Bridge deformation detection method based on three-dimensional laser scanning and multi-scale algorithm

Abstract

The invention provides a bridge deformation detection method based on three-dimensional laser scanning and a multi-scale algorithm, and the method specifically comprises the steps: carrying out the detection of the overall and local deformation of a bridge through employing the fusion of a three-dimensional laser scanning technology and the multi-scale algorithm, the method comprises the following steps: firstly, by introducing a geometric constraint registration algorithm, a cascade noise reduction model and a multi-stage slice center point offset analysis technology, realizing synchronous accurate quantification of local settlement of a bridge floor and overall inclination of a bridge pier; and in combination with load distribution verification, the spatial correlation between the deformation rule and the traffic load is revealed, and a scientific basis is provided for bridge safety assessment and reinforcement decision making. The bridge safety monitoring precision is remarkably improved, intelligent system development is supported, and guidance is provided for bridge maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional laser scanning monitoring of bridge structures, and in particular to a bridge deformation quantitative detection method based on multi-scale point cloud registration, noise reduction and slice analysis. BACKGROUND

[0002] As the core hub of transportation infrastructure, the structural safety of a bridge is directly related to the smoothness of the transportation network and public safety.

[0003] Traditional deformation monitoring techniques, such as total station and GNSS, rely on single-point displacement measurement and have significant limitations: total station monitoring is low in efficiency, GNSS is affected by multipath effects, resulting in limited accuracy, and the data collection rate is low, making it difficult to capture high-frequency dynamic deformation.

[0004] Although three-dimensional laser scanning technology (TLS) can obtain high-density point clouds, it still faces the following bottlenecks in practical applications: Low data processing efficiency: the noise reduction, registration and feature extraction of massive point clouds take too long, affecting real-time monitoring capabilities; Insufficient multi-time registration accuracy: existing 4PCS and ICP algorithms are prone to local optimal solutions when the point cloud overlap rate is low; Difficulty in local deformation quantification: traditional point cloud direct comparison and digital elevation model difference methods cannot simultaneously achieve high-precision analysis of detailed deformations such as bridge deck deflection and pier column inclination; Lack of dynamic load correlation: there is a lack of systematic coupling analysis of deformation data with traffic loads and geological conditions.

[0005] In addition, existing technologies such as the M3C2 algorithm can detect complex terrain changes, but are not optimized for bridge structure geometric features; cross-section extraction algorithms can analyze deflection and torsion angle, but do not solve the problem of multi-level slice perpendicularity quantification for bridge piers.

[0006] Therefore, there is an urgent need for a bridge intelligent monitoring method that integrates efficient data processing, multi-scale deformation analysis and load correlation analysis. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a bridge deformation monitoring method that integrates multi-scale registration, curvature adaptive noise reduction and multi-level slice quantitative analysis, thereby providing a scientific basis for bridge safety evaluation and reinforcement decision-making.

[0008] The technical problem of the present application is solved by the following technical solution: The bridge deformation detection method based on three-dimensional laser scanning and multi-scale algorithm provided by the application is specifically as follows: three-dimensional laser scanning technology and multi-scale algorithm are adopted to detect the overall and local deformation of a bridge; in the process, geometric constraint registration algorithm, cascaded noise reduction model and multi-level slice center point offset analysis technology are introduced to realize synchronous and accurate quantification of local settlement of a bridge deck and overall inclination of a pier; in combination with load distribution verification, the spatial correlation of deformation law and traffic load is revealed, and scientific basis is provided for bridge safety evaluation and reinforcement decision.

[0009] The geometric constraint registration algorithm can be introduced by the following method: First, a complete bridge point cloud model is generated by using Cyclone software, then statistical filtering and noise reduction are performed on the original point cloud in the model, then point cloud downsampling based on the octree method is performed to retain the point cloud with the smallest distance, then geometric constraint point cloud registration and deformation alignment are used to obtain a given rigid body transformation And The root mean square error of the source point cloud and the target point cloud.

[0010] The geometric constraint registration algorithm adopted in the application is used for high-precision registration of multi-source remote sensing images, and has the effects of improving registration accuracy and enhancing anti-noise performance.

[0011] The following method can be used to perform statistical filtering and noise reduction on the original point cloud: The statistical filtering and noise reduction are performed on the original point cloud, the average Euclidean distance of each point Pi and its nearest 50 neighborhood points is calculated, and the average Euclidean distance of each point Pi and its nearest n points in the point set is calculated: , In the formula, d i represents the average Euclidean distance of the i-th target point in the point cloud and its n nearest neighbor points, n represents the number of nearest neighbor points participating in the calculation, d in represents the Euclidean distance from the target point i to its n-th neighbor point.

[0012] The global mean μ and the standard deviation σ are calculated: , In the formula, d i represents the average neighborhood distance of the i-th point, N represents the total amount of point cloud, μ is the global mean, and σ is the standard deviation.

[0013] The threshold value , is set, and the size of and is compared, if it is greater than the threshold value, the corresponding point is removed.

[0014] The original point cloud is subjected to statistical filtering and noise reduction, which is used for removing outliers and noise interference, and has the effects of improving point cloud data quality and enhancing subsequent processing precision.

[0015] The geometric constraint point cloud registration and deformation alignment can be performed by the following method: A two-stage registration method of manual coarse registration and iterative closest point (ICP) fine registration is adopted, and no artificial setting of markers or target points is required during registration, and the point cloud is taken as the registration marker with small geometric deformation and artificial acquisition of the pier.

[0016] The cascade noise reduction model can be established by the following method: Due to the irregularity of the shape of the point cloud model, the statistical filtering SOR method is used for noise reduction processing; 6 adjacent points are selected as the reference when calculating the average distance, and the threshold is determined according to the standard deviation of the average distance of the point cloud, and finally set to 1; By analyzing the distance distribution of the target point and its adjacent points, the noise points deviating from the statistical law are identified and removed by using the local neighborhood statistical characteristics.

[0017] The cascade noise reduction model is used for multi-stage noise suppression and signal enhancement, and has the effects of gradually optimizing the noise reduction performance and preserving key details.

[0018] In the initial stage of the statistical filtering algorithm, the neighborhood number and distance threshold coefficient need to be manually inputted, so as to preliminarily filter out significant outliers and adapt to different point cloud density distribution; the subsequent stage is automatically processed by the least square method algorithm, so as to smooth and optimize the point cloud data and preserve effective geometric features.

[0019] The following method can be used to establish and analyze the multi-level slice center point: The front and rear pier columns are selected as the deformation monitoring objects, the point cloud of the pier is subjected to pruning and noise reduction processing based on the Open3D point cloud library, and then the multi-layer slicing method is used to extract the geometric features: the geometric center of each slice is extracted by polygon plane fitting, and the perpendicularity of the pier center axis is calculated by the least square method; when detecting the perpendicularity of the pier, the pier column point cloud is first subjected to continuous equidistant slicing and artificial processing, then the slice is subjected to plane fitting, and the fitting center point coordinates are automatically calculated by means of the algorithm.

[0020] The multi-level slice center point is used for segmented analysis of the geometric deformation and offset trend of the pier column, and has the effects of improving the detection accuracy and positioning the local tilt defects.

[0021] The present application can verify the load distribution of the local settlement of the bridge deck by the following method: After the point cloud preprocessing, the deformation detection is realized by multi-scale registration comparison: first, the point clouds before and after deformation are down-sampled based on the octree method to generate core point clouds with uniform density; then, coarse registration and ICP algorithm are used to complete fine registration, and the measured point cloud and the design point cloud are unified to the same coordinate system; In order to quantify the overall deformation, the M3C2 algorithm is used to optimize the neighborhood search efficiency by octree, and the displacement distribution is represented by chroma mapping, and the displacement distribution is represented by red, green and blue colors according to the reference color scale, and the green area corresponds to the smallest displacement value and the widest distribution, indicating that the overall deformation of the whole bridge is the smallest; the positive and negative signs of the displacement are determined by the direction of the normal vector of the fitting plane at each point in the compared point cloud; In the initial stage of the overall deformation detection of the bridge, the core distance and the projection radius of the M3C2 algorithm need to be manually set; then, the distance field between the point clouds is automatically calculated by using the starting M3C2 algorithm program.

[0022] The present application verifies the load distribution of the local settlement of the bridge deck, which is used for evaluating the structural bearing capacity and damage evolution law, has the effect of accurately positioning the weak area and optimizing the maintenance decision.

[0023] The present application can verify the load distribution of the overall inclination of the pier by the following method: The multi-scale analysis method is used to verify the load distribution of the overall inclination of the pier: multi-level continuous slices are extracted along the height direction of the pier, the coordinates of the center points of each slice are calculated by least square plane fitting, the deformation curve is constructed, the deformation difference of the pier before and after is compared, and the load influence is verified in combination with the traffic load distribution and the geological environment of the bridge.

[0024] The present application verifies the load distribution of the overall inclination of the pier, which is used for evaluating the stability and safety margin of the pier under complex stress conditions, has the effect of quantifying the inclination cause and warning the potential instability risk.

[0025] The bridge deformation detection method based on three-dimensional laser scanning and multi-scale algorithm provided by the present application is used for long-term safety monitoring of complex structures such as long-span bridges, cable-stayed bridges and suspension bridges.

[0026] Compared with the prior art, the present application has the following main advantages: The application innovatively proposes a multi-scale analysis method combining a multi-scale model inter-cloud contrast (M3C2) algorithm and a least square plane fitting, aiming to break through the bottleneck of low efficiency and limited spatial resolution of traditional single-point monitoring technology. The method realizes qualitative full-field deformation detection through the M3C2 algorithm, and completes quantitative feature extraction by combining the least square plane fitting. In the empirical research of the river-crossing bridge in Hubei Province, China, a Leica RTC360 scanner is used to obtain high-density point cloud data (>500pt / m²), and a curvature adaptive cascade denoising technology (noise removal rate >98%, structure feature retention rate >95%) and an octree simplification algorithm are used for data preprocessing. By extracting the multi-level slice features of the bridge deck and the piers, the global deformation trend and the local deformation are successfully analyzed simultaneously. The results show that the left bridge deck area presents an average settlement of 8.2mm, and the overall vertical deformation presents a "left low and right high" feature, and the piers show obvious inclination, especially the maximum offset of the rear pier column reaches 182.2mm, which is larger than the deformation of the front pier. The micro-settlement error of the bridge deck is ±1.2mm, and the inclination error of the pier is ±2.8mm, which meets the "China Highway Bridge Maintenance Specification" (JTG H11-2004) and the American Association of State Highway and Transportation Officials (AASHTO) standard, and the multi-scale algorithm reaches the engineering level precision. When the point cloud density is >500pt / m², the M3C2 algorithm realizes a spatial resolution of 0.5mm, and can perform sub-millimeter full-field analysis on complex scenes. The multi-scale analysis method significantly improves the precision of bridge safety monitoring, supports the development of intelligent systems, and provides guidance for bridge maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 Fig. 1 is a field observation real scene graph for a three-dimensional laser scanner, wherein: (a) is a field scanning real scene graph, and (b) is a three-dimensional laser scanning monitoring point.

[0028] Figure 2 Fig. 2 is a deformation detection research object, wherein: (a) is an original point cloud, and (b) is a research object.

[0029] Figure 3 Fig. 3 is a denoising and sampling point cloud model schematic diagram, wherein: (a) is a denoised point cloud model, and (b) is a sampled point cloud model.

[0030] Figure 4 Fig. 4 is an ICP registration graph, wherein: (a) is a registered point cloud model, (b) is an undeformed point cloud model, and (c) is a deformed point cloud model.

[0031] Figure 5 Fig. 5 is a M3C2 algorithm recognition result graph, wherein: (a) is a bridge deck plate settlement trend distribution graph, and (b) is a bridge deck plate uplift trend distribution graph.

[0032] Figure 6 Fig. 1 is a schematic diagram of a cross-section of a transverse diaphragm and longitudinal rib point cloud model, wherein: (a) is a schematic diagram of a cross-section of a transverse diaphragm point cloud model, and (b) is a schematic diagram of a cross-section of a longitudinal rib point cloud model.

[0033] Figure 7 Fig. 2 is a schematic diagram of a transverse diaphragm and longitudinal rib detection object number, wherein: (a) is a transverse diaphragm monitoring object number #1-#7, and (b) is a longitudinal rib detection object number #1-#10.

[0034] Figure 8 Fig. 3 is a schematic diagram of a transverse diaphragm and longitudinal rib plate bottom point cloud, wherein: (a) is a transverse diaphragm bottom point cloud number #1-#9, and (b) is a longitudinal rib bottom point cloud number #1-#8.

[0035] Figure 9 Fig. 4 is a coordinate value of a transverse diaphragm and longitudinal rib bottom point cloud fitting center point in the X-Z direction, wherein: (a) is a coordinate value of a transverse diaphragm bottom point cloud #1~#7 fitting center point in the X-Z direction, and (b) is a coordinate value of a longitudinal rib bottom point cloud #1~#10 fitting center point in the X-Z direction.

[0036] Figure 10 Fig. 5 is a transverse diaphragm deflection curve diagram.

[0037] Figure 11 Fig. 6 is a longitudinal rib plate deflection curve diagram.

[0038] Figure 12 Fig. 7 is a schematic diagram of a bridge pier column number #1~#4, wherein: (a) is a bridge pier number, and (b) is a pier column point cloud number.

[0039] Figure 13 Fig. 8 is a schematic diagram of a bridge pier column point continuous point cloud slice, wherein: (a) is a front pier column continuous point cloud slice #1~#2, and (b) is a rear pier column continuous point cloud slice #3~#4.

[0040] Figure 14 Fig. 9 is a deformation amount of a pier column #1~#4 slice center point in the X, Y direction, wherein: (a) is a deformation amount of a pier column #1~#4 slice center point in the X direction, and (b) is a deformation amount of a pier column #1~#4 slice center point in the Y direction.

[0041] Figure 15 Fig. 10 is a three-dimensional scatter diagram of a pier column #1~#4 slice fitting center, wherein: (a) is a comparison diagram of a #1 bridge pier column continuous slice center point before and after deformation, (b) is a comparison diagram of a #2 bridge pier column continuous slice center point before and after deformation, (c) is a comparison diagram of a #3 bridge pier column continuous slice center point before and after deformation, and (d) is a comparison diagram of a #4 bridge pier column continuous slice center point before and after deformation. DETAILED DESCRIPTION

[0042] The bridge deformation detection method based on three-dimensional laser scanning and multi-scale algorithm provided by the present invention is a high-precision detection method for the overall and local deformation of bridges that integrates three-dimensional laser scanning technology and multi-scale algorithm. This method first introduces a geometric constraint registration algorithm, a cascade noise reduction model and a multi-level slice center point offset analysis technology to achieve synchronous and accurate quantification of local settlement of the bridge deck and overall inclination of the bridge piers; combined with load distribution verification, it reveals the spatial correlation between deformation laws and traffic loads, providing a scientific basis for bridge safety assessment and reinforcement decisions.

[0043] The method of the present invention includes the following steps: 3D laser scanning data acquisition and system configuration, point cloud preprocessing and statistical filtering noise reduction, point cloud downsampling based on the octree method, and geometrically constrained point cloud registration and deformation alignment. 1. 3D laser scanning data acquisition and system configuration: A Leica RTC360 3D laser scanner (ranging accuracy ±1mm @ 10m, scanning rate ≥ 2 million points / second) was used to scan the entire bridge cross-section. Eight scanning stations were located along the bridge's centerline, front pier, and rear pier, with station spacing ≤50m and scanning time ≤5 minutes per station. Targets, spherical reflectors with a diameter of 100mm, were placed on the bridge's transverse diaphragms and pier foundations to facilitate automatic cloud stitching of multiple stations. A complete bridge point cloud model was generated using Cyclone software, with a point cloud density ≥10^4 points / m² and a total of approximately 79.54 million points (using a cross-river bridge in Hubei as an example).

[0044] 2. Point cloud preprocessing and statistical filtering noise reduction: Perform statistical filtering and noise reduction on the original point cloud and calculate each point P i The average Euclidean distance of each point P in the point set to its nearest 50 neighboring points i The average Euclidean distance to the nearest n points: , Where: d i represents the average Euclidean distance between the i-th target point and its n nearest neighbor points in the point cloud, n represents the number of nearest neighbor points involved in the calculation, d in Represents the Euclidean distance from target point i to its nth neighbor.

[0045] Calculate the global mean μ and standard deviation σ: , Where: d i Represents the average neighborhood distance of the i-th point, and N represents the total number of point clouds. Set the threshold , ,Compare and If the size is greater than the threshold, the corresponding point will be removed.

[0046] 3. Point cloud downsampling based on octree method: Firstly, the overall range of the point cloud is determined, and the limits of length, width and height are determined by the difference between the maximum and minimum values of the coordinate axes x, y and z. Secondly, the octree is represented in layers as follows: , In the formula, is the minimum cube size, is the maximum cube size, and then the position of the cube in which the point cloud is located is determined: , where p, l, i are index values, x, y, z are point cloud coordinates, and finally the minimum cube center coordinates are calculated according to the formula to retain the point cloud with the smallest distance.

[0047] 4. Geometrically constrained point cloud registration and deformation alignment: An improved ICP algorithm is used to align the point clouds before and after deformation, and the objective function is to minimize the error of the rotation matrix R and the translation vector t. , In the formula, si is any point i in the source point cloud; M is the target point cloud set; m c(i) is the nearest neighbor point in the target point cloud corresponding to the source point s i ; c(i) is the corresponding point of any point i in the source point cloud S in the target point cloud M; Ns is the total number of points in the source point cloud; the ICP algorithm uses iterative solution of the rotation matrix R and the translation vector t.

[0048] Given the rigid body transformation and The root mean square error of the source point cloud and the target point cloud: , In the formula: R k , t k is the rotation matrix and translation vector of the kth iteration; k is the root mean square error (RMSE) of the current iteration, reflecting the registration accuracy; min is the preset error threshold, when k ≤ min iteration is terminated.

[0049] where the iteration termination condition is that for a given root mean square . If ≤ , or the number of iterations reaches the maximum number of iterations, the algorithm terminates; otherwise, the iteration continues.

[0050] 5. Bridge multi-level slice deformation analysis: Bridge deck deflection detection: take the cross bulkhead bottom point cloud slice every 2m along the bridge longitudinal direction, adopt the least square method to fit the plane equation z=a 0x +a 1y +a2, calculate the coordinate deviation of the center point , the precision is ±0.8mm; the verticality of the pier is detected: take the pier slice every 0.5m along the pier height, fit the polygon geometric center and linearly regress the pier axis, and the verticality error is less than 0.01%.

[0051] The distance LM3C2 between the two periods of point clouds is calculated by combining the M3C2 algorithm, and the roughness analysis formula is:

[0052] In the formula, It is the distance between the kth point within the radius D / 2 and the best fitting plane; It is the average distance between the best fitting plane and all point clouds within the D / 2 radius; M is the total amount of point clouds distributed within the D / 2 radius.

[0053] The bridge deformation detection method based on three-dimensional laser scanning and multi-scale algorithm provided by the application is suitable for long-term safety monitoring of complex structures such as large-span bridges, cable-stayed bridges and suspension bridges.

[0054] Next, taking a certain approach bridge segment of a river-crossing bridge in Hubei Province as an example, the application will be further described in combination with the drawings, but this does not constitute any limitation on the application.

[0055] As shown by Figure 1 It can be seen that the main girder of the segment adopts an innovative "hybrid" structure, the middle part is a steel box girder (total length 906.4m), and the two ends are concrete box girders, forming a bolted and welded hybrid system, the research object segment is about 42m long and about 27m wide; Figure 4 The field scene clearly shows the structural characteristics and environmental differences of the segment, and the front pier is located on the shore and the rear pier is submerged in water, providing a typical scene for subsequent deformation monitoring.

[0056] As shown by Figure 2 It can be seen that in order to meet the high-precision deformation monitoring requirements of the approach bridge, the application adopts Leica RTC360 three-dimensional laser scanner to implement multi-station data acquisition, Figure 2 (a) shows the spliced original point cloud (initial data amount 79,540,078 points); in view of the challenges of scanning accuracy affected by distance and environmental interference, the application implements intelligent point cloud simplification processing, under the premise of retaining ≥95% key structural features, the data compression ratio reaches 97.52% (1,969,086 points after simplification), and finally obtains Figure 5 (b) shows the efficient box girder segment point cloud model.

[0057] In view of the irregularity of the shape of the point cloud model, a statistical filtering SOR method is used for point cloud noise reduction processing; 6 adjacent points are selected as the reference when calculating the average distance, and the threshold is determined according to the standard deviation of the average distance of the point cloud, and finally set to 1. This method is based on the statistical characteristics of the local neighborhood (calculating the neighborhood point distance mean μ and standard deviation σ), accurately identifying and removing noise points (such as outliers) that deviate from the statistical law. As a classic point cloud noise reduction algorithm, the SOR method has wide applicability, strong robustness and high precision, and is a reliable technical solution for processing irregular point cloud models. For example Figure 3 (a) shows that the denoised point cloud retains the complete geometric shape of the model while achieving effective filtering. The statistical filtering algorithm used for point cloud noise reduction requires manual input of the number of neighborhoods and the distance threshold coefficient in the initial stage, and the subsequent stage is automatically processed by the algorithm.

[0058] As shown in Figure 3 (b), in order to break through the bottleneck of massive point cloud calculation, the present application adopts an octree downsampling method, which realizes the balance between calculation efficiency and feature fidelity by recursively dividing the space grid and retaining representative points (such as grid center points); experiments show that this method compresses 97.52% of the data amount while the key feature point cloud retention rate is ≥95%, providing a lightweight high-fidelity data basis for subsequent analysis.

[0059] As shown in Figure 4 , the present application innovatively adopts a non-target two-stage registration method, first using a geometrically stable pier as a natural reference for coarse registration, and then performing ICP fine registration on the deformed measured point cloud model (Pm) after noise reduction and sampling processing Figure 4 (c) and the design point cloud model (Pd) before deformation Figure 4 (b). The registered point cloud is shown in Figure 4 (a). This method avoids manual calibration errors and realizes fully automatic high-precision spatial matching.

[0060] As shown in Figure 5 , the displacement field analysis based on the M3C2 algorithm shows that the overall deformation of the whole bridge is small (green dominant area), and at the same time reveals the deformation law of the bridge deck: taking the central axis of the bridge deck as the dividing line, Figure 5 (a) shows that the larger area on the left side of the bridge presents a positive displacement value, showing a settlement phenomenon compared with the undeformed bridge deck; and Figure 5 (b) shows that the area on the right side of the bridge appears a negative displacement value, indicating that this side has a certain degree of uplift relative to the left side. In addition, the bridge pier also shows a significant deviation.

[0061] The present application proposes a method of manually extracting feature point clouds, which is aimed at bridge deck structure deformation monitoring, and Figure 6 , Figure 7 ,Figure 8 It is known that the transverse diaphragm and the longitudinal rib bottom point cloud are monitored objects, and the center point coordinates are accurately extracted by least square plane fitting as the deflection monitoring points; 7 transverse diaphragm sections are selected Figure 7 (a) 9 monitoring objects are selected Figure 8 (a) The vertical displacement of the X-Z plane (transverse bridge deformation) is analyzed, and 10 longitudinal rib sections are selected Figure 7 (b) 8 monitoring objects are selected Figure 8 (b) The vertical displacement of the Y-Z plane (along the bridge direction) is analyzed, which realizes the grid-based accurate monitoring of the bridge deck deformation and eliminates the subjective error of manual point selection.

[0062] Load distribution verification of local settlement of bridge deck: After point cloud preprocessing, deformation detection is realized by multi-scale registration comparison. First, based on the octree method, the point clouds before and after deformation are down-sampled to generate a uniform density of core point clouds; then, coarse registration and ICP algorithm are used to complete fine registration, and the measured point cloud and the design point cloud are unified to the same coordinate system. In order to quantify the overall deformation, the M3C2 algorithm is used. The efficiency of neighborhood search is optimized by octree, and the displacement distribution is represented by chroma mapping. As shown in Figure 8 , the displacement distribution is represented by red, green and blue colors according to the reference color scale. The green area corresponds to the smallest displacement value and the widest distribution, indicating the smallest overall deformation of the whole bridge. The sign (positive or negative) of the displacement is determined by the normal vector of the fitting plane in the direction of each point in the compared point cloud. The positive M3C2 distance represents the settlement of the bridge deck, and the negative distance represents the lifting of the bridge deck. In the initial stage of overall deformation detection of the bridge, the core distance and projection radius of the M3C2 algorithm need to be manually set. Then, the distance field between the point clouds is automatically calculated by using the start algorithm program. The algorithm running time is 3 minutes.

[0063] The multi-scale analysis method is used to verify the load distribution of the overall inclination of the pier. Multi-level continuous slices are extracted along the height direction of the pier, the center point coordinates of each slice are calculated by least square plane fitting, the deformation curve is constructed, the deformation difference of the pier before and after is compared, and the load influence is verified combined with the traffic load distribution and the geological environment of the bridge.

[0064] As Figure 9 known, first, the Open3D library is used for cutting and noise reduction preprocessing of the pier point cloud, second, the multi-level slice method is used along the elevation direction (the slice interval and thickness parameters are shown in the schematic diagram), then the geometric center point of each slice is calculated by polygon plane fitting, and finally the center axis of the pier is fitted based on the least square method and the perpendicularity and offset are calculated. This process systematically solves the problem of quantifying complex structure deformation.

[0065] As Figure 10It can be seen that the deflection curves of diaphragms #1~#7 reveal that the elevation of the monitoring points in the transverse direction of the bridge (vertical to the central axis) shows an overall upward trend, that is, the elevation of the left side of the bridge deck is significantly lower than that of the right side, which is consistent with the Figure 5 (a) The observed subsidence trend on the left is consistent.

[0066] Depend on Figure 11 It can be seen that the deflection curves of longitudinal ribs #1~#10 show that the elevation of the monitoring point shows an overall downward trend along the bridge direction (along the central axis), which is consistent with the Figure 5 (b) The observed settlement trend on the left side is consistent, confirming that the bridge deck has an overall settlement along the bridge direction.

[0067] Depend on Figure 12 As shown in (a) and (b), differentiated point cloud segmentation was performed on the front pier columns #1 and #2 and the rear pier columns #3 and #4.

[0068] Depend on Figure 13 As shown in (a) and (b), after cropping and denoising the pier point cloud using the Open3D point cloud library, a multi-level slicing method was used to extract geometric features: the front pier column was sliced ​​into 14 slices with a thickness of 0.05m and a spacing of 0.13m, and the rear pier column was sliced ​​into 16 slices with a thickness of 0.125m and a spacing of 0.45m. Verticality was calculated based on the axis fitted at the slice center point to quantify the pier column inclination.

[0069] The geometric center of each slice is extracted through polygonal plane fitting, and the least squares method is used to calculate the verticality of the pier's central axis. To test the verticality of the pier, the pier column point cloud is first manually processed by continuously equidistantly slicing. Then, a plane fitting is performed on the slices, and the coordinates of the fitting center point are automatically calculated using an algorithm. The entire algorithm takes 6 minutes to run.

[0070] Depend on Figure 14 As shown in (a) and (b), as the number of point cloud slices increases along the elevation direction, the deformation of the slice center coordinates of piers #1 to #4 in the X and Y directions shows an increasing trend, which is in line with the deformation law of the piers. In addition, the deformation of the slice center in the Y direction is greater than that in the X direction. The deformation of the rear pier columns #3 and #4 in the X direction is within 60 mm, and the deformation in the Y direction is within 180 mm. The deformation of the front pier columns #1 and #2 in the X direction is within 7 mm, and the deformation in the Y direction is within 30 mm.

[0071] Depend on Figure 15 As shown in (a)-(d), the inclination angle, offset, and verticality of the rear pier columns #3 and #4 are significantly greater than those of the front pier columns #1 and #2. This indicates that the offset deformation of the front pier columns is smaller than that of the rear pier columns.

[0072] The bridge deformation monitoring method based on three-dimensional laser scanning and multi-scale algorithm provided by the application, aiming at the problems of insufficient single-point precision of traditional monitoring technology, low overlap point cloud registration error of three-dimensional laser scanning, asymmetric deformation sensitivity and limitations of multi-scale analysis capability, etc., innovatively proposes a multi-scale analysis method combining M3C2 algorithm and least square plane fitting. The method guarantees the local point cloud feature matching accuracy through M3C2 algorithm, establishes a global deformation reference surface by combining least square plane fitting, forms a double analysis mechanism of "macro trend decoupling-micro feature analysis", realizes the synchronous and accurate quantification of sub-millimeter local deformation and centimeter global deformation, breaks through the path dependence of existing methods on symmetric deformation, and significantly improves the accuracy and reliability of multi-scale deformation analysis in complex scenes.

[0073] The main performance is to solve the following technical problems: 1. The traditional registration algorithm has large error in low overlap point cloud, which leads to misjudgment of deformation trend; 2. The existing noise reduction method causes loss of key geometric features, affecting the local deformation quantification accuracy; 3. The traditional method cannot realize high-precision detection of bridge deck deflection and bridge pier verticality synchronously; 4. The dynamic coupling analysis of monitoring results and traffic load distribution is insufficient, the deformation-load correlation is missing, and it cannot guide the reinforcement decision; 5. The existing algorithm is not optimized for bridge geometric features, resulting in large error in cylindrical pier inclination detection.

Claims

1. A bridge deformation detection method based on 3D laser scanning and multi-scale algorithm is characterized by: The overall and local deformation of the bridge are detected by integrating 3D laser scanning technology with a multi-scale algorithm. During the process, a geometric constraint registration algorithm, a cascade noise reduction model, and a multi-level slice center point offset analysis technology are first introduced to achieve simultaneous and accurate quantification of the local settlement of the bridge deck and the overall inclination of the piers. Combined with load distribution verification, the spatial correlation between deformation patterns and traffic loads is revealed, providing a scientific basis for bridge safety assessment and reinforcement decisions.

2. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 1 is characterized in that: The geometric constraint registration algorithm is introduced using the following method: First, a complete bridge point cloud model is generated using Cyclone software. Then, the original point cloud in the model is subjected to statistical filtering and noise reduction. Then, the point cloud is downsampled based on the octree method, and the point cloud with the smallest distance is retained. Then, geometric constraint point cloud registration and deformation alignment are used to obtain the given rigid body transformation. and The root mean square error between the source point cloud and the target point cloud.

3. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 2 is characterized in that: The following method is used to perform statistical filtering and noise reduction on the original point cloud: Perform statistical filtering and noise reduction on the original point cloud, calculate the average Euclidean distance between each point Pi and its 50 nearest neighboring points, and the average Euclidean distance between each point Pi and the nearest n points in the point set: ， Where: d i represents the average Euclidean distance between the i-th target point and its n nearest neighbor points in the point cloud, n represents the number of nearest neighbor points involved in the calculation, d in Represents the Euclidean distance from the target point i to its nth neighbor point, Calculate the global mean μ and standard deviation σ: ， Where: d i represents the average neighborhood distance of the i-th point, and N represents the total amount of point cloud. Setting thresholds , ,Compare and If the size is greater than the threshold, the corresponding point will be removed.

4. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 2 is characterized in that: The following methods are used for geometrically constrained point cloud registration and deformation alignment: A two-stage registration method consisting of manual coarse registration and iterative closest point ICP fine registration is adopted. No manual setting of markers or target points is required during registration. Bridge piers with small geometric deformation and manually acquired features are used as registration markers for the point cloud. It should be noted that the original point cloud data needs to be downsampled before ICP registration. ICP registration is performed on the designed point cloud model before deformation and the measured point cloud model after deformation after noise reduction and sampling.

5. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 1 is characterized in that: The cascade noise reduction model is established using the following method: In view of the irregular shape of the point cloud model, the statistical filtering SOR method is used for noise reduction. When calculating the average distance, 6 neighboring points are selected as the benchmark, and the threshold is determined according to the standard deviation of the average distance of the point cloud, and is finally set to 1. By analyzing the distance distribution between the target point and its neighboring points, the local neighborhood statistical characteristics are used to identify and eliminate noise points that deviate from the statistical laws.

6. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 5 is characterized in that: The statistical filtering algorithm used requires manual input of the number of neighbors and the distance threshold coefficient in the initial stage, and the subsequent stages are automatically processed by the least squares algorithm.

7. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 1 is characterized in that: The following methods are used to establish and analyze the center points of multi-level slices: The front and rear piers were selected as deformation monitoring objects. After pruning and denoising the pier point cloud based on the Open3D point cloud library, a multi-layer slicing method was used to extract geometric features. The geometric center of each slice was extracted through polygonal plane fitting, and the verticality of the pier center axis was calculated using the least squares method. When detecting the verticality of bridge piers, the pier column point cloud is first manually processed by continuous equidistant slices, and then the slices are plane fitted, and the coordinates of the fitting center point are automatically calculated with the help of an algorithm.

8. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 1 is characterized in that: The following methods are used to verify the load distribution of the local settlement of the bridge deck: After point cloud preprocessing, deformation detection is achieved through multi-scale registration and comparison. First, the point cloud before and after deformation is downsampled based on the octree method to generate a core point cloud with uniform density. Then, coarse registration and the ICP algorithm are used to complete fine registration, unifying the measured point cloud and the designed point cloud into the same coordinate system. To quantify the overall deformation, the M3C2 algorithm was used. The neighborhood search efficiency was optimized through an octree, and the displacement distribution was represented by a chromaticity map. The displacement distribution was represented by red, green, and blue according to the reference color scale. The green area corresponds to the smallest displacement value and the widest distribution, indicating that the overall deformation of the entire bridge is the smallest. The sign of the displacement is determined by the direction of the normal vector of the fitting plane at each point in the comparison point cloud. In the initial stage of bridge overall deformation detection, the core distance and projection radius of the M3C2 algorithm need to be manually set; then, the distance field between point clouds is automatically calculated using the startup M3C2 algorithm program.

9. The bridge deformation detection method based on 3D laser scanning and multi-scale algorithm according to claim 1 is characterized in that: The load distribution of the overall inclination of the pier is verified by the following method: A multi-scale analysis method is used to verify the load distribution of the overall inclination of the pier: multiple continuous slices are extracted along the height direction of the pier, and the coordinates of the center point of each slice are calculated through least squares plane fitting. The deformation curve is constructed, and the deformation difference between the front and rear piers is compared. The load impact is verified by combining the traffic load distribution and the geological environment of the bridge.

10. A bridge deformation detection method based on 3D laser scanning and multi-scale algorithm is characterized by: Used for long-term safety monitoring of complex structures such as large-span bridges, cable-stayed bridges, and suspension bridges.

Images