Bridge erection deviation intelligent correction method based on point cloud data

Abstract

This invention discloses an intelligent method for correcting bridge erection deviations based on point cloud data, specifically relating to the field of bridge construction surveying and intelligent deviation correction technology. The method includes acquiring depth images of bridge components at each unloading stage during the lifting point unloading process, and performing 3D reconstruction on the depth images of each unloading stage to generate corresponding point cloud data for that stage. By performing stage division, point cloud reconstruction, key area extraction, and stable migration point identification on the multi-view depth images during the lifting point unloading process, the apparent point cloud formed by temporary loads, temporary constraints, and local occlusion is separated from the actual installation deviation. Furthermore, by registering the separated actual installation deviation with the bridge component design model and generating installation deviation correction data, intelligent deviation correction during the bridge erection process is achieved.

Description

Technical Field

[0001] This invention relates to the field of bridge construction surveying and intelligent deviation correction technology, and more specifically, to an intelligent deviation correction method for bridge erection based on point cloud data. Background Technology

[0002] In the process of correcting deviations during bridge erection, existing technologies mainly address the issue of whether the spatial position of beam segments, steel box girders, or precast bridge decks after hoisting and placement matches the design posture. The common approach is to acquire the three-dimensional data of the current component through laser scanning, visual measurement, or point cloud acquisition, and then register the acquired point cloud with the design model, reference point cloud, or point cloud from the previous moment. Based on this, translational deviations, rotational deviations, or local height differences are calculated, and edge computing is combined to quickly output the correction basis on-site. For example, in the scenario of hoisting and erecting segmental beams for high-pier bridges in mountainous areas or bridges spanning rivers, it is often necessary to complete the on-site deviation judgment before communication conditions are limited, the hoisting window is short, the tension of the hoisting points changes continuously, and the beam segment is not completely unloaded, in order to avoid prolonged suspension or repeated hoisting. However, under this constraint, the existing practice will consistently expose a key defect: before and after the lifting points are gradually unloaded, the deviation direction and amount obtained by point cloud registration will change significantly. On-site, it can be directly observed that after the first fine adjustment, the splicing edge that was originally close to closed will reopen, the local fitting state of the support area will change repeatedly, or the overall fitting error will decrease, but the actual assembly state will worsen. The reason is that the existing method usually takes the appearance of the point cloud formed by the combined influence of temporary loads, temporary constraints and local occlusion as the real installation state and participates in the deviation calculation, which leads to the output correction basis being mixed with non-real installation deviation components. The technical problem to be solved by this application is: how to separate the actual installation deviation from the point cloud appearance morphology affected by temporary loads and temporary constraints during the bridge erection process, and to perform intelligent correction based on the separated actual installation deviation. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent method for correcting bridge erection deviations based on point cloud data. This method involves segmenting multi-view depth images during the unloading process, reconstructing point clouds, extracting key regions, and identifying stable migration points. This separates the apparent point cloud data, influenced by temporary loads, temporary constraints, and local occlusions, from the actual installation deviations. Furthermore, by registering the separated actual installation deviations with the bridge component design model and generating installation deviation correction data, intelligent correction during bridge erection is achieved, thus solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent correction of bridge erection deviations based on point cloud data includes: S1. Collect depth images of each unloading stage of the bridge components during the unloading process at the lifting point, and perform three-dimensional reconstruction on the depth images of each unloading stage to generate point cloud data for the corresponding unloading stage. S2. Transform the point cloud data of each unloading stage to the same erection coordinate system, and extract the splicing end face point set, support area point set and lifting point neighborhood point set from the transformed point cloud data of each unloading stage, and output the key area point group. S3. Establish corresponding point pairs according to the point adjacency relationship of the corresponding key area point sets in the adjacent unloading stages, and perform displacement direction comparison on each corresponding point pair. Collect the corresponding point pairs with continuous and consistent displacement directions into a stable migration point set, and collect the corresponding point pairs with reverse displacement directions into a rebound point set. Output the stable migration point set and the rebound point set. S4. Perform point cloud registration between the stable migration point set and the corresponding region in the bridge component design model, and perform culling processing on the registered region based on the rebound point set to output the actual installation deviation set. S5. Perform deviation decomposition on the splicing end face deviation, support area deviation and lifting point neighborhood deviation of the actual installation deviation concentration, generate bridge component installation deviation correction data, and output the bridge component installation correction results based on the installation deviation correction data.

[0005] In a preferred embodiment, S1 includes: S1-1. Synchronously acquire multi-view depth images and corresponding lifting point load data at each sampling time. Calculate the lifting point load difference between adjacent sampling times in chronological order. When the lifting point load difference is less than zero, record the multi-view depth image of the next sampling time as the new unloading stage image. Output each unloading stage image group.

[0006] In a preferred embodiment, S1 further includes: S1-2. For the multi-view depth images in each unloading stage image group, extract the support area edge and the stitching end face edge respectively. Then, perform position overlap comparison between the support area edge in any view and the support area edge in other views one by one. Then, perform direction consistency comparison between the corresponding stitching end face edges one by one. Select the view transformation relationship with the largest position overlap and the smallest direction difference to perform alignment on each view depth image and output the fused depth image of each unloading stage. S1-3. Perform three-dimensional back projection on the depth values ​​of each pixel in the fused depth image of each unloading stage according to the imaging parameters of the corresponding viewpoint to generate spatial points corresponding to each pixel. Then, compare the spatial points with the outer contour range of the bridge component point by point and delete the spatial points located outside the outer contour range of the bridge component, and output the point cloud data of the corresponding unloading stage.

[0007] In a preferred embodiment, S2 includes: S2-1. For the point cloud data of each unloading stage, respectively count the support contact edges and suspension edges in the outer contour of the point cloud, calculate the intersection direction between each support contact edge and the connection direction between each suspension edge, and set the plane where the lowest point of the support contact edge is located as the reference plane, the connection direction as the longitudinal reference direction, and the direction in the reference plane that is perpendicular to the longitudinal reference direction as the transverse reference direction. In this way, construct the stage erection coordinate system corresponding to the point cloud data of each unloading stage, and then transform the point cloud data of each unloading stage to a unified erection coordinate system to output a unified coordinate point cloud group.

[0008] In a preferred embodiment, S2 further includes: S2-2. For the point cloud data of each unloading stage in the coordinate unified point cloud group, calculate the change in the cross-sectional area of ​​the point cloud segment by segment along the longitudinal reference direction, and record the point cloud cross-section with the area change greater than the area change on the adjacent two sides as the candidate cross-section of the end face. Then, perform plane fitting degree calculation on the points in each candidate cross-section of the end face according to the transverse reference direction and the vertical reference direction. Record the continuous point area with the largest plane fitting degree as the splicing end face point set, and output the splicing end face point set. S2-3. For the point cloud data of each unloading stage in the coordinate unified point cloud group, calculate the distance from each point to the reference plane and record the continuous point area with the smallest distance as the support area point set. Then, take the connection position between the suspension edge of each lifting point and the point cloud as the center, extract the continuous point area within the predetermined range as the lifting point neighborhood point set, and combine the splicing end face point set, the support area point set, and the lifting point neighborhood point set into the key area point group of the corresponding unloading stage.

[0009] In a preferred embodiment, S3 includes: S3-1. For the splicing end face point set, support area point set and lifting point neighborhood point set in the adjacent unloading stage, construct candidate corresponding score values ​​according to the normal distance from the point to the local tangent plane, the boundary distance from the point to the adjacent boundary line and the point density of the neighborhood where the point is located. Write the point pair with the largest score value into the candidate corresponding pair table. Then perform a two-way reverse lookup consistency check on each point pair in the candidate corresponding pair table. Write the point pairs with consistent reverse lookup results into the initial corresponding point pair set and write the point pairs with inconsistent reverse lookup results into the point pair set to be re-checked. S3-2. For each point pair in the initial set of corresponding point pairs, calculate the component signs of the point pair displacement vector in the longitudinal reference direction, the lateral reference direction, and the vertical reference direction, generate direction symbol codes, and construct direction windows in each key region according to the three adjacent unloading stages. Perform continuity counting, cross-viewpoint coincidence counting, and neighborhood co-directional counting on the direction symbol codes of the same point pair within the direction window to form the migration confidence value of the corresponding point pair, and then write the migration confidence value into the state table of the corresponding point pair.

[0010] In a preferred embodiment, S3 further includes: S3-3. For each point pair in the corresponding point pair state table, first generate a gating token based on its migration confidence value, and then perform regional consistency check, neighborhood propagation consistency check, and point pair back-check for the point pair with the gating token. When the regional consistency check and the neighborhood propagation consistency check pass, the point pair is written into the stable candidate table. When the regional consistency check fails and the point pair back-check shows that the point pair has a displacement component sign reversal in the later unloading stage, the point pair is written into the bounce candidate table. When the two types of check results conflict, the point pair is written into the state lock table and a rollback re-check is triggered. S3-4. Aggregate each point pair in the stable candidate table according to the neighborhood connectivity relationship. The set of point pairs whose number of aggregated point pairs is more than half of the total number of point pairs in the corresponding key area and whose total migration confidence value is greater than the total migration confidence value in the bounce candidate table in the same area is recorded as the stable migration point set. Aggregate each point pair in the bounce candidate table according to the same aggregation rule. The set of point pairs whose number of aggregated point pairs is more than one-third of the total number of point pairs in the corresponding key area and whose displacement component signs have been reversed at least twice is recorded as the bounce point set. Then, merge the point pairs in the state lock table into the side with the most contact number according to the difference between the number of neighborhood contacts with the stable migration point set and the bounce point set. Output the stable migration point set and the bounce point set.

[0011] In a preferred embodiment, S4 includes: S4-1. Map the set of stable migration points to the corresponding regions in the bridge component design model according to their respective key regions. Calculate the normal distance from each stable migration point to the fitting surface of the corresponding region and the tangential distance along the boundary direction of the corresponding region. Determine the model corresponding point of each stable migration point based on the criterion of minimizing the sum of the normal distance and the tangential distance, and output the corresponding point group of the region.

[0012] In a preferred embodiment, S4 further includes: S4-2. For each stable migration point in the corresponding point group of the region and the corresponding point of the model, iteratively solve the region transformation parameters according to the rule of minimizing the sum of squared normal distances and minimizing the change in spacing between adjacent point pairs. Then, delete the model region corresponding to the location of the bounce point set and its adjacent range from the corresponding point group of the region and re-execute the iterative solution to output the registration transformation results. S4-3. Apply the registration transformation results to the stable migration point set, calculate the coordinate differences between each stable migration point and the corresponding point in the model in the longitudinal, lateral, and vertical reference directions, and group the coordinate differences into splicing end face deviation, support area deviation, and lifting point neighborhood deviation according to key areas, and output the actual installation deviation set.

[0013] In a preferred embodiment, S5 includes: S5-1. Calculate the longitudinal reference direction component, transverse reference direction component, and vertical reference direction component for the splicing end face deviation, support area deviation, and lifting point neighborhood deviation in the actual installation deviation concentration, respectively. Construct a decomposition sequence according to the influence of splicing end face deviation on end face closure, the influence of support area deviation on support fit, and the influence of lifting point neighborhood deviation on lifting point force balance. Merge each direction component into end face correction amount, support correction amount, and lifting point correction amount according to the decomposition sequence, and output installation deviation correction data. S5-2. Generate a correction execution sequence by first eliminating the end face correction amount, support correction amount, and lifting point correction amount in the installation deviation correction data, then restoring the support fit, and finally correcting the lifting point neighborhood offset, and write the correction direction and correction amplitude corresponding to the correction execution sequence into the bridge component installation correction results.

[0014] The technical effects and advantages of this invention are as follows: By dividing the unloading stage according to the load change of the lifting point under edge computing conditions, establishing stable migration points and rebound points, and removing rebound areas from the registration, the apparent view cloud affected by temporary loads and temporary constraints can be separated from the actual installation deviation, thereby relatively suppressing the jump in deviation direction and deviation amount. By performing alignment and fusion on multi-view depth images at edge computing nodes according to the support area edge and the stitching end face edge, and reconstructing them into point cloud data of the corresponding unloading stage, the impact of single-view occlusion, local missing data and acquisition disturbance on the quality of stage point cloud can be relatively reduced, thereby relatively improving the data consistency between stages. By constructing a phased coordinate system based on the support contact edge and the suspension edge of the lifting point, and unifying the point clouds of each unloading stage into the same coordinate system, the situation where coordinate reference differences are misjudged as installation deviations can be reduced, thereby relatively improving the comparability of subsequent displacement comparisons and region extraction. By extracting the splicing end face point set, support area point set, and lifting point neighborhood point set from the point cloud under a unified erection coordinate system, the corresponding point establishment and deviation analysis can be limited to the area directly related to the erection state, thereby relatively reducing the interference of irrelevant surface points on the deviation judgment results. By constructing candidate correspondence scores based on normal distance, boundary distance, and point density difference, and combining bidirectional reverse lookup, directional window, and migration confidence value to filter corresponding points, the number of local mismatches, supplementary points, and outliers entering the effective registration basis can be relatively reduced, thereby relatively improving the stability of the correspondence relationship. By further decomposing the actual installation deviation into end face correction, support correction, and lifting point correction, and generating a correction execution sequence in the order of end face first, support then lifting point, the correction basis of the edge calculation output can be kept consistent with the evolution of the installation state, thereby relatively improving the pertinence and coherence of on-site adjustments. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides an intelligent method for correcting bridge erection deviations based on point cloud data, comprising: S1. Collect depth images of each unloading stage of the bridge components during the unloading process at the lifting point, and perform three-dimensional reconstruction on the depth images of each unloading stage to generate point cloud data for the corresponding unloading stage. In this embodiment, the purpose of steps S1-1 to S1-3 is to convert multi-view depth images of bridge components during the unloading process at the lifting points into point cloud data corresponding to each unloading stage. Since bridge components are affected by load changes, local occlusion, and viewing angle differences during the unloading process, data from a single moment or a single viewpoint cannot stably reflect the morphological correspondence between different unloading stages. Therefore, it is necessary to first divide the unloading stages according to the load changes at the lifting points, then align and fuse the multi-view depth images within each unloading stage, and finally complete the three-dimensional back projection based on the imaging parameters and delete out-of-contour spatial points to output the corresponding unloading stage point cloud data. This implementation process includes the following steps: In S1-1, multi-view depth images and corresponding lifting point load data are first synchronously acquired at each sampling time according to a preset sampling period, and the images and load data at the same sampling time are written to the same timestamp. Then, for adjacent sampling times arranged in chronological order, the lifting point load value at the later sampling time is subtracted from the lifting point load value at the previous sampling time to obtain the lifting point load difference. When the lifting point load difference is less than zero, the multi-view depth image at the later sampling time is recorded as a new unloading stage image. When multiple consecutive sampling times satisfy the condition that the lifting point load difference is less than zero, they are recorded as different unloading stages in chronological order, thereby outputting each unloading stage image group. To avoid misclassification caused by instantaneous fluctuations, the lifting point load difference at three adjacent sampling times can be continuously judged. Only when at least two adjacent differences are less than zero is it recorded as a new unloading stage image. In S1-2, viewpoint alignment and fusion are performed on the multi-view depth images in each unloading stage image group. Specifically, the support area edge and the splicing end face edge are extracted from the depth image of each viewpoint. The support area edge is obtained by extracting boundary points near the bottom of the bridge component with continuous depth changes, and the splicing end face edge is obtained by extracting boundary points in the end region of the bridge component with obvious depth abrupt changes and continuous boundaries. Then, one viewpoint is selected as the reference viewpoint, and the support area edges in the other views are transformed one by one to the reference viewpoint. The positional overlap between the transformed support area edge and the support area edge of the reference viewpoint is calculated. The positional overlap is expressed as the distance between the midpoints of the two sets of edge points. The proportion of points smaller than the predetermined edge tolerance to the total number of edge points is used to represent the distance. Subsequently, for viewpoint transformation relationships with high positional overlap, the directional difference between the corresponding splicing end face edge and the reference view splicing end face edge is further calculated. The directional difference is obtained by fitting the main direction lines of the two sets of splicing end face edges and calculating the included angle. Finally, the viewpoint transformation relationship with the largest positional overlap and the smallest directional difference is selected, and uniform alignment is performed on the depth images of each viewpoint. The images are then fused according to the depth values ​​of the same spatial position, and the corresponding unloading stage fused depth image is output. For spatial positions observed only from a single viewpoint, they are only retained if the continuous area around them is within the outline of the bridge component. In steps S1-3, 3D backprojection and contour constraint filtering are performed on the fused depth images of each unloading stage, outputting the corresponding unloading stage point cloud data. Specifically, for each pixel in the fused depth image, its pixel coordinates and pixel depth value are read, and 3D backprojection is performed in conjunction with the imaging parameters of the corresponding viewpoint to convert the pixel into spatial point coordinates. The imaging parameters include at least the focal length parameter, principal point position parameter, and installation pose parameter of the depth imaging device relative to the acquisition site. After backprojection of all pixels, the initial spatial point set for the unloading stage is obtained. Subsequently, this initial spatial point set is compared point by point with the outer contour range of the bridge component. In comparison, the outer contour range of bridge components is preferably determined based on the projected contour of the bridge component design model in the current acquisition direction, or it can be determined based on the outer boundary of the continuous region of the bridge component in the fused depth image. When a spatial point is projected outside the outer contour range of the bridge component, the spatial point is deleted. When a spatial point is located within the outer contour range of the bridge component and maintains a continuous connection with adjacent spatial points, the spatial point is retained as a valid point. All retained valid points are output according to the spatial coordinate set, which yields the corresponding unloading stage point cloud data. If necessary, neighborhood smoothing can also be performed on the output point cloud data to reduce isolated point offset. Through the above implementation process, the continuous sampling results can be divided into multiple unloading stages based on the load difference at the lifting points. Then, the multi-view depth images are aligned by jointly constraining the edges of the support area and the splicing end face. Point cloud data corresponding to the unloading stages is generated through 3D back projection and contour constraints, thus providing a consistent data foundation for subsequent coordinate unification, key area extraction, and deviation determination. Simultaneously, the above processing can reduce the impact of occlusion, local depth errors, and lifting disturbances on the quality of the stage point cloud. In practical applications, taking the segmental beam lifting setup as an example, when the bridge component is about to be positioned and the lifting points begin to unload in stages, multiple depth imaging devices are deployed on both sides and above the beam segment. Load sensors are deployed at the main lifting points. Multi-view depth images and lifting point load values ​​are acquired synchronously at each sampling time. When the total load of the lifting points at the next sampling time is detected to be lower than that at the previous sampling time, the next sampling time is recorded as a new unloading stage. The edges of the support area and the splicing end face are extracted from the multi-view depth images of this stage. The view alignment is completed based on the position overlap and direction difference. Then, three-dimensional back projection is performed pixel by pixel on the fused depth image. Spatial points that exceed the projection range of the beam segment design outline are deleted, and finally the point cloud data of this unloading stage is obtained. After processing multiple unloading stages in this way, a stage point cloud group can be formed for use in subsequent steps.

[0018] S2. Transform the point cloud data of each unloading stage to the same erection coordinate system, and extract the splicing end face point set, support area point set and lifting point neighborhood point set from the transformed point cloud data of each unloading stage, and output the key area point group. In this embodiment, the purpose of steps S2-1 to S2-3 is to unify the point cloud data of each unloading stage under the same erection coordinate system, and to extract the point sets of the splicing end face, the support area, and the neighboring point of the lifting point to form the key area point groups for the corresponding unloading stage. Since the point cloud data of each unloading stage is affected by component micro-sway and fusion errors, if the coordinate reference is not unified first, the spatial position and directional relationship between different unloading stages cannot be directly compared; if the key areas are not extracted, subsequent deviation identification will be mixed with irrelevant surface points. Therefore, it is necessary to first construct a stage erection coordinate system, and then perform coordinate unification and key area extraction. This implementation process includes the following steps: In S2-1, the outer contour of the point cloud is first extracted from the point cloud data of each unloading stage. In the outer contour of the point cloud, the boundary point area located in the lower part of the point cloud, continuously distributed and extending for a long time along the length of the component, is selected as the candidate area for support contact edge. Boundary lines are fitted to each candidate area, and the boundary lines with the same direction and near the lowest position are recorded as support contact edges. Then, the boundary points on all support contact edges whose vertical coordinates are within the lowest 10% range are selected to fit a plane, and this fitted plane is used as the reference plane. At the same time, the boundary point area located in the upper part of the point cloud, close to the connection position of the lifting device and corresponding to the force position of the lifting point is selected as the suspension edge of the lifting point. For each candidate area, the center point position is calculated, and then the center points are connected according to the arrangement order of the lifting points. The resulting main extension direction is used as the connecting direction between the suspension edges of the lifting points, and this connecting direction is set as the longitudinal reference direction. In the reference plane, the direction perpendicular to the longitudinal reference direction is taken as the transverse reference direction, and the direction perpendicular to the reference plane and orthogonal to the longitudinal and transverse reference directions is taken as the vertical reference direction. Thus, the stage erection coordinate system corresponding to the unloading stage point cloud data is constructed. Subsequently, the spatial points in the point cloud data of each unloading stage are converted to the unified erection coordinate system, and the coordinate unified point cloud group is output. In S2-2, the splicing end face point set is extracted from the point cloud data of each unloading stage in the coordinate unified point cloud group. Specifically, slicing is performed segment by segment along the longitudinal reference direction with a fixed step size. For each spatial point in the slice, the cross-sectional area of ​​the point cloud is calculated in the cross-sectional plane formed by the transverse and vertical reference directions. Subsequently, for each slice arranged according to the longitudinal reference direction, the area difference between the current slice and the previous slice, and between the current slice and the next slice, is calculated. When the area change between the current slice and the slices before and after is greater than the area change of the corresponding slices on both sides, the result is considered complete. The current slice is recorded as a candidate end face section. For spatial points in each candidate end face section, the end face plane is first fitted, then the absolute value of the distance from each spatial point to the end face plane is calculated, and its average value is obtained. At the same time, the proportion of spatial points with an absolute distance less than the predetermined fitting distance to the total number of points in the candidate end face section is counted, and the proportion and the reciprocal of the average distance are used as the plane fitting degree characterization value. When the plane fitting degree characterization value corresponding to a certain continuous point area is the largest, and the continuous point area meets the minimum size requirement of the end face, the continuous point area is recorded as the splicing end face point set. In S2-3, the support area point set and the lifting point neighborhood point set are further extracted from the point cloud data of each unloading stage in the coordinate unified point cloud group, and key area point groups are formed for the corresponding unloading stage. Specifically, the distance from each spatial point in the point cloud data of each unloading stage to the reference plane is calculated, and the point area with the smallest distance and continuous spatial distribution is selected as the candidate point area of ​​the support area. If there are multiple candidate point areas of the support area, the continuous point area that corresponds to the design support position of the bridge component in the longitudinal reference direction and has the largest extension length in the transverse reference direction is selected as the support area point set. Subsequently, the suspension edges of each lifting point are used as the support area point set. The connection point with the point cloud is used as the center to extract the neighborhood point set of the lifting point. The connection point is determined by searching downward along the vertical reference direction from the center point of the candidate area of ​​the suspension edge of the lifting point to find the position where it first makes continuous contact with the outer contour of the bridge component. After determining the connection point, a continuous point area within a predetermined range is extracted as the neighborhood point set of the lifting point. After the above extraction is completed, the splicing end face point set, support area point set, and lifting point neighborhood point set in the same unloading stage are combined according to the region category to output the key region point group corresponding to the unloading stage. After each unloading stage is executed in sequence, the key region point group of all unloading stages is obtained. Through the above implementation process, the point cloud data of each unloading stage can be unified under the same erection coordinate system. Then, under the unified reference, the splicing end face point set, support area point set, and lifting point neighborhood point set that directly reflect the erection state of the bridge components can be extracted, thereby reducing the interference of irrelevant area points on the subsequent establishment of corresponding points and deviation calculation. In practical applications: taking the erection of precast segmental beams as an example, after obtaining the point cloud data of each unloading stage in S1, the continuous low boundary in the lower outer contour boundary is first extracted as the support contact edge, and the low boundary point on the support contact edge is fitted to the reference plane. Then, the upper part corresponding to the two main lifting points is used as the reference plane. The direction of the line connecting the boundary center points is used as the longitudinal reference direction to establish a phased erection coordinate system. Subsequently, all unloading phase point cloud data are uniformly transformed into the erection coordinate system corresponding to the first unloading phase, and the point cloud of each phase is sliced ​​along the longitudinal reference direction. Candidate end face sections are screened by the change in cross-sectional area, and the splicing end face point set is determined by the continuous point area with the largest planar fit. After that, the support area point set is determined by the continuous area with the minimum distance from the point to the reference plane, and the lifting point neighborhood point set is determined by the continuous spatial points around the lifting point connection position. Finally, the key area point group of each unloading phase is formed for use in subsequent steps.

[0019] S3. Establish corresponding point pairs according to the point adjacency relationship of the corresponding key area point sets in the adjacent unloading stages, and perform displacement direction comparison on each corresponding point pair. Collect the corresponding point pairs with continuous and consistent displacement directions into a stable migration point set, and collect the corresponding point pairs with reverse displacement directions into a rebound point set. Output the stable migration point set and the rebound point set. In this embodiment, the purpose of steps S3-1 to S3-4 is to establish stable corresponding points between adjacent unloading stages and to separate short-term rebound points, occlusion-filling points, and local mismatch points from the subsequent registration base. Since bridge components undergo continuous displacement during unloading at the lifting points, but local areas may experience rebound, point filling, and reconstruction errors, establishing point pairs based solely on the closest distance can easily misjudge spatial points at different physical locations as corresponding points, thus affecting the subsequent calculation of actual installation deviations. Therefore, it is necessary to first construct candidate corresponding score values ​​and screen initial corresponding point pairs, then form migration confidence values ​​based on the direction window, and output stable migration point sets and rebound point sets through gating verification, aggregation, and merging. This implementation process includes the following steps: In S3-1, candidate corresponding point pairs are first established within the point sets of the same category of key areas in adjacent unloading stages. Specifically, let the previous unloading stage be the first stage and the next unloading stage be the second stage. The point sets of splicing end faces, support areas, and hoisting points in the first stage are matched with the point sets of the corresponding categories in the second stage, respectively, without cross-matching between different categories. For any point to be matched in the first stage, candidate points within a predetermined search radius are extracted from the point set of the corresponding category in the second stage. Subsequently, a candidate corresponding score is calculated for each candidate point: first, a local tangent plane is fitted with the spatial points in the neighborhood of the point to be matched, and the vertical distance from the candidate point to the local tangent plane is calculated, denoted as the normal distance; then, the shortest distances from the candidate point and the point to be matched to the boundary line of the corresponding key area are calculated respectively, and the absolute value of the difference between the two is denoted as the boundary distance; then, the number of spatial points in the neighborhood of the candidate point and the neighborhood of the point to be matched are counted respectively, and the absolute value of the difference between the two is denoted as the boundary distance. The point density difference is recorded as the point density difference. Next, the normal distance is divided by the predetermined search radius to obtain the normal normal value; the boundary distance is divided by the maximum span of the corresponding key area boundary to obtain the boundary normal value; and the point density difference is divided by the number of neighboring points of the point to be matched to obtain the density normal value. Then, the candidate corresponding score is calculated by subtracting the weighted sum of the normal normal value, boundary normal value, and density normal value from one, with the normal normal value having the largest weight, followed by the boundary normal value, and the density normal value having the smallest weight. All candidate points corresponding to the same point to be matched are sorted according to their candidate corresponding score values. The candidate point with the largest score value is selected to form a candidate corresponding point pair with the point to be matched, and this pair is written into the candidate pair table. After completing the candidate pair table, a bidirectional reverse lookup consistency check is performed on each point pair in the table. When the lookup point is the same as the original point to be matched, the point pair is written into the initial corresponding point pair set; when the lookup point is different from the original point to be matched, the point pair is written into the point pair set to be rechecked. In S3-2, displacement direction comparisons are performed on each point pair in the initial set of corresponding point pairs, and a corresponding point pair state table is generated. Specifically, for each point pair, the coordinates of the first point in the first stage and the coordinates of the second point in the second stage are read. The displacement vector is obtained by subtracting the coordinates of the first point from the coordinates of the second point. The displacement vector is then projected onto the longitudinal, lateral, and vertical reference directions to obtain displacement components in the three directions. When a displacement component in a certain direction is greater than a predetermined component threshold, it is recorded as positive; when it is less than the opposite value of the threshold, it is recorded as negative; and when it is in between, it is recorded as zero. Direction symbol codes are generated by combining them in the order of longitudinal, lateral, and vertical. Subsequently, directions are constructed in each key area according to the three adjacent unloading stages. Within a window, for the same point pair, continuous counting, cross-view coincidence counting, and neighborhood co-directional counting are performed. Continuous counting is used to count the number of times the component signs in the three directions remain consistent in two consecutive stage differences. Cross-view coincidence counting is used to count the number of times the point pair falls within the same key region projection range in two or more original views. Neighborhood co-directional counting is used to count the percentage of points whose main displacement direction is consistent with the main displacement direction of other points in their neighborhood. The above three counts are then normalized and weighted summed to form the migration confidence value of the point pair. Finally, the point pair number, its key region category, direction sign code, three-directional displacement components, and migration confidence value are written into the corresponding point pair status table. In S3-3, gating is performed on each point pair in the corresponding point pair state table. Specifically, gating tokens are first generated based on the migration confidence value. When the migration confidence value is not less than the first gating value, a high-confidence gating token is generated. When the migration confidence value is less than the first gating value but not less than the second gating value, a gating token to be verified is generated. When the migration confidence value is less than the second gating value, no gating token is generated and it is directly written to the candidate area of ​​the state lock table. For point pairs with gating tokens, regional consistency verification, neighborhood propagation consistency verification, and point pair re-examination verification are further performed: regional consistency verification is used to determine whether the main displacement direction of the current point pair is consistent with the main displacement direction of most point pairs in the critical region; neighborhood propagation consistency verification is used to determine whether the main displacement direction of the current point pair is... Whether it maintains the same direction as the high-confidence point pair in its neighborhood; the point pair to be re-examined back-check is used to determine whether the point pair to be re-examined adjacent to the current point pair has re-established the corresponding relationship in the next unloading stage, and at least one main displacement component has changed from positive to negative or from negative to positive; when the regional consistency check passes and the neighborhood propagation consistency check passes, the point pair is written into the stable candidate table; when the regional consistency check fails and the point pair to be re-examined back-check shows that the point pair has a displacement component sign reversal in the next unloading stage, the point pair is written into the bounce candidate table; when the two types of check results conflict, the point pair is written into the state lock table and a back-down re-check is triggered. Back-down re-check means to supplement and construct a new direction window and recalculate the direction sign code and migration confidence value, and then perform the above check again; In S3-4, aggregation is performed on each pair of points in the stable candidate table and the bounce candidate table, and merging is performed on each pair of points in the state lock table. Specifically, aggregation is performed on each pair of points in the stable candidate table according to the neighborhood connectivity relationship. When the spatial distance between the first points of any two pairs of points is less than the predetermined aggregation distance, and the two pairs of points belong to the same critical region category and have the same principal displacement direction, they are grouped into the same stable candidate cluster. When the number of point pairs in a stable candidate cluster is more than half of the total number of point pairs in the corresponding critical region, and the sum of the migration confidence values ​​of all point pairs in the cluster is greater than the sum of the migration confidence values ​​in the bounce candidate table of the same region, the stable candidate cluster is recorded as a stable migration point set. Point pairs in the bounce candidate list are aggregated according to the same rules to form bounce candidate clusters. When the number of point pairs in a bounce candidate cluster is more than one-third of the total number of point pairs in the corresponding key region, and at least one principal displacement component symbol in the cluster is reversed twice or more in the continuous direction window, the bounce candidate cluster is recorded as a bounce point set. For each point pair in the state lock list, the number of neighborhood contacts between it and the stable migration point set and the bounce point set is counted, and they are merged into the side with the most contacts. When the number of contacts is equal, the average difference of its migration confidence value with the two point sets is compared, and it is merged into the side with the smaller average difference. After the aggregation and merging are completed, the stable migration point set and the bounce point set are output. Through the above implementation process, an initial correspondence satisfying geometric similarity and bidirectional consistency can be established between adjacent unloading stages. Then, using directional windows, migration confidence values, and gating checks, valid corresponding points with continuous and consistent displacement directions are separated from springback points, mismatched points, and points to be rechecked. This allows subsequent registration to be based on more stable spatial points that better reflect the actual installation process. Simultaneously, by forming a closed-loop decision chain through regional consistency checks, neighborhood propagation consistency checks, and recheck points for backfilling checks, the sensitivity of a single decision condition to local noise can be reduced. In practical applications, taking the splicing end face region between two consecutive unloading stages of a segmental beam as an example, first, any point is selected from the splicing end face point set of the previous unloading stage. Then, candidate points within the adjacent range of that point are searched in the splicing end face point set of the subsequent unloading stage, and the normal distance, boundary distance, and point density difference are used as the basis for the search. The candidate corresponding score value is calculated, and the point with the largest score value is selected to form a candidate corresponding point pair. Then, the initial corresponding point pair set and the point pair set to be re-examined are obtained through bidirectional reverse lookup consistency verification. Subsequently, the three-directional displacement components are calculated for each point pair in the initial corresponding point pair set and the direction symbol code is generated. Then, the direction window is formed by combining the three consecutive unloading stages before and after. The continuity count, cross-view coincidence count, and neighborhood co-directional count are statistically analyzed to form the migration confidence value. After that, the point pairs whose migration confidence value meets the gating condition are sequentially subjected to regional consistency verification, neighborhood propagation consistency verification, and point pair re-examined backfilling verification. The point pairs that pass the verification are written into the stable candidate table or the bounce candidate table respectively. Then, the point pairs in the two tables are aggregated according to the neighborhood connectivity relationship, and the point pairs in the state lock table are merged into the stable migration point set or the bounce point set. Finally, the stable migration point set and the bounce point set are obtained for use in subsequent steps.

[0020] S4. Perform point cloud registration between the stable migration point set and the corresponding region in the bridge component design model, and perform culling processing on the registered region based on the rebound point set to output the actual installation deviation set. In this embodiment, the purpose of steps S4-1 to S4-3 is to use the stable migration point set as an effective registration basis, establish a correspondence between it and the corresponding region in the bridge component design model, and output the actual installation deviation set after removing the region corresponding to the rebound point set. Since the previous steps have already grouped points with continuous and consistent displacement directions into the stable migration point set and points with reverse displacement directions into the rebound point set, this step no longer performs overall registration of the entire point cloud. Instead, it establishes model correspondence and solves region transformation parameters for three key regions: splicing end face, support area, and lifting point neighborhood, and deletes the model region corresponding to the rebound point set to avoid local rebound interfering with the registration results. This implementation process includes the following steps: In S4-1, the stable migration point set is first mapped to the corresponding region in the bridge component design model according to its key region. Specifically, the bridge component design model, which has been transformed to a unified erection coordinate system, is read. For each point in the stable migration point set, its key region category is read: points belonging to the splicing end face set are mapped to the splicing end face region; points belonging to the support area set are mapped to the support contact region; and points belonging to the lifting point neighborhood set are mapped to the lifting point connection neighborhood region. Subsequently, local model patches are extracted within each corresponding region. For any stable migration point, the normal distance from its location to the fitting surface of each local model patch is calculated. The tangential distance along the boundary direction of the projection point of the stable migration point onto the fitting surface of the local model patch is calculated. The normal distance and the tangential distance are then normalized and added together to form the corresponding evaluation value. The nearest model point on the local model patch with the smallest corresponding evaluation value is selected as the model corresponding point of the stable migration point. The stable migration point, the model corresponding point, the category of the key region to which it belongs, and the corresponding evaluation value are written into the region corresponding point group. To avoid cross-regional mismapping, the model corresponding point is only allowed to be written into the region corresponding point group when it is located within the boundary of the key region to which it belongs and is not less than the predetermined boundary protection distance from the outer boundary of the key region. In S4-2, the corresponding point group in the region is subjected to elimination processing and iterative registration solution, and the registration transformation result is output. Specifically, the model region to be eliminated and its adjacent range are first determined based on the rebound point set: the key region category to which each point in the rebound point set belongs and its spatial position in the unified coordinate system are read, and then the spatial position is projected onto the corresponding region in the bridge component design model to determine its corresponding model position; with the local model patch where the model position is located as the center, the predetermined range is expanded along the boundary direction and the adjacent normal direction to form the rebound elimination region, and The model corresponding points and their associated stable migration points falling within the rebound rejection region are removed from the region corresponding point group to obtain the effective region corresponding point group. Subsequently, the region transformation parameters are solved for each stable migration point and model corresponding point in the effective region corresponding point group. These parameters include translations along the longitudinal, lateral, and vertical reference directions, as well as rotations around these three reference directions. During the solution process, the joint objective is to minimize the sum of squared normal distances and the change in spacing between adjacent point pairs. The sum of squared normal distances is calculated by applying a method to all stable migration points... The normal distances from the pre-transformed point to the local model patch fitting surface where the corresponding point is located are squared and summed. The change in the distance between adjacent point pairs is obtained by calculating the absolute values ​​of the differences between the distances between the transformed points and the distances between the corresponding model points for spatially adjacent stable migration point pairs within the same key region, and summing them. In each iteration, all stable migration points are first transformed to temporary positions using the current region transformation parameters, and then the above two quantities are recalculated. The three translations and three rotations are adjusted in the direction that reduces the joint objective. When the joint objective decreases after a certain round of adjustments... If the value is small, the result of that round is retained as the new region transformation parameter. If the joint target does not decrease after a round of adjustment, the adjustment step size is reduced and the calculation is repeated. If the decrease in the joint target is less than the predetermined termination value for two consecutive rounds, or the number of iteration rounds reaches the predetermined upper limit, the iteration is stopped and the current region transformation parameter is output as the registration transformation result. To reflect the impact of the elimination process on the registration result, an initial iteration can be performed on the corresponding point group of the region in the non-elimination state, and then the corresponding points of the region in the bounce elimination region are deleted, and the initial registration transformation result is used as the starting value for resolving. In S4-3, the registration transformation results are applied to the stable migration point set, and the actual installation deviation set is output according to the key area. Specifically, for each stable migration point in the stable migration point set, its spatial coordinates are translated and rotated according to the registration transformation results to obtain the transformed stable migration point position. Then, the model corresponding point position of the stable migration point in the corresponding point group of the region is read, and the transformed stable migration point position is subtracted from the model corresponding point position to calculate the coordinate difference in the longitudinal, lateral, and vertical reference directions, respectively. Subsequently, the coordinate difference of all stable migration points in the same key area category is grouped by region, and the signed average value in the three directions is calculated to represent the overall offset direction and total offset of the key area. The offset is calculated, and then the point with the largest absolute value of the coordinate difference in the three directions and its maximum difference are counted to characterize the local extreme value deviation of the key area. Then, the three-direction signed average coordinate difference of all stable migration points in the splicing end face area and its maximum coordinate difference in the three directions are combined to form the splicing end face deviation. The corresponding results of all stable migration points in the support area are combined to form the support area deviation. The corresponding results of all stable migration points in the lifting point neighborhood area are combined to form the lifting point neighborhood deviation. These are written into the real installation deviation set according to a unified data format. The unified data format includes at least the key area category identifier, longitudinal reference direction deviation component, lateral reference direction deviation component, vertical reference direction deviation component, and the local maximum deviation value and corresponding point identifier in the three directions. Through the above implementation process, an effective correspondence can be established between the stable migration point set and the bridge component design model. After the model region corresponding to the rebound point set and its adjacent range are eliminated, a registration transformation result that better reflects the current actual installation state of the bridge component can be obtained, thus avoiding the mixing of local rebound into the registration process. At the same time, by using the sum of squares of the normal distance and the change in the spacing between adjacent point pairs to jointly constrain the solution of the region transformation parameters, both the local fitting accuracy and the geometric preservation relationship within the key area can be taken into account. Furthermore, by aggregating the coordinate differences after registration according to the splicing end face, support area, and lifting point neighborhood, a direct basis can be provided for subsequent deviation decomposition and correction output. In practical applications: taking the unloading process after the bridge segment beam is nearly in place as an example, after obtaining the stable migration point set and the rebound point set, the stable migration points in the splicing end face are first mapped... The system projects the data onto the splicing end face region in the design model, mapping the stable migration points in the support area to the support contact region in the design model, and mapping the stable migration points in the lifting point neighborhood to the lifting point connection neighborhood region in the design model. The corresponding model points for each stable migration point are determined by comparing the normal and tangential distances. Subsequently, the model positions corresponding to the springback point set and their adjacent ranges are removed from the corresponding point group in the region. The translation and rotation amounts are then iteratively solved for the remaining stable migration points and their corresponding model points until the joint target no longer decreases significantly. Afterward, the obtained registration transformation results are applied to all stable migration points, and the coordinate differences between them and their corresponding model points in the longitudinal, lateral, and vertical reference directions are calculated. These differences are then aggregated according to the splicing end face, support area, and lifting point neighborhood to form a set of actual installation deviations.

[0021] S5. Perform deviation decomposition on the splicing end face deviation, support area deviation and lifting point neighborhood deviation of the actual installation deviation concentration, generate bridge component installation deviation correction data, and output the bridge component installation correction results based on the installation deviation correction data. In this embodiment, the purpose of steps S5-1 to S5-2 is to convert the actual installation deviation set into installation deviation correction data that can be directly used for on-site adjustments, and further form bridge component installation correction results with clear order, direction, and amplitude. Since the actual installation deviation set only reflects the deviation status of three key areas—the splicing end face, the support area, and the lifting point neighborhood—it cannot directly correspond to which part should be adjusted first, in which direction, and by how much on-site. Therefore, it is necessary to first calculate the impact of the three types of deviations on end face closure, support fit, and lifting point force balance, and then form the correction amount and correction execution sequence accordingly. This implementation process includes the following steps: In S5-1, the longitudinal reference direction deviation component, lateral reference direction deviation component, vertical reference direction deviation component, and local maximum deviation value are first read for the splicing end face deviation, support area deviation, and lifting point neighborhood deviation in the actual installation deviation concentration. For the splicing end face deviation, the absolute value of the longitudinal reference direction deviation component is added to the absolute value of the lateral reference direction deviation component to obtain the end face position offset. Then, the end face position offset is combined with the absolute value of the vertical reference direction deviation component in a weighted summation manner to form the end face closure influence quantity, in which the vertical reference direction deviation component has a higher weight than the other two direction components. For the support area deviation, the average value of the vertical reference direction deviation component at each point in the support area is calculated to obtain the overall support suspension amount. Then, it is added to the local maximum vertical deviation value to form the support fit influence quantity. For the lifting point neighborhood deviation, different... The average deviation difference between continuous point areas near the lifting point connection location in the longitudinal and vertical reference directions is calculated, and the absolute values ​​of both are combined with the local maximum deviation value to form the lifting point force balance influence quantity. After completing the calculation of the three types of influence quantities, the deviation decomposition order is arranged from the largest to the smallest influence quantity. When the difference between the two types of influence quantities is less than the predetermined influence quantity interval, the order of end face closure priority, support fit center, and lifting point force balance remains unchanged. Subsequently, the components of the splicing end face deviation that directly affect the end face fit are merged into the end face correction quantity, the components of the support area deviation that directly affect the support fit are merged into the support correction quantity, and the components of the lifting point neighborhood deviation that directly affect the residual offset and force balance of the lifting point connection area are merged into the lifting point correction quantity. The key area category, main correction direction, correction component value of each direction, and priority identifier are written into the installation deviation correction data. In S5-2, the end face correction, support correction, and lifting point correction in the installation deviation correction data are sequentially arranged and their amplitudes are converted to output the bridge component installation correction results. Specifically, a correction execution sequence is generated in the order of first eliminating end face openings, then restoring support fit, and finally correcting lifting point neighborhood offsets. For each type of correction in the correction execution sequence, the correction direction and correction amplitude are determined: when the correction component in a certain direction is greater than zero, the correction direction is reversed; when the correction component in a certain direction is less than zero, the correction direction is forward; when the absolute value of the correction component in a certain direction is less than a predetermined correction threshold, the correction direction is reversed. The correction action is not output separately; the correction amplitude is taken as the absolute value of the correction component in the corresponding direction; when the absolute value is less than the predetermined safe correction amplitude, it is directly used as the single correction amplitude; when the absolute value is greater than or equal to the predetermined safe correction amplitude, it is split into two or more sub-correction amplitudes, and the retest requirement is retained after each sub-correction; then, the key area category, execution order, correction direction in each direction, correction amplitude in each direction, and whether it is executed in stages are written into the bridge component installation correction results; when the retest result after a certain step of correction does not meet the predetermined deviation convergence condition, the subsequent correction is paused, and the corresponding correction amount is updated according to the retest result; Through the above implementation process, the actual installation deviation set can be converted into installation deviation correction data with clear direction, clear amplitude, and clear sequence, and further form bridge component installation correction results that can be directly used for on-site fine-tuning. Simultaneously, by calculating the end-face closure influence, support fit influence, and lifting point force balance influence respectively, the correction actions can be prioritized around the deviations that have the greatest impact on the current installation state. Combined with correction amplitude limits and phased execution rules, the feasibility and stability of on-site correction can be improved. In practical applications: taking the erection and adjustment of bridge segment beams nearing their placement as an example, after obtaining the deviations of the splicing end face, support fit, and lifting point force balance influence, the correction actions can be prioritized around the deviations that have the greatest impact on the current installation state. After calculating the deviations in the bearing area and the adjacent area of ​​the lifting point, the influence of the end face closure, the influence of the support fit, and the influence of the lifting point force balance are calculated separately. Then, the deviation decomposition order is determined according to the magnitude of the influence, and the end face correction, support correction, and lifting point correction are obtained. The correction execution sequence is generated in the order of end face first, then support, and then lifting point. Finally, the correction direction is determined according to the positive and negative states of the correction components in each direction, and the correction amplitude is determined according to their absolute values. The correction amount that exceeds the single safe correction range is split into multiple executions, and the execution order, correction direction, and correction amplitude are written into the bridge component installation correction results for on-site personnel to adjust in sequence.

[0022] Furthermore, it also includes: an intelligent method for correcting bridge erection deviations based on point cloud data.

[0023] Working principle: The scheme first continuously records the state of bridge components during the gradual unloading process at the lifting points. Then, relying on edge computing, it identifies the information that truly reflects the installation deviation from these records. Finally, it converts the deviation into directly executable correction results. Specifically, it first synchronously collects multi-view depth images and lifting point load data at various times. It divides the unloading process into multiple stages according to the descent of the lifting point load, and aligns and fuses the multi-view depth images of each stage to reconstruct the point cloud data for that stage. Then, it establishes a unified erection coordinate system based on the support contact edge and the lifting point suspension edge, placing the point clouds of each stage under the same coordinate reference, and extracting the splicing end face point set, support area point set, and lifting point neighbor point set from them. The system identifies three key regions: domain point sets, field point sets, and point pairs. Then, corresponding point pairs between adjacent unloading stages are established on edge computing nodes. Stable migration points and rebound points are distinguished based on displacement direction, continuity, and neighborhood consistency. Stable migration points reflect the actual installation state, while rebound points are considered interference points to be eliminated. Next, stable migration points are registered with corresponding regions in the bridge component design model. After eliminating the influence of rebound areas, the deviations of the splicing end face, support area, and lifting point neighborhood are calculated. Finally, these deviations are decomposed into end face correction, support correction, and lifting point correction, and the bridge component installation correction results are output in the order of end face first, then support, and finally lifting point. For example, in the scenario of segmental beam hoisting and placement, after the beam segment is hoisted to near the design position, it is not completely unloaded at once, but rather placed while being measured. At this time, the system first deploys depth imaging equipment on both sides and above the beam segment, and load sensors at the main lifting points. As the load at the lifting points gradually decreases, edge computing nodes automatically divide the entire unloading process into multiple stages, and reconstruct point clouds from the multi-view depth images captured at each stage. Then, the system unifies these stage point clouds into the same erection coordinate system, focusing on extracting the splicing end face area at the beam segment's end and the support area at the bottom. The system compares the points in the vicinity of the lifting points with those in adjacent stages to determine whether they move continuously in the same direction or rebound in the opposite direction. Points that change steadily indicate a closer approximation of the actual installation deviation, while rebound points indicate a greater influence from short-term force changes and are not suitable as a registration basis. Based on this, the system aligns the actual and valid points with the design model, calculates whether the current beam segment has an opening at the end face, a misalignment of the support, or residual offset near the lifting points, and further provides the correction results on which part should be adjusted first, in which direction, and by how much each adjustment, so that on-site personnel can fine-tune the beam segment position in sequence.

[0024] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent correction of bridge erection deviations based on point cloud data, characterized in that, include: S1. Collect depth images of each unloading stage of the bridge components during the unloading process at the lifting point, and perform three-dimensional reconstruction on the depth images of each unloading stage to generate point cloud data for the corresponding unloading stage. S2. Transform the point cloud data of each unloading stage to the same erection coordinate system, and extract the splicing end face point set, support area point set and lifting point neighborhood point set from the transformed point cloud data of each unloading stage, and output the key area point group. S3. Establish corresponding point pairs according to the point adjacency relationship of the corresponding key area point sets in the adjacent unloading stages, and perform displacement direction comparison on each corresponding point pair. Collect the corresponding point pairs with continuous and consistent displacement directions into a stable migration point set, and collect the corresponding point pairs with reverse displacement directions into a rebound point set. Output the stable migration point set and the rebound point set. S4. Perform point cloud registration between the stable migration point set and the corresponding region in the bridge component design model, and perform culling processing on the registered region based on the rebound point set to output the actual installation deviation set. S5. Perform deviation decomposition on the splicing end face deviation, support area deviation and lifting point neighborhood deviation of the actual installation deviation concentration, generate bridge component installation deviation correction data, and output the bridge component installation correction results based on the installation deviation correction data.

2. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 1, characterized in that, S1 includes: S1-1. Synchronously acquire multi-view depth images and corresponding lifting point load data at each sampling time. Calculate the lifting point load difference between adjacent sampling times in chronological order. When the lifting point load difference is less than zero, record the multi-view depth image of the next sampling time as the new unloading stage image. Output each unloading stage image group.

3. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 2, characterized in that, S1 also includes: S1-2. For the multi-view depth images in each unloading stage image group, extract the support area edge and the stitching end face edge respectively. Then, perform position overlap comparison between the support area edge in any view and the support area edge in other views one by one. Then, perform direction consistency comparison between the corresponding stitching end face edges one by one. Select the view transformation relationship with the largest position overlap and the smallest direction difference to perform alignment on each view depth image and output the fused depth image of each unloading stage. S1-3. Perform three-dimensional back projection on the depth values ​​of each pixel in the fused depth image of each unloading stage according to the imaging parameters of the corresponding viewpoint to generate spatial points corresponding to each pixel. Then, compare the spatial points with the outer contour range of the bridge component point by point and delete the spatial points located outside the outer contour range of the bridge component, and output the point cloud data of the corresponding unloading stage.

4. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 3, characterized in that, S2 includes: S2-1. For the point cloud data of each unloading stage, respectively count the support contact edges and suspension edges in the outer contour of the point cloud, calculate the intersection direction between each support contact edge and the connection direction between each suspension edge, and set the plane where the lowest point of the support contact edge is located as the reference plane, the connection direction as the longitudinal reference direction, and the direction in the reference plane that is perpendicular to the longitudinal reference direction as the transverse reference direction. In this way, construct the stage erection coordinate system corresponding to the point cloud data of each unloading stage, and then transform the point cloud data of each unloading stage to a unified erection coordinate system to output a unified coordinate point cloud group.

5. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 4, characterized in that, S2 also includes: S2-2. For the point cloud data of each unloading stage in the coordinate unified point cloud group, calculate the change in the cross-sectional area of ​​the point cloud segment by segment along the longitudinal reference direction, and record the point cloud cross-section with the area change greater than the area change on the adjacent two sides as the candidate cross-section of the end face. Then, perform plane fitting degree calculation on the points in each candidate cross-section of the end face according to the transverse reference direction and the vertical reference direction. Record the continuous point area with the largest plane fitting degree as the splicing end face point set, and output the splicing end face point set. S2-3. For the point cloud data of each unloading stage in the coordinate unified point cloud group, calculate the distance from each point to the reference plane and record the continuous point area with the smallest distance as the support area point set. Then, take the connection position between the suspension edge of each lifting point and the point cloud as the center, extract the continuous point area within the predetermined range as the lifting point neighborhood point set, and combine the splicing end face point set, the support area point set, and the lifting point neighborhood point set into the key area point group of the corresponding unloading stage.

6. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 5, characterized in that, S3 includes: S3-1. For the splicing end face point set, support area point set and lifting point neighborhood point set in the adjacent unloading stage, construct candidate corresponding score values ​​according to the normal distance from the point to the local tangent plane, the boundary distance from the point to the adjacent boundary line and the point density of the neighborhood where the point is located. Write the point pair with the largest score value into the candidate corresponding pair table. Then perform a two-way reverse lookup consistency check on each point pair in the candidate corresponding pair table. Write the point pairs with consistent reverse lookup results into the initial corresponding point pair set and write the point pairs with inconsistent reverse lookup results into the point pair set to be re-checked. S3-2. For each point pair in the initial set of corresponding point pairs, calculate the component signs of the point pair displacement vector in the longitudinal reference direction, the lateral reference direction, and the vertical reference direction, generate direction symbol codes, and construct direction windows in each key region according to the three adjacent unloading stages. Perform continuity counting, cross-viewpoint coincidence counting, and neighborhood co-directional counting on the direction symbol codes of the same point pair within the direction window to form the migration confidence value of the corresponding point pair, and then write the migration confidence value into the state table of the corresponding point pair.

7. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 6, characterized in that, S3 also includes: S3-3. For each point pair in the corresponding point pair state table, first generate a gating token based on its migration confidence value, and then perform regional consistency check, neighborhood propagation consistency check, and point pair back-check for the point pair with the gating token. When the regional consistency check and the neighborhood propagation consistency check pass, the point pair is written into the stable candidate table. When the regional consistency check fails and the point pair back-check shows that the point pair has a displacement component sign reversal in the later unloading stage, the point pair is written into the bounce candidate table. When the two types of check results conflict, the point pair is written into the state lock table and a rollback re-check is triggered. S3-4. Aggregate each point pair in the stable candidate table according to the neighborhood connectivity relationship. The set of point pairs whose number of aggregated point pairs is more than half of the total number of point pairs in the corresponding key area and whose total migration confidence value is greater than the total migration confidence value in the bounce candidate table in the same area is recorded as the stable migration point set. Aggregate each point pair in the bounce candidate table according to the same aggregation rule. The set of point pairs whose number of aggregated point pairs is more than one-third of the total number of point pairs in the corresponding key area and whose displacement component signs have been reversed at least twice is recorded as the bounce point set. Then, merge the point pairs in the state lock table into the side with the most contact number according to the difference between the number of neighborhood contacts with the stable migration point set and the bounce point set. Output the stable migration point set and the bounce point set.

8. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 7, characterized in that, S4 includes: S4-1. Map the set of stable migration points to the corresponding regions in the bridge component design model according to their respective key regions. Calculate the normal distance from each stable migration point to the fitting surface of the corresponding region and the tangential distance along the boundary direction of the corresponding region. Determine the model corresponding point of each stable migration point based on the criterion of minimizing the sum of the normal distance and the tangential distance, and output the corresponding point group of the region.

9. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 8, characterized in that, S4 also includes: S4-2. For each stable migration point in the corresponding point group of the region and the corresponding point of the model, iteratively solve the region transformation parameters according to the rule of minimizing the sum of squared normal distances and minimizing the change in spacing between adjacent point pairs. Then, delete the model region corresponding to the location of the bounce point set and its adjacent range from the corresponding point group of the region and re-execute the iterative solution to output the registration transformation results. S4-3. Apply the registration transformation results to the stable migration point set, calculate the coordinate differences between each stable migration point and the corresponding point in the model in the longitudinal, lateral, and vertical reference directions, and group the coordinate differences into splicing end face deviation, support area deviation, and lifting point neighborhood deviation according to key areas, and output the actual installation deviation set.

10. The intelligent correction method for bridge erection deviation based on point cloud data according to claim 9, characterized in that, S5 includes: S5-1. Calculate the longitudinal reference direction component, transverse reference direction component, and vertical reference direction component for the splicing end face deviation, support area deviation, and lifting point neighborhood deviation in the actual installation deviation concentration, respectively. Construct a decomposition sequence according to the influence of splicing end face deviation on end face closure, the influence of support area deviation on support fit, and the influence of lifting point neighborhood deviation on lifting point force balance. Merge each direction component into end face correction amount, support correction amount, and lifting point correction amount according to the decomposition sequence, and output installation deviation correction data. S5-2. Generate a correction execution sequence by first eliminating the end face correction amount, support correction amount, and lifting point correction amount in the installation deviation correction data, then restoring the support fit, and finally correcting the lifting point neighborhood offset, and write the correction direction and correction amplitude corresponding to the correction execution sequence into the bridge component installation correction results.

Images