A method for three-dimensional reconstruction and sedimentation morphology analysis of multibeam sonar pipes
By introducing confidence metrics of residual median and interquartile range, as well as loop closure constraints, into sonar detection, the global geometric drift problem of 3D reconstruction results in long-distance pipeline inspection was solved, achieving consistency of elevation benchmarks and accuracy of quantitative analysis of siltation across the entire pipeline length.
Patent Information
- Application Number
- CN202610456093.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-08
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-04-08
AI Technical Summary
In sonar inspection of long-distance municipal drainage pipelines and industrial water transmission pipelines, the error of successive rigid registration between adjacent frames accumulates unidirectionally along the pipeline axis, resulting in global geometric drift in the three-dimensional reconstruction results. This affects the location of severely silted sections and the targeting of dredging solutions.
By generating confidence metrics based on residual median and residual interquartile range during rigid registration of adjacent frames, and searching for non-temporally adjacent but spatially similar frame pairs to establish closure constraints, a confidence-weighted pose graph is constructed, global optimization is performed, and a globally consistent 3D point cloud model is generated.
It effectively eliminates the unidirectional accumulation of frame-by-frame registration errors in long-distance pipeline inspection, ensures the consistency of elevation benchmarks throughout the entire pipeline length, ensures the accuracy of siltation thickness and quantity calculations, and provides a reliable basis for dredging scheduling plans.
Smart Images

Figure CN122335944B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of pipeline non-destructive testing and image processing technology, specifically relating to a method for three-dimensional reconstruction of pipelines and analysis of siltation morphology using multibeam sonar. Background Technology
[0002] Municipal drainage networks and industrial water pipelines are crucial components of urban infrastructure, undertaking key functions such as sewage discharge, rainwater collection, and industrial water delivery. During long-term operation, these pipelines inevitably accumulate sediment, organic deposits, and various debris carried by the water flow. The continuous accumulation of sediment reduces the effective flow cross-section and water delivery capacity, and in severe cases, can lead to blockages and overflows. Therefore, accurate detection and quantitative assessment of sediment accumulation within pipelines are fundamental technical requirements for network maintenance scheduling and dredging decisions.
[0003] Multibeam sonar, capable of non-contact, high-density measurements in turbid water, has become a primary technique for acquiring the three-dimensional morphology of pipeline inner walls and sediment surfaces. In actual inspection operations, the sonar platform advances along the pipeline axis, collecting multiple frames of cross-sectional scan data at regular intervals. The spatial overlap area of the point clouds from adjacent frames is selected, and an iterative nearest-point algorithm is used to solve for the rigid transformation matrix that minimizes the distance between point pairs in the overlapping area. The data from the next frame is then aligned to the coordinate system of the previous frame, and all frames are sequentially stitched together to form a three-dimensional point cloud model of the pipeline interior. This method is widely adopted due to its lack of reliance on additional positioning hardware and low implementation threshold.
[0004] However, in actual pipeline inspection operations, the sonar-mounted platform is situated in a continuously flowing water environment within the pipeline. Due to the combined effects of water turbulence and pipeline geometric constraints, the platform's attitude inevitably undergoes irregular swaying. The rigid registration method for adjacent frames only utilizes the geometric constraints between local frames to solve for pose transformation in each registration step. Registration errors are propagated unidirectionally along the pipeline axis between frames without any global constraints or backtracking correction mechanisms. When the inspection operation advances more than 10 meters along the pipeline, the small registration deviations accumulated frame by frame will superimpose into a centimeter-level or even decimeter-level overall spatial drift, causing a systematic deviation of the geometric shape of the pipeline's far end in the 3D reconstruction results relative to its true location. This deviation makes it impossible for pipeline maintenance personnel to obtain spatially comparable data on siltation thickness and volume between different pipeline sections when developing dredging plans based on the inspection results. This makes it difficult to accurately locate the position and extent of severely silted sections, thus affecting the targeted development of dredging plans and the rational allocation of maintenance resources. Summary of the Invention
[0005] The purpose of this invention is to propose a method for three-dimensional reconstruction and siltation morphology analysis of pipelines using multi-beam sonar, in order to solve the technical problem of global geometric drift in the three-dimensional reconstruction results caused by the unidirectional accumulation of errors in the rigid registration of adjacent frames along the pipeline axis during sonar inspection of long-distance municipal drainage pipelines and industrial water transmission pipelines.
[0006] Therefore, the technical solution provided by this invention for a three-dimensional reconstruction and siltation morphology analysis method for multibeam sonar pipelines is as follows: A method for three-dimensional reconstruction and sedimentation morphology analysis of a multibeam sonar pipeline includes the following steps: S1. Perform polar coordinate to rectangular coordinate transformation on the raw echo data collected frame by frame along the pipeline axis by the sonar mounting platform to generate a cross-sectional point cloud containing multiple reflection points, and attach the echo intensity value of each beam as an attribute label to the corresponding reflection point. S2. Perform rigid registration on the point clouds of two adjacent frames to solve the pose transformation matrix. Based on the residual distribution of the paired point pairs in the overlapping area after registration, extract the residual median and residual interquartile range. Combine the two to form the confidence measure of this registration. S3. Concatenate and multiply the pose transformation matrices calculated frame by frame, project the cross-sectional point clouds of all frames onto the temporary global coordinate system, search for frame pairs that are not adjacent in time but are close in space as candidate frame pairs for loop closure, and confirm them as loop closure constraints after performing registration verification. S4. Construct a pose graph with each frame pose as a node, and adjacent frame registration constraints and loop closure constraints as edges. Convert the confidence measure of each edge into an information matrix as constraint weights. Perform global optimization with the goal of minimizing the weighted sum of squared residuals of all edges. Generate a globally consistent 3D point cloud model based on the cross-sectional point cloud corresponding to the optimized pose projection of each frame. S5. Slice the three-dimensional point cloud model section by section along the pipeline axis, and verify the radial distance discrimination based on the nominal cross-section geometric prior of the pipeline and the echo intensity characteristics. Separate the reflection points of the pipe wall surface and the siltation surface, and calculate the siltation morphology parameters for the separated siltation surface reflection points.
[0007] This invention generates confidence metrics based on the median and interquartile range of residuals during rigid registration of adjacent frames. It also establishes loop constraints by searching for temporally non-adjacent but spatially similar frame pairs in the frame sequence. Furthermore, it replaces the traditional open-loop, frame-by-frame serial stitching with confidence-weighted pose graph global optimization. This effectively solves the problem of global geometric drift in the 3D reconstruction results of the far end of the pipeline caused by the accumulation of frame-by-frame registration errors in long-distance pipeline sonar detection. It maintains spatial consistency of the elevation benchmark throughout the entire pipeline length, thereby ensuring the geometric reference reliability of siltation thickness estimation and siltation volume calculation. This allows pipeline maintenance personnel to obtain spatially comparable quantitative siltation data between different pipeline sections, providing accurate decision-making basis for the targeted formulation of dredging scheduling plans.
[0008] Furthermore, in step S1, a two-dimensional rectangular coordinate system is established with the geometric center of the current frame of the sonar platform as the local origin. The direction angle and echo distance of each beam are converted into reflection point coordinates according to the geometric mapping relationship from polar coordinates to rectangular coordinates. When there is a waterless area above the liquid surface in the pipe section, coordinate transformation is only performed on the beam direction that obtains effective echo.
[0009] This invention establishes a coordinate system with the geometric center of the sonar platform as the local origin and performs coordinate transformation only on the effective echo beams. Under conditions where the pipe cross-section has an incomplete cross-section such as a waterless area above the liquid surface, it can adaptively process the effective echo data without introducing false coordinate points in the direction of invalid beams. This avoids the interference of invalid data mixed in the cross-section point cloud under non-full pipe conditions on the subsequent registration accuracy and ensures the reliability of the 3D reconstruction results in partially filled pipe inspection scenarios.
[0010] Further, in step S2, the rigid registration adopts the iterative nearest point registration algorithm. In each iteration, for each point in the source cross section point cloud, the nearest corresponding point in the target cross section point cloud is searched. Based on the spatial positional relationship of all corresponding point pairs, the rigid transformation parameter that minimizes the sum of the distances between the point pairs is solved. The rigid transformation parameter is applied to the source cross section point cloud and the iteration is repeated until the change in the rigid transformation parameter is less than the preset convergence threshold.
[0011] This invention employs an iterative nearest-point registration algorithm to search for the nearest corresponding point and solve for the optimal rigid transformation parameters in each iteration. This enables the spatial alignment process between adjacent frames to have a self-correcting ability that gradually converges. Compared with the single-transformation estimation method, it can make fuller use of the geometric information of the overlapping area. Even in actual detection scenarios where there are local protrusions, depressions, or siltation on the inner wall of the pipe, resulting in irregular cross-sectional contours, it can still obtain high-precision inter-frame pose transformation results.
[0012] Furthermore, when the registration process of two adjacent frame cross-sectional point clouds still fails to meet the convergence threshold after reaching the preset maximum number of iterations, the registration is marked as non-converged, and the confidence metric of the registration is forcibly set to the preset minimum confidence level. In the global optimization of step S4, the pose of the frame is determined by the constraints in the pose graph that are adjacent to the frame and whose confidence metric is not the minimum confidence level, in order to suppress the constraint weights with the minimum confidence level.
[0013] Further, in step S3, the frame pair is set to represent as To mark a frame pair as a loopback candidate, it must simultaneously meet the following two conditions: frame With frames The number of frames separated in the time sequence of the frame sequence exceeds the preset minimum frame interval threshold; With frames The spatial distance in the temporary global coordinate system is less than a preset ratio of the nominal diameter of the pipe.
[0014] This invention identifies candidate frame pairs for loop closure by simultaneously setting two screening conditions: a frame spacing threshold and a spatial distance threshold. The frame spacing threshold excludes neighboring frames that have already been processed as adjacent frames to avoid redundant constraints, while the spatial distance threshold ensures that loop closure constraints are established only for frame pairs that are actually close in space. This ensures that the loop closure detection process neither misses effective loop closure opportunities nor introduces false loop closures that are too far apart in space, thus guaranteeing the effective establishment of loop closure constraints in actual detection sections where paths overlap, such as pipe bends and junctions.
[0015] Furthermore, after performing rigid registration on the candidate frame pairs of loop closures, the median of the residuals and the interquartile range of the residuals for that registration are extracted respectively. Only when the median of the residuals for that registration is less than a preset median of residuals threshold and the interquartile range of the residuals is less than a preset interquartile range threshold, is the loop closure constraint confirmed and included in the pose graph of step S4 to participate in global optimization.
[0016] Furthermore, in step S4, the information matrix is constructed as follows: the median of the residual and the reciprocal of the interquartile range of the residual are used as the basic values of the diagonal elements of the information matrix. When the median of the residual or the interquartile range of the residual increases, the corresponding information matrix value decreases, and the constraint weight of the edge in the global optimization decreases accordingly.
[0017] This invention constructs the diagonal elements of the information matrix using the median and the reciprocal of the interquartile range of the residuals, thereby achieving automatic mapping of confidence metrics to constraint weights. This allows high-registration-quality inter-frame constraints to bear a greater weight in pose localization during global optimization, while the constraint weights of low-registration-quality sections are automatically weakened. In actual working conditions where the degree of water flow turbulence and the richness of cross-sectional geometric features vary in different sections of the pipeline, global optimization can adaptively rely on reliable data segments and tolerate low-quality data segments, thus improving the overall positional accuracy of 3D reconstruction over the entire length.
[0018] Further, in step S5, separating the reflection points from the pipe wall surface and the siltation surface includes two stages; the first stage includes: performing preliminary classification based on the comparison result of the radial distance deviation of each reflection point from the nominal pipe wall profile and the preset pipe wall tolerance threshold, separating a preliminary subset of pipe wall reflection points and a preliminary subset of siltation surface reflection points; the second stage includes: for reflection points whose radial distance deviation is in the fuzzy area near the pipe wall tolerance threshold, calculating the difference between the echo intensity value of the reflection point and the median echo intensity of the preliminary subset of pipe wall reflection points and the preliminary subset of siltation surface reflection points, and classifying the reflection point into the category with the smaller difference value.
[0019] Furthermore, the frame sampling interval of the sonar mounting platform is one frame acquired every 10 mm to 50 mm of advance; the operating frequency range of the multi-beam sonar transducer array is 600 kHz to 2000 kHz; and the preset tube wall tolerance threshold range for radial distance determination in step S5 is 8 mm to 30 mm.
[0020] Furthermore, the sediment morphology parameters include: The siltation accumulation height is defined as the maximum radial accumulation distance from the lowest point of the pipe bottom at the siltation surface reflection point at the corresponding cross-section location. The siltation cross-sectional area is defined as the area of the closed region enclosed by the reflection point of the siltation surface at the corresponding cross-sectional location. The effective flow area ratio is defined as the ratio of the effective flow area obtained by subtracting the siltation cross-sectional area from the nominal cross-sectional area of the pipe to the nominal cross-sectional area of the pipe.
[0021] The beneficial effects of this invention are as follows: This invention can effectively eliminate the global geometric drift caused by the unidirectional accumulation of frame-by-frame registration errors in long-distance pipeline inspection, ensure the spatial consistency of the elevation benchmark within the entire length of the pipeline, and improve the global positional accuracy of the three-dimensional reconstruction results inside the pipeline. Attached Figure Description
[0022] Figure 1 This is a flowchart illustrating the method for three-dimensional reconstruction and siltation morphology analysis of the multibeam sonar pipeline in this embodiment. Figure 2 This is a schematic diagram of a non-contact detection operation scenario where the sonar platform in this embodiment performs non-contact detection inside a pipe with turbulent water flow. Figure 3 This is a schematic diagram comparing the separation effect of the present invention and conventional methods on the reflection points of the pipe wall surface and the reflection points of the siltation surface. Detailed Implementation
[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0024] In this embodiment, the multi-beam sonar pipeline 3D reconstruction and siltation morphology analysis method operates on a sonar inspection system mounted on a pipeline crawling robot platform. The pipeline crawling robot platform consists of a chassis walking module and a sonar mounting cabin. The chassis walking module is driven axially along the bottom of the pipeline by rubber tracks. The propulsion speed is set to 50 mm / s in this embodiment, but in practical applications, it can be adjusted within the range of 20 mm / s to 200 mm / s depending on the water flow velocity and detection accuracy requirements within the pipeline. The sonar mounting cabin is fixedly installed above the chassis walking module, and a multi-beam sonar transducer array is installed inside the cabin.
[0025] In this embodiment, the multibeam sonar transducer array comprises 256 elements arranged in an arc-shaped, equally spaced pattern. The effective sector angle of the array is 360°, covering the entire circumferential range of the pipe cross-section. In application scenarios where the number of elements is insufficient to cover the entire circumference, the number of elements is configured in the range of 128 to 512, and the corresponding sector angle can be adjusted in the range of 180° to 360°. The operating frequency of the multibeam sonar transducer array is set to 1200 kHz in this embodiment, a frequency selection that balances underwater acoustic attenuation and angular resolution. In scenarios with large pipe diameters and high water turbidity, the operating frequency should preferably be selected in the range of 600 kHz to 900 kHz to ensure detection range; in scenarios with small pipe diameters and high requirements for detail resolution, the operating frequency can be increased to 1500 kHz to 2000 kHz. The angular resolution per frame scan in this embodiment is 1.4°, corresponding to the uniform distribution of the 256 beams within a 360° range.
[0026] At the data acquisition and processing end, the sonar transducer array communicates with a main control unit based on a hybrid architecture of a field-programmable gate array (FPGA) and a digital signal processor (DSP) via a high-speed differential signal interface. In the main control unit, the FPGA performs real-time beamforming calculations for the sonar echo signals, while the DSP performs subsequent data processing tasks such as point cloud generation, registration calculations, and optimization. The DSP operates at a clock frequency of 600 MHz and supports double-precision floating-point operations. The main control unit also communicates with a large-capacity non-volatile memory module via a serial peripheral interface to store the raw echo data and processed point cloud data for each scan frame. In this embodiment, the non-volatile memory module has a storage capacity of 64 gigabytes, which can be expanded to 256 gigabytes in applications where the detection pipeline length exceeds 500 meters.
[0027] At the communication level, the main control unit establishes a wired data link with the ground-based upper-level monitoring computer via armored communication cables laid inside the pipeline. The communication cables employ a shielded twisted-pair structure, using the industrial Ethernet protocol, with a transmission bandwidth of 100 megabits per second in this embodiment, capable of supporting real-time transmission of sonar scan data. When the pipeline length exceeds the effective transmission distance of the communication cable, an offline acquisition mode can be adopted, where the robot platform autonomously stores the data and transmits it in batches after detection.
[0028] Specifically, when an individual element in the sonar transducer array malfunctions, resulting in no echo data being returned in the corresponding beam direction, the main control unit executes the following protection logic: The beam direction corresponding to the faulty element is marked as an invalid beam; the coordinate calculation for this beam direction is skipped during subsequent cross-sectional point cloud generation; and arc-shaped interpolation is performed between adjacent valid beams on both sides of this beam direction. During the execution of the protection logic, the system's cross-sectional scanning function remains normal, with only a slight reduction in local angular resolution in the faulty beam direction. When the number of faulty elements exceeds 10% of the total number of array elements, the system sends an equipment maintenance alarm to the upper-level monitoring computer, prompting operators to perform maintenance on the transducer array after the current inspection is completed.
[0029] Example 1 like Figure 1 As shown, the method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipes in this embodiment includes the following steps: S1. Perform polar coordinate to rectangular coordinate transformation on the raw echo data collected frame by frame along the pipeline axis by the sonar mounting platform to generate a cross-sectional point cloud containing multiple reflection points, and attach the echo intensity value of each beam as an attribute label to the corresponding reflection point.
[0030] like Figure 2 As shown, this illustrates a real-world scenario where a sonar-mounted platform performs non-contact inspections inside a pipe with turbulent water flow. The figure labels the sonar platform, the pipe wall, the sediment deposits at the bottom, and the spatial coverage pattern of the multi-beam sonar emitted downwards by the platform. As the pipe-crawling robot platform advances along the pipe's axis, the multi-beam sonar transducer array continuously collects cross-sectional echo data at a preset frame sampling interval. In this embodiment, the frame sampling interval is set to one frame every 20 millimeters of advancement, meaning a complete cross-sectional scan is triggered every 20 millimeters the sonar platform moves forward. This frame sampling interval balances axial point cloud density and data storage capacity. In fine-scale inspection scenarios requiring high axial resolution, the frame sampling interval can be shortened to 10 millimeters; in large-scale rapid survey scenarios, the frame sampling interval can be increased to 50 millimeters. The raw measurement data generated by each frame scan includes 256 pairs of beam azimuth angles and corresponding echo distance values, as well as the echo signal strength value for each beam. The original measurement data uses the geometric center of the current frame of the sonar platform as the local origin, and the beam azimuth angles increase clockwise with the vertical upward as the zero-degree reference.
[0031] After data acquisition, this step first performs a polar coordinate to Cartesian coordinate transformation on each frame of raw measurement data. A two-dimensional Cartesian coordinate system is established with the geometric center of the current frame on the sonar platform as the local origin, where the horizontal direction is the first coordinate axis and the vertical direction is the second coordinate axis. For the first... A beam, its direction angle and echo distance The coordinates of the reflection points are converted to local rectangular coordinates. This conversion follows the standard geometric mapping relationship between polar coordinates and rectangular coordinates, and belongs to the basic coordinate transformation method known in the art. After this conversion, the 256 sets of polar coordinate measurements from each frame of scanning are converted into 256 two-dimensional reflection point coordinates with well-defined spatial locations, in millimeters.
[0032] Furthermore, the echo intensity value is extracted from the echo signal of each beam, and this intensity value is attached as an acoustic attribute label to the coordinate data of the corresponding reflection point. The physical meaning of the echo intensity value is the signal amplitude of the sound wave returning to the transducer array after reflection from the pipe wall or siltation surface, expressed in decibels. Due to differences in acoustic impedance, the echo intensity varies statistically among different reflective surfaces: the pipe wall material is usually concrete, metal, or plastic, which has a higher acoustic impedance and generally higher echo intensity; the siltation material is usually mud, sand, organic sediment, or a mixture of debris, which has a relatively lower acoustic impedance and generally lower echo intensity. The echo intensity information is not involved in the coordinate calculation in this step, but is stored as an additional attribute along with the coordinate data, to be used as an auxiliary constraint for inter-frame feature matching and a joint criterion for separating the pipe wall surface and the siltation surface in subsequent steps.
[0033] After the above processing, each frame of data outputs a cross-sectional point cloud containing 256 reflection points. Each reflection point carries two types of information: two-dimensional spatial coordinates and echo intensity values. Through this structured reorganization step, the original echo data is transformed from polar coordinate measurements with the transducer array as a reference into a cross-sectional point cloud with a clear spatial location, centered on the geometric center of the mounting platform. At the same time, acoustic intensity information is preserved, providing input data with both geometric and acoustic features for inter-frame registration and surface separation in subsequent steps.
[0034] It should be noted that in certain application scenarios, when the pipe cross-section is not a complete closed ring but contains a waterless area above the liquid surface, the beam direction above the liquid surface will not be able to obtain a valid pipe wall echo signal. Under such conditions, the coordinate transformation in this step is only performed on the beam direction that obtains a valid echo; the reflection points corresponding to the invalid beam direction are not included in the cross-sectional point cloud. This processing method does not affect the registration and analysis operations based on the subset of valid reflection points in subsequent steps.
[0035] S2. Perform rigid registration on the point clouds of two adjacent frames to solve the pose transformation matrix. Based on the residual distribution of the paired point pairs in the overlapping area after registration, extract the residual median and residual interquartile range. Combine the two to form the confidence measure of this registration.
[0036] As the sonar platform propels itself through the water inside the pipe, it is affected by various factors, including water turbulence, non-uniform flow field at pipe bends, and changes in friction between the track and pipe wall. Consequently, the platform's spatial pose changes irregularly between adjacent acquisition frames. If the point clouds of adjacent frames are directly superimposed in their respective local coordinate systems, the geometry between the two frames cannot be correctly aligned. Therefore, spatial registration of the point clouds of adjacent frames is required to solve the pose transformation relationship that aligns the point cloud of the later frame to the coordinate system of the previous frame.
[0037] For the first Frame section point cloud and the first First, the spatial overlap region between two frames is determined for the frame cross-sectional point cloud. In the pipeline sonar detection scenario, since the frame sampling interval is much smaller than the detection range of the sonar scan, there is a large proportion of spatial overlap between the cross-sectional point clouds of adjacent frames along the pipeline axis. In this embodiment, the frame sampling interval is 20 mm, while the axial detection range of the sonar covers a pipeline cross-sectional area of approximately 50 mm to 300 mm in front of the sonar. Therefore, the overlap rate between adjacent frames is not less than 80%. Within the overlap region, an iterative nearest-point registration algorithm is performed on the two frame point clouds. The iterative nearest-point registration algorithm is a well-known technique in the art for rigid point cloud alignment. Its core process is as follows: in each iteration, for each point in the source cross-sectional point cloud, the nearest corresponding point in the target cross-sectional point cloud is searched. Based on the spatial positional relationship of all corresponding point pairs, the rigid transformation parameters that minimize the sum of the distances between point pairs are solved. Then, this transformation is applied to the source cross-sectional point cloud, and the above process is repeated until the change in the transformation parameters is less than a preset convergence threshold. In this embodiment, the convergence threshold is set to a change of less than 0.01 mm in the translation component and less than 0.01° in the rotation component of the transformation parameters, and the maximum number of iterations is set to 50. In scenarios with low overlap or high point cloud noise, the maximum number of iterations can be increased to 100 to improve convergence stability.
[0038] After registration convergence, the residual distribution of all paired point pairs within the overlapping region is calculated. The residual is defined as the spatial distance between each paired point pair after the registration transformation is applied, in millimeters. Two statistics are extracted from the residual distribution: the first statistic is the median of the paired point pair residuals, which physically represents the overall deviation level of the registration alignment, reflecting the accuracy of the registration result; the second statistic is the interquartile range of the residuals, which physically represents the spatial uniformity of the registration alignment, reflecting the certainty of the registration result. The median, rather than the mean, is used as the first statistic because the median is insensitive to a few large residual anomalies at the edges of the overlapping region caused by inconsistencies in point cloud coverage, and will not be significantly skewed by the presence of individual outliers.
[0039] The first and second statistics mentioned above are combined to form the confidence measure for this registration. When the overlapping region has rich geometric features, i.e., when there are obvious local morphological changes on the pipe wall surface or when the accumulation of sediment causes the cross-sectional profile to exhibit asymmetrical features, the iterative nearest-point registration algorithm can obtain sufficient geometric constraints, and the median and interquartile range of the residuals of the registration results are both small, corresponding to a high confidence level. When the geometric features of the overlapping region degenerate, i.e., when the pipe cross-section is close to a standard circle and there are no obvious sedimentation or local deformation features, the geometric recognizability of the cross-sectional profile in the rotation direction decreases, and the convergence direction of the iterative registration process becomes ambiguous. Even if the median of the residuals is still acceptable on the surface due to the symmetry of the cross-section, the interquartile range will increase significantly to reflect the uncertainty of the registration result in the spatial direction. Therefore, the confidence measure reflects both the accuracy and certainty of the registration result; any anomaly in either dimension will reduce the confidence level.
[0040] In this step, the iterative nearest-point registration algorithm is merely one specific method for achieving spatial alignment of point clouds in adjacent frames. Those skilled in the art will understand that any registration algorithm capable of determining the optimal rigid transformation parameters for alignment of overlapping regions from two sets of spatially overlapping point cloud data can be applied to this step without departing from the technical concept of this invention.
[0041] After the above processing, each pair of adjacent frames outputs a local pose transformation matrix with a confidence metric label. The pose transformation matrix is a homogeneous transformation matrix with three rows and three columns, containing two-dimensional translation and rotation components. Through registration and confidence evaluation in this step, not only is a spatial alignment relationship established between adjacent frames, but the reliability of each alignment result is also quantified, enabling subsequent global optimization steps to distinguish between high-reliability and low-reliability registrations and assign differentiated constraint weights accordingly.
[0042] Specifically, when the registration process of a pair of adjacent frames fails to meet the convergence threshold after reaching the maximum number of iterations, the system marks the registration as non-converged, and its confidence metric is forcibly set to the lowest level. In subsequent global optimization steps, the impact of this low-confidence constraint on the global pose will be automatically suppressed, and the pose of the frame will be determined by the constraints of the high-confidence neighboring frames on both sides of the frame.
[0043] S3. Concatenate and multiply the pose transformation matrices calculated frame by frame, project the cross-sectional point clouds of all frames onto a temporary global coordinate system, search for frame pairs that are not temporally adjacent but spatially close as loop closure candidate frame pairs, and confirm them as loop closure constraints after performing registration verification.
[0044] After step S2 is completed, the system obtains the local pose transformation matrices and their confidence metrics between all adjacent frame pairs. If only these adjacent frame constraints are used for serial stitching, the registration error will accumulate unidirectionally along the frame sequence without any correction mechanism. In order to introduce backtracking correction capability into the flow topology of registration information, frame pairs that are not adjacent in time but are close in spatial location are found in the frame sequence, and the geometric consistency between these frame pairs is used to establish lapsing constraints spanning multiple frames.
[0045] First, using the local pose transformation matrix obtained from frame-by-frame registration in step S2, the cross-sectional point clouds of all frames are projected into a temporary global coordinate system by sequentially multiplying the matrices of all preceding frames in chronological order. This temporary global coordinate system is based on the local coordinate system of the first frame, and the global pose of each subsequent frame is obtained by sequentially multiplying the local transformation matrices of all its preceding frames. The pose estimation in this temporary global coordinate system, because it includes the accumulated registration error from frame-by-frame, is not the final output result; it is only used for determining the spatial proximity of loop closure candidate frame pairs in this step.
[0046] Subsequently, a spatial neighborhood search is performed on each frame of the point cloud within the temporary global coordinate system. For the first... Frame, iterate through all other frames, when a certain frame Frame pairs will be executed when both of the following conditions are met. Marking as a loopback candidate targets: the first condition is a frame. With frames If the number of frames separated by a frame sequence in time exceeds a preset minimum frame interval threshold (in this embodiment, this threshold is set to 50 frames), the purpose of setting this threshold is to exclude neighboring frames that have already been processed as adjacent frames in step S2, thus avoiding duplicate constraint establishment; the second condition is frames. With frames The spatial distance in the temporary global coordinate system is less than the nominal diameter of the pipe. In this embodiment, the nominal diameter of the pipe being detected is 1000 mm, therefore the spatial distance threshold is 500 mm. When the nominal diameter of the pipe varies between 400 mm and 3000 mm, the spatial distance threshold should be adjusted accordingly to the nominal diameter of the pipe. to In scenarios where the frame sampling interval differs from that in this embodiment, the minimum frame spacing threshold can be adjusted within the range of 30 to 100 frames. The lower bound of its value should ensure that the cumulative advance distance between two frames along the pipe axis exceeds the axial detection range of a single sonar frame scan.
[0047] For each marked loop closure candidate frame pair, the same iterative nearest-point registration process as in step S2 is performed, and the registration confidence metric is calculated using the same method. A loop closure constraint is only confirmed and included in subsequent global optimization if the first statistic in the confidence metric of the loop closure registration is less than a preset residual median threshold and the second statistic is less than a preset interquartile range threshold. In this embodiment, the residual median threshold is set to 2 mm, and the interquartile range threshold is set to 3 mm. The selection principle for these thresholds is that the quality threshold of the loop closure constraint should be higher than that of adjacent frame registration, because the role of the loop closure constraint is to provide a high-confidence long-distance verification condition for global optimization; including low-quality loop closure constraints may introduce incorrect global constraints.
[0048] In straight pipe sections, because the sonar platform advances unidirectionally along the pipe axis and will not return to a spatially proximate detected area, loopback candidate frame pairs satisfying the above two conditions are usually few or even non-existent. However, when there are bends, junctions, branch nodes, or overlapping detection paths due to operational needs in the pipe, the sonar platform passes through spatially proximate pipe areas at different times, and loopback candidate frame pairs will naturally appear. Each confirmed loopback constraint is equivalent to a spatial consistency check condition independent of the frame-by-frame serial transmission chain, providing an anchor point for subsequent global optimization to pull back the pose error of distant frames.
[0049] Furthermore, in another implementation, instead of solely using spatial distance as the criterion for loop closure candidate selection, the echo intensity features retained in step S1 can be used to construct a cross-sectional feature description vector, and similar frame pairs can be searched in the feature space as a supplementary source of loop closure candidates. This variant approach has the advantage of discovering potential loops in long straight pipe sections with highly repetitive pipe geometries.
[0050] After this step, the system obtains several additional cross-frame loop constraints in addition to the adjacent frame registration constraints. Each loop constraint carries a corresponding pose transformation matrix and confidence metric. The introduction of these loop constraints transforms the flow topology of inter-frame information from the original unidirectional serial chain structure to a graph structure with cross-connections.
[0051] S4. Construct a pose graph using the pose of each frame as nodes and the registration constraints and loop closure constraints of adjacent frames as edges. Convert the confidence measure of each edge into an information matrix as constraint weights. Perform global optimization with the goal of minimizing the weighted sum of squared residuals of all edges. Generate a globally consistent 3D point cloud model based on the cross-sectional point cloud corresponding to the optimized pose projection of each frame.
[0052] In order to ensure that the siltation morphology analysis in the subsequent step S5 can be based on a three-dimensional point cloud model with accurate and consistent spatial location, this step integrates all adjacent frame registration constraints output in step S2 and all loop closure constraints output in step S3 into a unified pose graph structure, and performs global consistency optimization on this pose graph.
[0053] In this embodiment, the pose graph is constructed as follows: the global pose of each frame in the frame sequence is taken as a node in the pose graph. The state variables of the node are the two-dimensional translation coordinates and rotation angle of the frame in the global coordinate system, for a total of three degrees of freedom. The registration result of each pair of adjacent frames in step S2 is taken as an edge connecting the corresponding two nodes. The information carried by each edge includes: the local pose transformation matrix obtained from the registration solution and the corresponding confidence metric. Each confirmed loop closure constraint in step S3 is also taken as an edge connecting the corresponding two non-adjacent nodes. The information structure of the edge is the same as that of the adjacent frame edge. In this embodiment, if the detection pipeline length is 100 meters and the frame sampling interval is 20 millimeters, the frame sequence contains approximately 5000 frames, the number of nodes in the pose graph is 5000, the number of adjacent frame edges is 4999, and the number of loop closure edges depends on the pipeline geometry and the overlap of the detection path.
[0054] When performing global optimization on the pose graph, the confidence metric of each edge is converted into the corresponding edge's information matrix. The physical meaning of the information matrix is the constraint strength of that edge on the node pose estimation: edges with higher confidence correspond to larger information matrix values, exerting a stronger pull on the pose of the connected nodes during optimization; edges with lower confidence correspond to smaller information matrix values, and their pull on the node pose is automatically weakened. Specifically, the conversion method is as follows: the median of the residuals and the reciprocal of the interquartile range are used as the base values for the diagonal elements of the information matrix. When the median of the residuals or the interquartile range increases, the corresponding information matrix value decreases, and the constraint weight of that edge decreases accordingly.
[0055] In this embodiment, the objective function of global optimization is the weighted sum of squared residuals of all edge constraints. The weighted residual is defined as follows: for each edge, take the current pose estimates of the two nodes connected by the edge, calculate the theoretically expected pose relationship between the two nodes according to the pose transformation matrix carried by the edge, and use the deviation between the theoretical relationship and the current estimate as the residual of the edge, multiplying it by the information matrix of the edge to obtain the weighted residual. The goal of global optimization is to find the pose values of all nodes such that the weighted sum of squared residuals of all edges reaches its minimum. This optimization problem belongs to nonlinear least squares problems, and in this embodiment, the Levenberg-Marquardt iterative algorithm is used to solve it. The Levenberg-Marquardt algorithm is a well-known algorithm in this field for handling nonlinear least squares optimization problems. In this embodiment, the initial value of its damping factor is set to 0.01, and the iteration termination condition is set to the relative change in the objective function value being less than 0.01. Or the number of iterations reaches 200.
[0056] During the global optimization process, the introduction of loop closure constraints has the following key effects: At the loop closure point, the pose deviation caused by the cumulative registration error frame by frame is forced to zero by the loop closure constraint. The correction amount generated by this zeroing is evenly distributed along the frame sequence covered by the loop towards both ends, rather than being concentrated on a single frame. This means that the large drift that was originally concentrated at the end of the frame sequence is dispersed into small residuals distributed along the entire length of the pipe, and the spatial positioning deviation at the far end of the pipe changes from cumulative drift to bounded residuals. Even in pipe segments without loop closure constraints, the confidence weighting mechanism ensures that low-reliability registration results do not dominate the pose estimation of that segment: high-confidence inter-frame constraints undertake the main spatial positioning function, and the pose of low-confidence segments is determined by the high-confidence segments at both ends, similar to the suspended segment between two high-strength support points in a suspension bridge, whose spatial position is determined by the constraint interpolation at both ends rather than by its own unreliable local measurements.
[0057] It should be noted that when there are no loop closure constraints in the pose graph, global optimization still adjusts the pose of each frame globally through a confidence-weighted mechanism. In this case, the effect of global optimization is mainly reflected in: suppressing the propagation of registration errors from low-confidence segments to the global pose, making the pose estimation of high-confidence segments more robust, but it cannot completely eliminate systematic drift over long distances. When there is at least one loop closure constraint in the pipeline detection path, the drift suppression effect will be significantly enhanced.
[0058] After global optimization, all frames are assigned spatial poses corrected for global consistency. The point clouds of each frame section are projected onto a unified global coordinate system according to the corrected poses, forming a complete 3D point cloud model of the pipeline's interior. Each point in the complete 3D point cloud model carries 3D spatial coordinates and the echo intensity value retained in step S1. Through this global optimization step, the spatial positional consistency of the pipeline along its entire length in the 3D point cloud model is significantly improved, and the stability of the elevation datum is ensured, thus providing a reliable geometric reference for the siltation morphology analysis in step S5.
[0059] S5. Slice the three-dimensional point cloud model section by section along the pipeline axis, and verify the radial distance discrimination based on the nominal cross-section geometric prior of the pipeline and the echo intensity characteristics. Separate the reflection points of the pipe wall surface and the siltation surface, and calculate the siltation morphology parameters for the separated siltation surface reflection points.
[0060] The 3D point cloud model output in step S4 contains reflection points from two types of geometric entities: the inner wall surface of the pipe and the surface of the sediment. Without separating these two types of surfaces, it is impossible to independently obtain information on the spatial distribution and accumulation thickness of the sediment. Therefore, it is necessary to distinguish the reflection points from the pipe wall surface and the sediment surface in a globally consistent 3D point cloud.
[0061] In this embodiment, the separation operation utilizes the geometric prior constraints of the pipe cross-section. The geometric prior refers to the nominal cross-sectional geometric parameters and contour information given in the pipe design. In this embodiment, the nominal cross-section is a circular cross-section with a nominal inner diameter of 1000 mm. In application scenarios where the pipe has a rectangular cross-section, an arched cross-section, or other non-circular cross-sections, the nominal cross-sectional shape should be replaced with the corresponding design geometric parameters. The globally consistent 3D point cloud model is sliced at equal intervals along the pipe axial direction. The slice interval is set to 10 mm in this embodiment, but can be relaxed to 50 mm in scenarios with lower axial resolution requirements. At each slice location, all reflection points within a 5 mm range before and after that location are extracted to form a subset of the cross-sectional point cloud at that location.
[0062] For each subset of cross-sectional point clouds, joint discrimination and separation of the pipe wall surface and the siltation surface are performed. The separation strategy consists of two stages. The first stage is preliminary assignment based on spatial distance. Specifically, using the center and radius of the nominal pipe cross-section as references, the radial distance deviation from each reflection point in the cross-sectional point cloud subset to the nominal pipe wall profile is calculated. When the radial distance deviation from a reflection point to the nominal pipe wall profile is less than a preset pipe wall tolerance threshold, the reflection point is initially assigned as a candidate pipe wall reflection point; when the radial distance deviation exceeds the pipe wall tolerance threshold and the deviation direction points inward into the pipe, the reflection point is initially assigned as a candidate siltation surface reflection point. In this embodiment, the pipe wall tolerance threshold is set to 15 mm, which covers the pipe construction error and the amount of pipe wall deformation during the service life. In older pipes with longer service life and higher deformation risk, this threshold can be relaxed to 30 mm; in newly built pipes, it can be narrowed to 8 mm.
[0063] like Figure 3 As shown in the enlarged view, the section reveals a localized area of blurred boundaries where pipe wall deformation occurs and sediment adheres tightly to the pipe wall. Rigid registration between adjacent frames and conventional separation strategies exhibit typical mis-separation phenomena in this area; however, the method of this invention, relying on a joint criterion of distance and intensity, accurately cuts and distinguishes the pipe wall surface from the sediment surface, demonstrating higher classification consistency and anti-ambiguity judgment capability.
[0064] The second stage involves joint verification based on echo intensity characteristics. In the preliminary attribution results from the first stage, for reflection points in ambiguous regions where the radial distance deviation is near the pipe wall tolerance threshold, the echo intensity information retained in step S1 is further introduced for joint discrimination. Specifically, the echo intensity distribution characteristics of the subsets clearly assigned as pipe wall reflection points and the subsets clearly assigned as siltation surface reflection points are statistically analyzed, and the median echo intensity of the two subsets is extracted as their respective intensity benchmarks. For reflection points in ambiguous regions, a first difference value is calculated between the echo intensity value of the reflection point and the median echo intensity of the preliminary pipe wall reflection point subset, and a second difference value is calculated between the echo intensity value of the reflection point and the median echo intensity of the preliminary siltation surface reflection point subset. If the first difference value is less than the second difference value, the reflection point is assigned as a pipe wall reflection point; otherwise, it is assigned as a siltation surface reflection point.
[0065] Using both spatial distance and echo intensity features as the separation criterion, compared to relying solely on a single distance threshold, maintains higher separation consistency in areas of localized pipe wall deformation and in ambiguous regions where sediment adheres closely to the pipe wall. In areas of localized pipe wall deformation, although the radial distance deviation of the pipe wall reflection point may exceed the tolerance threshold and be initially classified as a sedimentation surface candidate, its echo intensity still retains the high-intensity characteristics of the pipe wall material. This can be correctly corrected to a pipe wall reflection point through the second-stage joint verification. Similarly, in areas where sediment adheres closely to the pipe wall, although the radial distance deviation of the sedimentation surface reflection point may be less than the tolerance threshold and be initially classified as a pipe wall candidate, its echo intensity exhibits the low-intensity characteristics of the sediment material. This can also be correctly corrected through joint verification.
[0066] After separating the pipe wall from the siltation surface, three siltation morphology parameters are calculated cross-sectionally along the pipe axis for the subset of siltation surface reflection points. The first parameter is the siltation accumulation height, defined as the maximum radial accumulation distance from the lowest point of the pipe bottom among the siltation surface reflection points at that cross-section location, in millimeters. The second parameter is the siltation cross-sectional area, defined as the area of the closed region enclosed by the siltation surface reflection points at that cross-section location, in square millimeters. The third parameter is the effective flow area ratio, defined as the ratio of the effective flow area (obtained by subtracting the siltation cross-sectional area from the nominal cross-sectional area of the pipe) to the nominal cross-sectional area, expressed as a percentage, used to visually reflect the degree of water conveyance capacity reduction of the pipe at that cross-section location.
[0067] By arranging the three siltation morphology parameters mentioned above at all cross-sectional locations along the pipeline's axis in axial order, the distribution curves of siltation accumulation height, siltation cross-sectional area, and effective flow cross-sectional area ratio along the entire pipeline length can be generated. These three distribution curves together constitute the pipeline's siltation morphology distribution map, providing a quantitative basis for downstream dredging scheduling decisions based on the continuous spatial distribution along the pipeline's axis. Pipeline management personnel can use this to determine the precise location and extent of severely silted sections and formulate targeted dredging operation route plans and resource allocation schemes.
[0068] Specifically, when the number of effective reflection points in the sub-set of cross-sectional points at a certain cross-sectional location is lower than a preset minimum number of points threshold (in this embodiment, this threshold is set to 20 reflection points), the system considers the data coverage at that cross-sectional location insufficient to support reliable separation and discrimination between the pipe wall surface and the sedimentation surface, and marks the three sedimentation morphology parameters at that cross-sectional location as missing values. When generating the axial distribution curve, the cross-sectional location corresponding to the missing value is filled by linear interpolation of the effective cross-sections on both sides, and is marked in the distribution map with a labeling method different from the normal values, indicating to the pipeline network management personnel that the sedimentation parameters of this section are interpolated estimates rather than direct measurements.
[0069] This invention, through the registration confidence metric introduced in step S2 and the non-adjacent frame loop closure constraint detection mechanism introduced in step S3, enables the system to transform the pose drift, which was originally transmitted unidirectionally along the pipeline axis and continuously increased, into a bounded residual that is forcibly amortized at the loop closure point when facing the situation of continuous accumulation of frame-by-frame registration errors in long-distance pipeline sonar detection operations. The underlying mechanism is that the loop closure constraint provides an independent spatial consistency verification condition for the pose map across multiple frames. When the loop closure constraint detects that the accumulated drift causes inconsistency between two observations at the same spatial location, the global optimization process evenly distributes this inconsistency along the frame sequence covered by the loop, thereby transforming the large deviation concentrated at the far end of the pipeline into a small correction amount distributed over the entire length.
[0070] Regarding the consistency of elevation datum, since the spatial poses of each cross-section in the globally optimized 3D point cloud model have reached overall optimality under global constraints, the elevation reference system between different sections of the pipeline maintains spatial coherence. This means that in the siltation morphology analysis in step S5, the siltation accumulation heights calculated at different cross-sections are all based on a consistent geometric datum, and the siltation height data at the beginning and end of the pipeline are directly comparable. In contrast, if the traditional open-loop frame-by-frame serial stitching method is used, the elevation datum at the far end of the pipeline will systematically deviate from the true position due to cumulative drift, resulting in the estimated value of the siltation height at the far end containing the systematic error component caused by drift, and the siltation data at the far end and near end losing spatial comparability.
[0071] In the separation process between the pipe wall and the siltation surface, the joint criterion of spatial distance features and echo intensity features used in step S5 ensures that the separation accuracy does not entirely depend on a single radial distance threshold. In actual pipeline operation, local deformation of the pipe wall due to factors such as soil settlement, groundwater erosion, or changes in bearing capacity is common. The reflection points of the deformed pipe wall may overlap with the reflection points of the siltation surface in the radial distance dimension. The joint criterion, by introducing discriminative information from the echo intensity dimension, provides a second discrimination channel for these geometrically ambiguous areas, preventing systematic misseparation due to the limitations of a single feature.
[0072] At the process chain level, the siltation morphology distribution map and siltation spatial distribution curve output in step S5 are built on the 3D point cloud model after global optimization in step S4. The reliability of its quantitative analysis results directly benefits from the improved accuracy of upstream reconstruction. Based on this, pipeline maintenance and scheduling personnel can refine the granularity of dredging decisions from the traditional pipe section level to the cross-section level, identify the precise location and range of concentrated siltation, avoid indiscriminate full-line dredging of the entire pipe section, and thus improve the targeting and resource utilization efficiency of dredging operations.
[0073] Example 2 In this embodiment, steps S1 and S2 are executed in the same way as in embodiment 1. The difference lies in the way the global constraints are established in step S3, which is different from that in embodiment 2. This embodiment no longer relies on spatial proximity search between non-adjacent frames to automatically discover loop constraints. Instead, it uses axial position constraints based on the known geometric prior of the pipeline to achieve global consistency correction, thereby changing the constraint edges involved in optimization in step S4.
[0074] Specifically, before the pipeline inspection operation begins, the axial alignment information of the pipeline is obtained from the pipeline design drawings or an existing pipeline geographic information system database, including the pipeline's horizontal orientation, elevation changes, and bend locations. After completing the local registration of all adjacent frames in step S2, this embodiment applies global constraints to the frame-by-frame accumulated pose sequence using the pipeline axial alignment information. The constraint method is as follows: several key frames are selected in the frame sequence. The selection criterion for key frames is that the axial position of the pipeline corresponding to the frame sequence has clear spatial coordinate annotations in the pipeline design drawings. In this embodiment, the key frames are selected at intervals of approximately 10 meters. For each key frame, its annotated coordinates in the design drawings are used as the prior pose constraint of that frame in the global coordinate system and incorporated into the pose graph in the form of high-confidence edges.
[0075] In the global optimization of step S4, these prior pose constraint edges and the adjacent frame registration constraint edges from step S2 participate in the optimization solution. The value of the information matrix of the prior pose constraint edges is determined by the annotation accuracy of the pipeline design drawing coordinates: when the design drawing coordinate accuracy is high, the information matrix value is large, and the pulling force on the key frame pose is strong; when the design drawing coordinate accuracy is low, the information matrix value decreases accordingly. The objective function and solution algorithm for global optimization are the same as step S4 in Example 1.
[0076] This embodiment is more suitable for application scenarios where the pipeline inspection path is unidirectional and there is no chance of loopback, but the pipeline design drawing information is complete and available, such as the inspection of long straight pipe sections or trunk pipes with a single direction. In such scenarios, Embodiment 1 cannot establish loopback constraints due to the lack of loopback candidate frame pairs, resulting in limited global drift suppression capabilities; while this embodiment introduces prior position information from the pipeline design drawings as an external constraint source, and can still achieve global correction of frame-by-frame cumulative drift without relying on loopback, thereby ensuring the spatial position accuracy of the inspection results for long straight pipe sections.
[0077] Furthermore, this embodiment can be used in conjunction with Embodiment 1: in the detection path where there are bends and overlapping sections in the pipeline, the loop constraints of automatic detection and the prior position constraints of the design drawings are used simultaneously. The two types of constraints provide global consistency correction capability in the pose diagram, further improving the global accuracy of 3D reconstruction.
[0078] The technical concept of this invention is not limited to the sonar detection scenarios of municipal drainage pipelines or industrial water transmission pipelines described in the specific embodiments above. Without departing from the core concept of this invention, this solution is applicable to any application scenario where an acoustic or optical measurement platform advancing along a linear channel faces frame-by-frame registration error accumulation during 3D reconstruction of the inner wall, including but not limited to the detection of inner wall morphology and foreign object accumulation analysis of tunnels, culverts, integrated utility tunnels, oil pipelines, natural gas pipelines, and urban heating pipelines. Similarly, the registration confidence assessment mechanism and loop closure constraint detection mechanism in this solution are not limited to multi-beam sonar data sources; they are also applicable to other sensor types capable of acquiring cross-sectional point cloud data, such as structured light scanners and lidar cross-sectional scanners.
Claims
1. A method for three-dimensional reconstruction and siltation morphology analysis of a multibeam sonar pipeline, characterized in that, Includes the following steps: S1. Perform polar coordinate to rectangular coordinate transformation on the raw echo data collected frame by frame along the pipeline axis by the sonar mounting platform to generate a cross-sectional point cloud containing multiple reflection points, and attach the echo intensity value of each beam as an attribute label to the corresponding reflection point. S2. Perform rigid registration on the point clouds of two adjacent frames to solve the pose transformation matrix. Based on the residual distribution of the paired point pairs in the overlapping area after registration, extract the residual median and residual interquartile range. Combine the two to form the confidence measure of this registration. S3. Concatenate and multiply the pose transformation matrices calculated frame by frame, project the cross-sectional point clouds of all frames onto the temporary global coordinate system, search for frame pairs that are not adjacent in time but are close in space as candidate frame pairs for loop closure, and confirm them as loop closure constraints after performing registration verification. S4. Construct a pose graph with each frame pose as a node, and adjacent frame registration constraints and loop closure constraints as edges. Convert the confidence measure of each edge into an information matrix as constraint weights. Perform global optimization with the goal of minimizing the weighted sum of squared residuals of all edges. Generate a globally consistent 3D point cloud model based on the cross-sectional point cloud corresponding to the optimized pose projection of each frame. S5. Slice the three-dimensional point cloud model section by section along the pipeline axis, and verify the radial distance discrimination based on the nominal cross-section geometric prior of the pipeline and the echo intensity characteristics. Separate the reflection points of the pipe wall surface and the siltation surface, and calculate the siltation morphology parameters for the separated siltation surface reflection points.
2. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, In step S1, a two-dimensional rectangular coordinate system is established with the geometric center of the current frame of the sonar platform as the local origin. The direction angle and echo distance of each beam are converted into reflection point coordinates according to the geometric mapping relationship from polar coordinates to rectangular coordinates. When there is a waterless area above the liquid surface in the pipe section, coordinate transformation is only performed on the beam direction that obtains effective echo.
3. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, In step S2, the rigid registration adopts the iterative nearest point registration algorithm. In each iteration, for each point in the source section point cloud, the nearest corresponding point in the target section point cloud is searched. Based on the spatial position relationship of all corresponding point pairs, the rigid transformation parameter that minimizes the sum of the distances between the point pairs is solved. The rigid transformation parameter is applied to the source section point cloud and the iteration is repeated until the change in the rigid transformation parameter is less than the preset convergence threshold.
4. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 3, characterized in that, When the registration process of two adjacent frame section point clouds fails to meet the convergence threshold after reaching the preset maximum number of iterations, the registration is marked as non-converged, and the confidence metric of the registration is forcibly set to the preset minimum confidence level. In the global optimization in step S4, the pose of the frame is determined by the constraints in the pose graph that are adjacent to the frame and whose confidence metric is not the minimum confidence level, in order to suppress the constraint weights with the minimum confidence level.
5. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, In step S3, the frame pair is set to represent as To mark a frame pair as a loopback candidate, it must simultaneously meet the following two conditions: frame With frames The number of frames separated in the time sequence of the frame sequence exceeds the preset minimum frame interval threshold; With frames The spatial distance in the temporary global coordinate system is less than a preset ratio of the nominal diameter of the pipe.
6. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 5, characterized in that, After performing rigid registration on the candidate frame pairs of loop closures, the median of the residuals and the interquartile range of the residuals for that registration are extracted respectively. The loop closure constraint is confirmed and included in the pose graph of step S4 to participate in global optimization only when the median of the residuals for that registration is less than a preset median of residuals threshold and the interquartile range of the residuals is less than a preset interquartile range threshold.
7. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, In step S4, the information matrix is constructed as follows: the median of the residual and the reciprocal of the interquartile range of the residual are used as the basic values of the diagonal elements of the information matrix. When the median of the residual or the interquartile range of the residual increases, the corresponding value of the information matrix decreases, and the constraint weight of the edge in the global optimization decreases accordingly.
8. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, In step S5, separating the reflection points from the pipe wall surface and the siltation surface includes two stages. The first stage includes: performing preliminary classification based on the comparison between the radial distance deviation of each reflection point from the nominal pipe wall profile and a preset pipe wall tolerance threshold, separating a preliminary subset of pipe wall reflection points and a preliminary subset of siltation surface reflection points. The second stage includes: for reflection points whose radial distance deviation is in the fuzzy area near the pipe wall tolerance threshold, calculating the difference between the echo intensity value of the reflection point and the median echo intensity of the preliminary subset of pipe wall reflection points and the preliminary subset of siltation surface reflection points, and classifying the reflection point into the category with the smaller difference value.
9. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, The frame sampling interval of the sonar mounting platform is one frame acquired every 10 mm to 50 mm of advance; the operating frequency range of the multi-beam sonar transducer array is 600 kHz to 2000 kHz; the preset tube wall tolerance threshold range for radial distance determination in step S5 is 8 mm to 30 mm.
10. The method for three-dimensional reconstruction and siltation morphology analysis of multibeam sonar pipelines according to claim 1, characterized in that, The sedimentation morphology parameters include: The siltation accumulation height is defined as the maximum radial accumulation distance from the lowest point of the pipe bottom at the siltation surface reflection point at the corresponding cross-section location. The siltation cross-sectional area is defined as the area of the closed region enclosed by the reflection point of the siltation surface at the corresponding cross-sectional location. The effective flow area ratio is defined as the ratio of the effective flow area obtained by subtracting the siltation cross-sectional area from the nominal cross-sectional area of the pipe to the nominal cross-sectional area of the pipe.
Citation Information
Patent Citations
Multi-robot collaborative mapping method for post-disaster rescue scene
CN116503576A
Method and device for screening point cloud, equipment and storage medium
CN119494912A
