0009] Step S1, integrating high-precision inertial navigation, differential GNSS, airborne laser scanning equipment and RTK technology, and using a two-level flight line planning method to obtain a large-scale fine point cloud model:

[0010] First-level flight line planning: Based on the airborne laser scanning equipment, flight line planning is carried out through the roughly scanned point cloud model,

[0011] Second-level flight line planning: Based on the roughly scanned flight line obtained from the first-level flight line, a high-precision inertial navigation system and differential GPS fusion method are used to achieve high-precision point cloud acquisition of kilometer-level long-span suspension bridges;

[0012] Step S2, based on the SCF-Net network and combined with the structural characteristics of the suspension bridge, a SCF-Bridge-Net based on sling attention mechanism is proposed;

[0013] The sling attention mechanism part mainly extracts all sling area points at the same height through the structural characteristics of the suspension bridge, realizes spatial area classification in the two-dimensional plane based on Euclidean clustering and nearest neighbor algorithms, locks the target to be recognized within a small range, and improves the recognition accuracy and time;

[0014] The SCF-Bridge-Net part is mainly composed of the above sling attention mechanism module and the SCF-Net network;

[0015] The core idea of the SCF-Net is to improve the detection and classification of 3D targets by capturing the spatial context information in the point cloud. The point cloud of the classified spatial area is input into the network to obtain a higher-precision point cloud component of the suspension bridge.

[0016] Step S3, through the point cloud components of the suspension bridge obtained in S2, the inclination angle, elevation and mileage information of all cable clamps are realized through geometric information calculation methods, and the alignment measurement of the bridge is realized.

[0017] Project the segment point cloud onto the local plane, converting the cylinder recognition in the point cloud into circle recognition in the plane point cloud. Randomly sample 3 points from the projected plane point cloud and calculate the center coordinates (x i , y i ) and radius R i of the circle formed by these three points; Repeat this step N times to obtain samples of the radius;

[0018] Perform probability density statistics on all the obtained radius samples, where the horizontal axis is the number of times and the vertical axis is the density. Take the radius value R max and the center (x 0 , y 0 ) corresponding to the maximum probability density;

[0019] After obtaining the segmented point cloud, set a gradient along the bridge deck direction, calculate the number of point clouds contained in each gradient, and then select the density peak with a local maximum as the position of the sling, and calculate the density gradient of the sling area;

[0020] Calculate the plane centers on both sides of the cable clamp respectively, take the straight line where the two centers are connected as the straight line where the cable clamp is located, and then calculate the obtained sling density peak. The intersection point of the sling center line and the aforementioned straight line is the geometric center of the cable clamp calculated;

[0021] The automatic calculation method for the main cable alignment of a long-span suspension bridge based on airborne laser scanning according to the present invention, step S2, the SCF-Bridge-Net based on the sling attention mechanism includes: SCF-Net and the sling attention mechanism module;

[0022] The SCF-Net network uses the adjacent point cloud information to obtain local information, and at the same time can obtain global information based on the volume ratio of the region to the full point cloud, so as to be applicable to large-scale point cloud segmentation scenarios;

[0023] The sling attention mechanism module is mainly implemented based on Euclidean clustering and the nearest neighbor algorithm through spatial region classification in a two-dimensional plane. The specific process is as follows:

[0024] The automatic calculation method for the main cable alignment of a long-span suspension bridge based on airborne laser scanning according to the present invention, the geometric information calculation method in step S3 specifically includes:

[0025] Step S31, project the segment point cloud onto the local plane, convert the cylinder recognition in the point cloud into circle recognition in the plane point cloud, randomly sample 3 points from the projected plane point cloud, and calculate the center coordinates (x i , y i ) and radius R i of the circle formed by these three points. Repeat this step N times to obtain samples of the radius;

[0026] Step S32: Conduct probability density statistics on all obtained radius samples, where the horizontal axis represents the number of times and the vertical axis represents the density, and take the radius value R corresponding to the maximum probability density. max and the center of the circle (x 0 , y 0 );

[0027] Step S33: After obtaining the segmented point cloud, set a gradient along the bridge deck direction, calculate the number of point clouds contained in each gradient, then select the density peak with a local maximum as the position of the sling, and calculate the density gradient of the sling area.

[0028] Step S34: In Step S31, the plane centers of the two sides of the clamp are calculated respectively. The straight line connecting the two centers is used as the straight line where the clamp is located. Then, using the sling density peak obtained by the calculation in S33, the intersection point of the sling center line and the aforementioned straight line is the geometric center of the clamp.

[0029] As a preferred technical solution, the two-level route planning in Step S1 is specifically as follows:

[0030] To establish a point cloud model of the bridge, it is necessary to first determine the scanning plan, including parameters such as the area where the suspension bridge is located, the scanning height, and the route overlap rate. To ensure the quality of model reconstruction, it is necessary to plan the scanning route according to the structural characteristics of the suspension bridge. The airborne laser scanning adopted in the present invention has an absolute accuracy of 1 cm, but it only has 100,000 point scanning data per second. To obtain enough point cloud data of the target, a certain distance needs to be ensured. Especially for a specific component such as the main cable, since the main cable line shape is in a catenary shape, it is difficult to ensure the accuracy requirements and safety by manually controlling the UAV for scanning. Therefore, it is necessary to plan the route in advance and perform automated flight. A UAV scanning route calculation method for the structure of a long-span suspension bridge is proposed for the main cable line shape of the suspension bridge.

[0031] The highest point of the route is located 30 meters above the tower top. Since the lowest point of the clamp is only 4 m above the bridge deck, it is necessary to set a route that changes along the height of the main cable in the flight direction along the bridge deck in the scanning cable system to ensure that the key area can be scanned within a safe distance l a and the angle is within 45 degrees. This part of the scanning area is called the effective area. Therefore, in order to scan the target area more effectively and with high quality, it is necessary to accurately set the waypoints. The present invention adopts a pre-scanning method. A rough scanning model of the suspension bridge is obtained through fast route scanning. Then, 10 coordinate points on the main cable are uniformly selected in the WGS-84 coordinate system. Then, accurate route points are set by ensuring that all targets are within the effective area as described above. Then, a three-dimensional model of the suspension bridge including the main cable is obtained through route reconstruction.

[0032] As a preferred technical solution, the high-precision inertial navigation system in Step S1 is specifically as follows:

[0033] In the GNSS / INS integrated positioning system, the scanner obtains the distance and angle from the light spot to the platform. The WGS-84 coordinates of the antenna center are obtained through GNSS, and the attitude information of the platform can be obtained through INS. The combination of INS and GNSS constitutes a positioning and attitude system, which can provide the instantaneous position and attitude information of the platform;

[0034] As a preferred technical solution, the integrated positioning and navigation of GNSS and INS mainly uses the Kalman filtering algorithm to estimate the error state of INS and feedback-correct INS.

[0035] The general model of the discrete Kalman filtering system is as follows:

[0036]

[0037] Where X k and X k-1 represent the states of the system at times k and k - 1 respectively; A represents the system state displacement matrix; w k-1 represents the random noise in the state transition process; Y k is the observed value of the system; H is the observation vector of the system; v k is the observation noise. w k and w k satisfy the following conditions:

[0038]

[0039] Where: Q k is the covariance matrix of w k , R k is the covariance matrix of v k , and δ kj is the Dirac function.

[0040] As a preferred technical solution, the differential GPS in step S1 is specifically:

[0041] The GPS positioning system uses RTK technology to eliminate positioning errors;

[0042] RTK positioning technology is a real-time kinematic differential positioning technology based on high-precision carrier phase observations, including a reference station and a rover station. The principle of the differential real-time kinematic positioning technology is based on the spatial correlation between the reference station and the rover station. The specific process is as follows:

[0043] The reference station first transmits the carrier phase observations and station coordinates it obtains in real time to the surrounding dynamic users via a data communication link. The rover data processing module uses the method of kinematic differential positioning to determine the coordinates of the rover relative to the reference station, and then calculates its instantaneous coordinates based on the coordinates of the reference station. Furthermore, it realizes the accurate depiction of the UAV flight path and the precise acquisition of scanned point clouds.

[0044] As a preferred technical solution, the SCF-Bridge-Net based on the sling attention mechanism in step S2 includes: SCF-Net; sling attention mechanism module;

[0045] As a preferred technical solution, SCF-Net learns the context information of the point cloud in the local and global spaces. The SCF-Bridge-Net based on the sling attention mechanism in step S2 includes: SCF-Net, sling attention mechanism module.

[0046] As a preferred technical solution, the spatial region classification in the two-dimensional plane in step S2 is implemented based on the Euclidean clustering and nearest neighbor algorithms. The specific process is as follows:

[0047] The SCF-Bridge-Net based on the sling attention mechanism in step S2 includes: SCF-Net, sling attention mechanism module;

[0048] The SCF-Net network uses the adjacent point cloud information to obtain local information, and at the same time can obtain global information based on the volume ratio of the region to the whole point cloud, so as to be applicable to large-scale point cloud segmentation scenarios;

[0049] The sling attention mechanism module is mainly implemented through the spatial region classification in the two-dimensional plane based on the Euclidean clustering and nearest neighbor algorithms. The specific process is as follows:

[0050] (1). Intercept the region between the lowest point of the sling clamp and the main girder along the coordinate direction of the bridge tower, and select the sling region within this z coordinate range;

[0051] (2). Based on the selected sling region, calculate the center based on the Euclidean clustering and nearest neighbor algorithms;

[0052] First, classify all double slings by the Euclidean clustering method; the basic principle of Euclidean clustering is that for any point P, use the cross-tree nearest neighbor search to find its nearest k points, and when the distance from the nearest point to point P is less than the set threshold, cluster them into this set; when the number of points in the set no longer increases, select the points outside the set and repeat the process;

[0053] (3). Calculate the average value of all categories, use the average value of each cluster as the center of the square, and form an x-y region with a side length of 2m as the square side length to obtain the final core region.

[0054] Among them, the accuracy of threshold selection determines the clustering accuracy. The basis for the threshold selection in the present invention is that adjacent two sling ropes need to be grouped as one category according to requirements, and the threshold must be greater than the sling rope spacing of 0.5 m. Since the distance between two groups of sling ropes is 18 m, any number within the range of 0.5 - 18 m can be selected as the threshold to meet the requirements, and the present invention selects 10 m.

[0055] The automatic calculation method for the main cable alignment of a long-span suspension bridge based on airborne laser scanning according to the present invention

[0056] Step S3, the specific geometric information calculation method for the point cloud components of the suspension bridge includes:

[0057] Step S31, project the segment point cloud in the point cloud components of the suspension bridge onto a local plane, so that the cylinder recognition in the point cloud is transformed into circle recognition in the plane point cloud. Randomly sample 3 points from the projected plane point cloud, and calculate the center coordinates (x i , y i ) and radius R i formed by these three points; repeat this step N times to obtain a sample of the radius;

[0058] Step S32, perform probability density statistics on all the obtained radius samples, where the horizontal axis is the number of times and the vertical axis is the density, and take the radius value R max corresponding to the maximum probability density and the center (x 0 , y 0 );

[0059] Step S33, after obtaining the segmented point cloud from the point cloud components of the suspension bridge, set a gradient along the bridge deck direction, calculate the number of point clouds contained in each gradient, and then select the density peak with a local maximum as the position where the sling rope is located, and calculate the density gradient of the sling rope area;

[0060] Step S34, in Step S31, the plane centers on both sides of the cable clamp are calculated respectively. The straight line connecting the two centers is used as the straight line where the cable clamp is located. Using the sling rope density peak obtained by calculation in S33, the intersection point of the sling rope center straight line and the aforementioned straight line is the geometric center of the cable clamp.

[0061] Beneficial effects

[0062] 1) The automatic calculation method for the main cable alignment of a long-span suspension bridge based on airborne laser scanning provided by the present invention is aimed at obtaining the point cloud of the large-scale suspension bridge structure during the construction process. By adopting the method of fusing a high-precision inertial navigation system and differential GPS and combining with an airborne laser scanning device, and through the route planning method with the cable clamp as the core path established by the rough scanning model, the refined scanning of the long-span suspension bridge is realized.

[0063] 2) Aiming at the complex problem of large-scale structural point cloud segmentation, a large-scale point cloud processing network based on the structural characteristics of suspension bridges and SCF-Net is proposed to achieve the spatial positioning of cable clamps and quickly realize the automatic calculation of geometric information such as the inclination angle and elevation of cable clamps. The proposed method can still accurately identify and segment point cloud components even when there is partial occlusion of key components, and accurately calculate geometric information. Brief Description of the Drawings

[0064] Figure 1 Schematic diagram of the overall framework proposed by the present invention;

[0065] Figure 2 Schematic diagram of the hardware system in the data acquisition system proposed by the present invention;

[0066] Figure 3 Schematic diagram of the SCF-Bridge-Net proposed by the present invention;

[0067] Figure 4 Schematic diagram of the fine-scanning route planning proposed by the present invention;

[0068] Figure 5 Schematic diagram of the flow of the GNSS / INS integrated navigation and positioning algorithm based on Kalman filtering proposed by the present invention;

[0069] Figure 6 Detail diagram of the SCF-Net proposed by the present invention;

[0070] Figure 7 Schematic diagram of the spatial region classification proposed by the present invention;

[0071] Figure 8 Flow chart of geometric information recognition considering structural characteristics proposed by the present invention;

[0072] Figure 9 Schematic diagram of the experimental arrangement proposed in the example;

[0073] Figure 10 Schematic diagram of the implementation process of the route planning proposed in the example;

[0074] Figure 11 Schematic diagram of the comparison of the segmented structures proposed in the example;

[0075] Figure 12 Comparison table of the results of the algorithm framework proposed by the present invention and other advanced segmentation algorithms;

[0076] Figure 13 Comparison diagram of the elevation of the cable clamp proposed in the example with the total station and the drawing, and the angle error diagram of the cable clamp. Detailed Implementation Manner

[0077] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0078] The automatic calculation method for the main cable alignment of long-span suspension bridges based on airborne laser scanning in the present invention. First, for the acquisition of the structural point cloud of large suspension bridges during the construction process, a high-precision inertial navigation system and differential GPS fusion method are adopted, combined with airborne laser scanning equipment, and a route planning method with the cable clamp as the core path is established through a rough scanning model to achieve fine scanning of long-span suspension bridges. Subsequently, based on the original SCF-Net network, combined with the structural characteristics of suspension bridges, SCF-Bridge-Net based on the sling attention mechanism is proposed. The sling attention part mainly extracts all sling area points at the same height through the structural characteristics of the suspension bridge, and realizes spatial region classification in the two-dimensional plane based on the Euclidean clustering and nearest neighbor algorithms, greatly reducing the number of point clouds and improving the processing efficiency. Finally, the inclination angle, elevation, and mileage information of all cable clamps are realized through geometric information calculation methods to achieve the alignment measurement of the bridge.

[0079] The automatic calculation method for the main cable alignment of long-span suspension bridges based on airborne laser scanning in the present invention. The overall framework of this solution is as Figure 1 shown and includes the following components:

[0080] Component 1: Obtain a large-scale fine point cloud model based on two-level route planning; based on airborne laser scanning equipment, perform route planning through the rough scanned point cloud model, and adopt a high-precision inertial navigation system and differential GPS fusion method to achieve high-precision point cloud acquisition of kilometer-level long-span suspension bridges. The hardware system of the data acquisition system is as Figure 2 shown.

[0081] Component 2: Realize rapid detection of bridge alignment based on the bridge segmentation of SCF-Bridge-Net; based on the original SCF-Net network, combined with the structural characteristics of suspension bridges, SCF-Bridge-Net based on the sling attention mechanism is proposed. The sling attention part mainly extracts all sling area points at the same height through the structural characteristics of the suspension bridge, and realizes spatial region classification in the two-dimensional plane based on the Euclidean clustering and nearest neighbor algorithms. The inclination angle, elevation, and mileage information of all cable clamps are realized through geometric information calculation methods to achieve the alignment measurement of the bridge; The schematic diagram of SCF-Bridge-Net is as Figure 3 shown.

[0082] Furthermore, the specific details of the two-level route planning in Component 1 are as follows:

[0083] To establish a point cloud model of the bridge, it is necessary to first determine the scanning plan, including parameters such as the area where the suspension bridge is located, the scanning height, and the flight line overlap rate. In order to ensure the quality of model reconstruction, it is necessary to plan the scanning flight lines according to the structural characteristics of the suspension bridge. The airborne laser scanning used in the present invention has an absolute accuracy of 1 cm, but it only has 100,000 point scanning data per second. In order to obtain sufficient point cloud data of the target, a certain distance needs to be ensured. Especially for a specific component such as the main cable, since the shape of the main cable is catenary, it is difficult to ensure the accuracy requirements and safety by manually controlling the UAV for scanning. Therefore, it is necessary to plan the flight line in advance and perform automated flight. A method for calculating the UAV scanning flight line for the structure of a long-span suspension bridge is proposed for the shape of the main cable of the suspension bridge.

[0084] The highest point of the flight line is 30 meters above the top of the tower. Since the lowest point of the cable clamp is only 4 m above the bridge deck, it is necessary to set a flight line that changes along the height of the main cable in the flight direction along the bridge deck in the scanning cable system to ensure that the key area can be scanned within a safe distance l a range and at an angle within 45 degrees. This part of the scanning area is called the effective area. Therefore, in order to scan the target area more effectively and with high quality, it is necessary to accurately set the waypoints. The present invention adopts a pre-scanning method. By quickly scanning the flight line, a rough scanning model of the suspension bridge is obtained. Then, 10 coordinate points on the main cable are uniformly selected in the WGS-84 coordinate system. Then, precise flight line points are set by ensuring that all targets are within the effective area as described above. Then, a three-dimensional model of the suspension bridge including the main cable is obtained through flight line reconstruction. The schematic diagram of the fine-scanning flight line planning is as Figure 4 shown.

[0085] The software operation of flight line planning mainly relies on the KML markup language. It is an encoding specification based on the XML markup language. Through the KML document, various parameters of the UAV flight line can be defined, such as the flight path, waypoints, height, speed, trigger actions, etc. By defining the position of the waypoints and the connection relationship between the waypoints, the flight path of the UAV can be planned. Each waypoint can specify position information such as longitude, latitude, and height, as well as navigation parameters such as speed, direction, and tilt angle. In addition, various task actions can be triggered at specific waypoint positions or times, such as sensor acquisition, task switching, etc. By specifying the waypoints and action parameters, complex tasks can be achieved.

[0086] Furthermore, the specific details of the high-precision inertial navigation system in Component One are as follows:

[0087] The high-precision inertial navigation system is a GNSS / INS integrated positioning system. The scanner obtains the distance and angle from the light spot to the platform. The WGS-84 coordinates of the antenna center are obtained through GNSS, and the attitude information of the platform can be obtained through INS. The combination of INS and GNSS constitutes a positioning and attitude system, which can provide the instantaneous position and attitude information of the platform.

[0088] The combined positioning and navigation of GNSS and INS mainly uses the Kalman filtering algorithm to estimate the error state of INS and feedback-correct INS. The flow chart of the GNSS / INS integrated navigation and positioning algorithm based on Kalman filtering is as Figure 5 shown. The general model of the discrete Kalman filtering system is as follows:

[0089]

[0090] where X k , X k-1 represent the states of the system at times k and k - 1 respectively; A represents the system state displacement matrix; w k-1 represents the random noise in the state transition process; Y k is the observation value of the system; H is the observation vector of the system; v k is the observation noise. w k and w k satisfy the following conditions:

[0091]

[0092] where: Q k is the covariance matrix of w k , R k is the covariance matrix of v k , δ kj is the Dirac function.

[0093] Furthermore, the specific details of the differential GPS in Component 1 are as follows:

[0094] The GPS positioning system uses RTK technology to eliminate positioning errors. RTK positioning technology is a real-time kinematic differential positioning technology based on high-precision carrier phase observations, including a reference station and a rover station. The principle of the differential real-time kinematic positioning technology is based on the spatial correlation between the reference station and the rover station.

[0095] Furthermore, the specific process of the differential GPS in Component 1 is as follows:

[0096] The reference station first transmits the carrier phase observations and station coordinates it has obtained to the surrounding dynamic users in real time through the data communication link. The mobile station data processing module uses the method of kinematic differential positioning to determine the coordinates of the mobile station relative to the reference station, and then calculates its instantaneous coordinates based on the coordinates of the reference station. Furthermore, it realizes the accurate description of the UAV flight path and the accurate acquisition of the scanned point cloud.

[0097] Furthermore, the bridge segmentation based on SCF-Bridge-Net in Component 2 for realizing the rapid detection of the bridge alignment includes the following steps: Step 1: SCF-Net based on the cable attention mechanism, as Figure 6 shown; Step 2: Geometric information recognition considering structural features.

[0098] Furthermore, Step 1 of the bridge segmentation based on SCF-Bridge-Net in Component 2 for realizing the rapid detection of the bridge alignment: The specific details of the SCF-Bridge-Net based on the cable attention mechanism are as follows:

[0099] Furthermore, Step 1 of the bridge segmentation based on SCF-Bridge-Net in Component 2 for realizing the rapid detection of the bridge alignment: The schematic diagram of the spatial region classification of the SCF-Bridge-Net based on the cable attention mechanism in the two-dimensional plane is as Figure 7 shown, and the specific method is as follows:

[0100] First, intercept the area from the lowest point of the cable clamp to the main girder along the z coordinate direction, and the sling area within this z coordinate range can be selected.

[0101] Then, based on the selected sling area, calculate the center using the Euclidean clustering and nearest neighbor algorithms. First, classify all double slings through the Euclidean clustering method. The basic principle of Euclidean clustering is that for any point P, use the KD-Tree nearest neighbor search to find its nearest k points. When the distance from the nearest point to point P is less than the set threshold, it is clustered into this set. When the number of points in the set no longer increases, select the points outside the set and repeat the process.

[0102] Finally, calculate the average value of all categories, use the average value of each cluster as the center of the square, and form an x-y region with a 2m side length of the square to obtain the final core region.

[0103] Among them, the accuracy of the threshold selection determines the clustering accuracy. The basis for the threshold selection in the present invention is that according to the requirement to classify two adjacent slings as one category, the threshold must be greater than the sling spacing of 0.5m. And the distance between two groups of slings is 18m. Therefore, any number in the range of 0.5 - 18m can be selected as the threshold to meet the requirement. In the present invention, 10m is selected.

[0104] Further, step 2 of the bridge alignment rapid detection realized by the bridge segmentation based on SCF-Bridge-Net in Component 2: The process of geometric information recognition considering structural features is as follows Figure 8 shown, and the specific method is as follows:

[0105] First, project the segment point cloud onto a local plane to convert the cylinder recognition in the 3D point cloud into circle recognition in the plane point cloud. Randomly sample 3 points from the projected plane point cloud and calculate the center coordinates (x i , y i ) and radius R i of the circle formed by these three points. Repeat this step N times to obtain samples of the radius; then perform probability density statistics and take the radius value R max and the center (x 0 , y 0 ) corresponding to the maximum probability density; subsequently, calculate the density gradient of the sling area; finally, calculate the geometric center of the cable clamp.

[0106] Example 1

[0107] In the Xianxinlu Yangtze River Bridge, the proposed scheme was tested, as shown Figure 9 below. The main girder of the bridge adopts a steel box girder form, with a total of 97 segments, 190 cable clamps, and 190 slings. Figure 9 (a) and 9(b) show that the DJ-M300 is equipped with a mobile GPS and a laser scanning system. Reflective sheets with a size of 10 cm * 10 cm are evenly arranged in the bridge span direction, and then a reflective prism is arranged at the center of the reflective sheet. The total station is used to measure the spatial position information here, and the relative geometric relationship is as shown Figure 9 (c). Figure 9 (d) is a Leica Ts40 Total station with an angular measurement accuracy of 8" and a distance measurement accuracy of 1.2 mm.

[0108] The method of measuring the line shape using an airborne radar has two sets of data acquisition terminals, namely an unmanned aerial vehicle (UAV) - mounted high - performance INS / GNSS system and an airborne lidar. For the INS / GNSS system, the collected data includes acceleration data, UAV attitude data, position data collected by the airborne GPS, and observation station data collected by the ground static base station. Correspondingly, the data processing is also divided into two parts, namely the data processing of the INS / GNSS system and the data processing of the airborne radar data. The processing of the INS / GNSS system is divided into two steps. First, the observation data of the static GPS base station is imported and differenced with the airborne mobile GPS data to obtain a differential positioning result that eliminates GPS errors. The positioning error of this result is generally less than 1 cm. Then, the differential positioning result is fused with the acceleration data of the INS / GNSS system to further improve the positioning accuracy and the scanning data accuracy. Finally, in the extraction of the positioning result, it is set to export only the positioning trajectories with an observation error less than 1 cm, which serves as the basis for accurately establishing the point cloud model of the entire bridge.

[0109] Based on the rough - scanned point cloud model obtained above, the point cloud of this model is relatively sparse and there are a large number of un - scanned areas. Based on this, 20 flight path points are evenly selected in the mid - span, and the flight paths are set according to the aforementioned effective areas, as Figure 10 (a) shows that the red represents the flight path above the main cable of the bridge, the green represents the flight path above the main girder, and the blue represents the flight path above the downstream main cable. The flight speed is 5 m / s, and the total flight time is about 40 minutes. Figure 10 (b) shows the flight state of the UAV on the green flight path on the girder, Figure 10 (c) is the actual on - site picture of the UAV on the blue flight path downstream of the bridge.

[0110] After obtaining the refined point cloud model, principal component analysis is performed on the point cloud of the bridge to find the normal vector perpendicular to the bridge deck, and the local area of the entire bridge is extracted according to the direction of the normal vector. The entire point cloud model (about 30 million points) is input into the network SCF - Bridge - Net. The data set consists of 650 cable clamp point clouds and 650 double suspension cables of three under - construction or in - service suspension bridges. First, it goes through the point cloud reduction module. After determining the three main directions of the bridge, the z - coordinates of the highest point of the main girder and the lowest point of the main cable are selected, and then 50 cm in this direction is selected to find all the local suspension cables. Secondly, all the suspension cable areas are divided into 190 clusters using Euclidean clustering, and the point cloud mean values of all clusters are calculated. The Euclidean clustering distance parameter is set to 0.5 m. Finally, the input point cloud of the entire bridge is reduced to a point cloud with a quantity of three million through the mean center, and then sent to the SCF module for subsequent recognition. The network recognition results of the original SCF - Net for processing large - scale point clouds are compared, as Figure 11As shown. The identification results of the cable clamps numbered 25, 45, and 65 are selected for comparison. It can be seen from the figure that the network proposed in this paper has better identification results. To better compare the accuracy and efficiency of the proposed SCF-Bridge-Net in this paper, other network models for large-scale point cloud processing are compared, and the identification results are shown in Figure 12 as follows.

[0111] As can be seen from Figure 12 , the algorithm proposed in this paper has 94.5 and 93.2 in terms of cable clamp, sling, and average accuracy rate respectively, which are the highest among all networks. In addition, due to the addition of a point cloud reduction module with very low computational complexity, the original number of point clouds is downsampled to 10% of the original number, so the computational time is greatly reduced, and the large-scale point cloud identification process is completed in only 320 s.

[0112] To verify the accuracy of the algorithm proposed in this paper, reflective sheets are uniformly arranged on the entire bridge, and the total station is used to measure the center position of the reflective sheet and the coordinates at the other two endpoints of the cable clamp (for calculating the inclination angle), and the total station data in the construction coordinate system is converted into the center position of the cable clamp through the geometric characteristics of the cable clamp. Figure 13 shows the inclination angles and total station coordinates of all cable clamps with reflective sheets calculated by the algorithm proposed in this problem, and compares them with the design coordinates. The maximum calculation error from the design results is 0.41°, and the average error is 0.25°, while the maximum error from the total station coordinates is only 0.2°, and the average error is 0.11°. The measurement accuracy meets the specification requirement of 0.5°, which proves the effectiveness of the method in measuring the inclination angle.

[0113] In summary, the specific embodiments verify the effectiveness of the solution proposed in the present invention and its applicability to complex projects.

[0114] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. Automatic calculation method of main cable shape of long-span suspension bridge based on airborne laser scanning, characterized by: Step S1, integrating high-precision inertial navigation, differential GNSS, airborne laser scanning equipment and RTK technology, and using a two-level route planning method to obtain a large-scale fine point cloud model: First-level route planning: Based on the airborne laser scanning equipment, route planning is carried out through the roughly scanned point cloud model. Second-level route planning: Based on the rough scanning route obtained by the first-level route, a high-precision inertial navigation system and differential GPS fusion method are used to achieve high-precision point cloud acquisition of kilometer-level large-span suspension bridges; Step S2, based on the SCF-Net network and combined with the structural characteristics of the suspension bridge, a SCF-Bridge-Net based on the suspension cable attention mechanism is proposed; The cable attention mechanism extracts all cable area points at the same height based on the structural characteristics of the suspension bridge, and implements spatial area classification in a two-dimensional plane based on Euclidean clustering and nearest neighbor algorithms, locking the target to be identified in a small range to improve recognition accuracy and time. SCF-Bridge-Net consists of a sling attention mechanism module and an SCF-Net network; SCF-Net improves the detection and classification of 3D objects by capturing the spatial context information in the point cloud. The classified spatial region point cloud is input into the network to obtain the accurate point cloud components of the suspension bridge. Step S3, using the suspension bridge point cloud components obtained in S2, the inclination angle, elevation and mileage information of all cable clamps are realized by a geometric information calculation method to achieve linear measurement of the bridge.

2. The method for automatically calculating the main cable shape of a long-span suspension bridge based on airborne laser scanning according to claim 1 is characterized in that: In step S1, the large-scale fine point cloud model is obtained based on the two-level route planning by pre-scanning: The first-level route planning obtains a rough scan model of the suspension bridge through rapid route scanning, and evenly selects coordinate points on the main cable in the WGS-84 coordinate system. The second-level route planning sets precise route points by ensuring that all targets are within the valid area.

3. The automatic calculation method of the main cable line shape of a long-span suspension bridge based on airborne laser scanning according to claim 1 is characterized in that: The high-precision inertial navigation system in step S1 is: a GNSS / INS combined positioning system, Get the WGS-84 coordinates of the antenna center through GNSS. The attitude information of the device can be obtained through INS.

4. The method for automatically calculating the main cable shape of a long-span suspension bridge based on airborne laser scanning according to claim 1 is characterized in that: In step S1, the differential GNSS is specifically a positioning system that uses RTK technology to eliminate positioning errors.

5. The method for automatically calculating the main cable shape of a long-span suspension bridge based on airborne laser scanning according to claim 1, characterized in that: The SCF-Bridge-Net based on the sling attention mechanism in step S2 includes: SCF-Net, sling attention mechanism module; The SCF-Net network uses neighboring point cloud information to obtain local information, and can also obtain global information based on the volume ratio of the region to the entire point cloud, making it suitable for large-scale point cloud segmentation scenarios; The sling attention mechanism module is implemented by classifying spatial regions in a two-dimensional plane based on Euclidean clustering and the nearest neighbor algorithm. The specific process is as follows: (1) Intercept the area between the lowest point of the cable clamp and the main beam along the bridge tower coordinate direction, and select the cable area within this z coordinate range; (2) Calculate the center of the selected sling area based on Euclidean clustering and nearest neighbor algorithm; First, all double slings are classified by Euclidean clustering method; the basic principle of Euclidean clustering is to use cross-tree nearest neighbor search to find the nearest k points for any point P, and cluster them into the set when the distance from the nearest point to point P is less than the set threshold; When the number of set points stops increasing, select a point outside the set and repeat the process; (3) Calculate the average value of all categories, take the average value of each cluster as the center of the square, and use 2m as the side length of the square to form an xy area to obtain the final core area.

6. The method for automatically calculating the main cable shape of a long-span suspension bridge based on airborne laser scanning according to claim 1, characterized in that: The step S3, the method for calculating the geometric information of the suspended bridge point cloud component specifically includes: Step S31, project the segment point cloud in the suspension bridge point cloud component onto a local plane, transform the cylinder recognition in the point cloud into the circle recognition in the plane point cloud, randomly sample 3 points from the projected plane point cloud, and calculate the center coordinates (x i ,y i ) and radius R i ; Repeat this step N times to get the radius sample; Step S32: Perform probability density statistics on all radius samples obtained, where the horizontal axis is the number of times and the vertical axis is the density. The radius value R corresponding to the maximum probability density is taken. max and the center of the circle (x0,y0); Step S33, after obtaining the segmented point cloud from the suspension bridge point cloud component, set a gradient along the bridge deck direction, calculate the number of point clouds contained in each gradient, and then select the density peak with the local maximum value as the location of the suspension cable, and calculate the density gradient of the suspension cable area; Step S34, in step S31, the plane centers of the circles on both sides of the cable clamp are calculated respectively, and the straight line connecting the two center lines is taken as the straight line where the cable clamp is located. Using the peak value of the cable density calculated in S33, the intersection of the cable center line and the aforementioned straight line is the geometric center of the cable clamp.

Images