A method for estimating the height of a retaining wall and the slope of a road surface in a working area of a dump

Abstract

A kind of dump facing retaining wall height and operating area road slope joint estimation method, point cloud preprocessing and ground and non-ground initial segmentation;Retaining wall candidate extraction and path reference construction;Subsection robust estimation and cross-frame fusion;Road slope estimation and joint output;The present application is through the collaborative design of point cloud preprocessing, adaptive candidate extraction, unified reference construction, subsection robust estimation and cross-frame fusion, road slope joint estimation, constructs the joint estimation framework of two core indexes under the condition of complex point cloud of dump, realizes the accurate, stable measurement of retaining wall height and road slope, makes two indexes align under the same arc length index, provides reliable measurement basis for risk classification and graded push of intelligent safety supervision of dump, effectively solves the problems existing in the prior art, such as candidate point set pollution, reference inconsistency, cross-frame jump, etc., adapts to the actual business needs of intelligent monitoring of dump safety production.

Description

Technical Field

[0001] This invention relates to a method for jointly estimating the height of retaining walls and the slope of road surfaces, specifically a method for jointly estimating the height of retaining walls and the slope of road surfaces in work areas for spoil heaps, belonging to the field of three-dimensional point cloud geometric understanding and structured measurement technology in intelligent safety supervision of mines. Background Technology

[0002] With the deepening of intelligent mine construction, the need for real-time, refined, and traceable safety supervision of open-pit coal mine spoil heaps, as high-risk operation areas, is becoming increasingly urgent. Spoil heap terrain continuously evolves with spoil disposal operations, and the dense activity of vehicles and personnel creates a complex geometric environment with multiple structures such as slopes, stockpiles, retaining walls, and roads. The standardization of retaining wall height and the rationality of road slope in the operation area directly affect the driving safety of vehicles and are core monitoring indicators for intelligent safety supervision of spoil heaps.

[0003] In the safety supervision of spoil heaps, retaining walls serve as a critical safety boundary for operating vehicles. Insufficient height, continuous damage, or abnormal wall posture significantly increase the risk of vehicles crossing boundaries, overturning, or rolling over. Road surface slope in the operating area is a core geometric indicator for assessing vehicle traction and braking stability; abnormal slope is highly correlated with the risk of violations such as speeding, sharp turns, and brake instability. Both must be used together for risk classification. For example, the risk level of insufficient retaining wall height under steep road surface conditions needs to be further increased. Therefore, accurate joint measurement of retaining wall height and road surface slope is fundamental to spoil heap risk assessment.

[0004] Traditional methods for obtaining retaining wall height and road slope data rely heavily on manual measurement or infrequent inspections. These methods suffer from drawbacks such as insufficient monitoring coverage, discrete measurement results, and poor traceability. Furthermore, manual on-site measurement poses significant operational safety risks and fails to meet the operational needs of real-time safety monitoring and risk-based data dissemination at spoil heaps. With the widespread deployment of high-definition cameras and high-beam lidar at spoil heaps, 3D point cloud data can provide a precise 3D geometric basis for measuring retaining wall height and road slope. Structured measurement methods based on point clouds have become the mainstream technology for intelligent monitoring of spoil heaps.

[0005] However, point cloud data in spoil heap scenarios have significant non-ideal characteristics, including drastic changes in point density with distance and occlusion, a large amount of outlier noise caused by dust and reflection effects, and high similarity in local geometric features between retaining walls and steep slopes / piles. These factors can easily lead to contamination of candidate point sets and fitting bias during the measurement process, posing challenges to accurate measurement. Existing point cloud-based methods for estimating retaining wall height and road slope typically employ a process of ground and non-ground segmentation, local plane fitting, height difference calculation, and index output. While these methods can achieve basic measurements, they still have significant shortcomings in complex scenarios such as spoil heaps: First, the candidate point set is heavily polluted. Retaining walls are spatially adjacent to non-target structures such as slopes, stockpiles, and vehicles. Extracting candidate points for retaining walls using only fixed normal or height thresholds easily introduces a large number of non-retaining wall points, leading to systematic biases in subsequent fitting results. Second, measurement benchmarks are inconsistent. Retaining wall height and road slope are often estimated independently in different spatial segments or reference systems, lacking a unified structural coordinate system. This results in the two indicators at the same location not being accurately aligned, making it difficult to form a consistent basis for risk assessment. Third, cross-frame estimation stability is poor. During online monitoring, changes in point cloud occlusion and sparsity can cause jumps in local fitting results. Existing methods lack reasonable temporal consistency constraints, easily generating false alarms or missed alarms, and failing to meet the needs of continuous monitoring. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the present invention aims to provide a joint estimation method for retaining wall height and road surface slope in spoil heaps. By adaptive retaining wall candidate extraction, unified structural coordinate system construction, and cross-frame consistency fusion, the method achieves accurate and stable joint estimation of retaining wall height and road surface slope in the work area. This method effectively solves the problems of candidate point set contamination, benchmark inconsistency, and cross-frame jumps, thereby improving the measurement accuracy and stability of intelligent safety supervision of spoil heaps.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for jointly estimating the height of retaining walls and the slope of the road surface in a spoil heap includes the following steps: S1. Point cloud preprocessing and initial segmentation of ground and non-ground surfaces; Input point cloud A finite screening was performed to obtain an effective point set, in which Representing the coordinate vectors of 3D points, multi-scale downsampling and robust denoising are performed to form a stable point set. Robust estimation of the local plane is then performed under spatial partitioning conditions. Based on deviations and consistency checks from points to the local plane, the ground point set is completed. Non-ground point set The initial segmentation; S2. Retaining wall candidate extraction and path baseline construction; non-ground point set In this process, a regionalized robust plane estimation is performed, and a structure confidence score is constructed through multiple features to adaptively filter candidate regions for retaining walls, which are then aggregated to obtain a set of candidate points for retaining walls. In the candidate point set of the retaining wall Estimate the main direction of the structure and construct the central path. A unified index is obtained through arc length parameterization, which serves as a common benchmark for the joint estimation of retaining walls and road surfaces; S3, segmented robust estimation and cross-frame fusion; Candidate point set for retaining wall under arc length index The system is divided into multiple segmented point sets. Robust plane estimation and analytical refinement are performed on each segment to obtain local attitude parameters of the wall and estimates of the retaining wall height. Uncertainty metrics are calculated, and a fusion strategy of gated update and confidence weighting is adopted on the time series to suppress anomalous jumps and maintain the ability to respond to real structural changes. S4. Road slope estimation and joint output; Extract the boundary zone ground point set of the retaining wall neighborhood from the ground point set. The arc length index of the center path is reused to achieve accurate alignment with the retaining wall segments. The local plane of the road surface is fitted for each segment to obtain the road slope estimate. Finally, the retaining wall height, wall attitude, road slope and corresponding confidence metric are jointly output under the same arc length reference.

[0008] Step S1 of the present invention is as follows: S11. Let the input point cloud be... , Given a three-dimensional coordinate vector, if the point cloud carries intensity information, then... , As an intensity component, it can assist in the identification of outlier noise; S12, Limited Screening: Elimination , , For points with invalid values, obtain the set of valid points. To avoid the impact of numerical instability on subsequent processing; S13. Multi-scale voxel downsampling: Set voxel side length For the effective point set Perform voxel mapping, selecting the point closest to the voxel center for each voxel as a representative point to obtain a set of downsampling points. This reduces computational complexity while maintaining the macroscopic geometric structure. S14. Statistical Outlier Suppression: Set the neighborhood size With standard deviation threshold ,right Construct each point Nearest neighbor set: Calculate the mean and standard deviation of the Euclidean distances within the neighborhood, and remove neighborhoods with distances greater than the mean. The outliers, multiplied by the standard deviation, yield the set of denoised points. This is used to construct the neighborhood set of each point and statistically analyze the neighborhood distance distribution to eliminate outlier noise points; S15. Spatial Blocking: Set block dimensions in the XY plane. , denoise set Divided into region subsets , The number of blocks is used to divide the grid in the XY plane according to the preset block size, and the stable point set is divided into multiple region subsets. S16. Ground Seed Selection: Set Elevation Level. For each region subset According to elevation Sort in ascending order and take the first few. Each point serves as a ground seed. , for The points are used to sort each region subset in ascending order of elevation, and the low-quantile subset is taken as the ground seed to reduce interference from non-ground structures. S17, Local Ground Plane Estimation: For ground seeds Perform least-squares plane fitting, the plane equation is: ,in It is a unit normal vector and , For the plane offset term, the algebraic distance from the point to the plane. ,because Normalization is used, and algebraic distance is equivalent to geometric distance. Based on this, least-squares plane fitting is performed on the ground seed to obtain plane parameters. S18. Ground and Non-Ground Segmentation: Set Adaptive Distance Threshold ,like ,but ,otherwise Complete the ground point set Non-ground point set The initial segmentation is then performed, based on the geometric distance from each point to the local plane, and combined with an adaptive threshold to divide the point cloud into ground point sets. Non-ground point set .

[0009] Step S2 of the present invention is as follows: S21. Non-ground regionization: Transforming non-ground point sets... Block Scale in the XY Plane Divide into blocks to obtain region subsets The number of points to be removed is less than the minimum threshold. This subset is used to obtain the subset of region points that satisfies the minimum number of points threshold, i.e., the effective subset of region points. This ensures the statistical sufficiency of a robust fit; S22, Robust Plane Fitting of Regions: For each subset of valid region points... Perform Random Consistent Sampling (RANSAC) estimation and set a distance threshold. With the number of iterations To obtain the plane parameters With interior point set ; S23. Multi-feature calculation: Calculate the planar consistency measure, residual scale measure, normal geometric measure, and point density measure for each region. Specifically: Planar consistency characteristics: Interior point ratio , for The number of points describes the planar consistency of the region; Residual scaling characteristics: Quantile intervals of inlier residuals , , for Quantile operator to suppress non-Gaussian tail effects; Normal geometric metric features: , for The vertical component, The smaller the value, the closer the plane is to vertical, which conforms to the geometric priors of the retaining wall; Point density metric features: , for The minimum bounding box area in the XY plane represents the sufficiency of local sampling, which is used to comprehensively characterize the regional structural features. S24. Construction of structural confidence: Constructing regional structural confidence by fusing multiple features including planar consistency measure, residual scale measure, normal geometric measure, and point density measure. in: , These are the weighting coefficients; This is the adaptive residual scaling normalization term; The coefficient of variation is the residual. As a global adaptive density normalization benchmark; S25. Adaptive Candidate Filtering: Setting Confidence Thresholds With normal threshold ,like and ,but For each candidate point of the retaining wall, a set of candidate retaining wall points is obtained by aggregating all candidate retaining wall points. This is used to filter candidate regions for retaining walls based on structural confidence and normal threshold, and then aggregate them to obtain a set of candidate points for retaining walls. ; S26. Center Path Construction: Calculate the candidate point set for the retaining wall. 3D mean From the covariance matrix The principal direction of the structure is obtained from the eigenvector corresponding to the largest eigenvalue. For each Calculate the projected scalar ,according to The sequence is obtained by sorting in ascending order. The Ramer–Douglas–Peucker criterion was used to simplify the path and obtain the central path control point. Connect the control points to obtain the central path ; S27, Arc Length Parametric: Calculate Center Path arc length sequence , Total path length Each control point corresponds to an arc length index. This forms a unified structural coordinate system.

[0010] Step S3 of the present invention specifically includes: S31. Valid Candidate Constraints: Valid candidate constraints: For each Calculate its path to the center lateral distance in the XY plane Construct distance set Take quantiles As the width of the corridor Remove The points are used to obtain the set of effective retaining wall candidate points. This allows for adaptive estimation of corridor width, eliminating points far from the main retaining wall structure, and improving the purity of the candidate point set. S32, Arc Length Segment Mapping: Sets segment length , central path Divided into There are segments, and the segment index is... For each Calculate its position on the central path Arc length corresponding to the projection point on Mapped to segments The segmented point set is obtained. ; S33, Piecewise Robust Plane Fitting: For each piecewise point set... Perform RANSAC robust plane fitting to obtain the initial plane parameters. With interior point set ; S34. Analytical Refinement: For the set of interior points Constructing the covariance matrix The eigenvector corresponding to the smallest eigenvalue is taken as the refinement normal. If one diligently cultivates the Dharma direction vertical component Then let This ensures that the normal orientation faces the inside of the spoil heap, thereby unifying the normal orientation. S35. Wall attitude and height estimation: based on refined normal. Generate wall pose related quantities Wall posture related quantities , The retaining wall height estimate is generated by using the elevation statistics of points within segments. Retaining wall height estimation ,in for mean elevation The average ground elevation of the area surrounding the retaining wall. Spatial correlation weights are introduced, along with a temporal smoothing correction term. This is to improve the robustness of the estimation; S36. Uncertainty Measure Construction: An adaptive uncertainty measure is constructed by integrating the global residual mean, in-point residual skewness, and in-point elevation distribution entropy. ,in , The mean of the global residual scale. For interior point residual skewness, The information entropy of the elevation distribution of interior points; S37. Cross-frame confidence weighted fusion: Setting a gate threshold ,like Then the adaptive smoothing coefficient ,otherwise A confidence-weighted adaptive gating fusion formula is used to fuse the current frame estimate with the historical state. The cross-frame fusion formula is as follows: Quantities related to wall posture Using the same fusion strategy, time-smoothed results were obtained. ;in The adaptive smoothing coefficient is modulated by the uncertainty and consistency test results to achieve a balance between stability and responsiveness in time series estimation.

[0011] Step S4 of the present invention specifically includes: S41. Boundary Zone Ground Extraction: Set Neighborhood Radius Set up with points inside the retaining wall To query the set, on the ground point set Perform a K-nearest neighbor search to obtain the set of ground points on the boundary zone of the retaining wall's neighborhood. ; S42. Neighbor exclusion and pollution suppression: Set exclusion radius. Remove the boundary zone ground point set Set of points inside the retaining wall Distance less than exclusion radius The points are used to obtain a pure set of ground points. This is to eliminate ground points whose distance from the retaining wall is less than a preset threshold, thereby reducing the bias of the bottom echo of the retaining wall on the road surface measurement. S43. Segment Alignment: Reuse the arc length index of the center path. Set up a clean ground point set The points in the graph are mapped to the corresponding segments to obtain a set of segmented ground points that are consistent with the retaining wall segments. ; S44. Road segment slope estimation: For each segment's ground point set... A robust plane fitting is performed to obtain the road surface normal vector. Road surface segment slope Introducing pollution correction factors With the spatial smoothing compensation term, a robust estimate is obtained. This is to improve the accuracy of the estimation; S45. Overall road surface slope estimation: A weighted robust global estimation combined with regional reliability weighting is used to generate the overall road surface slope. Overall road surface slope ,in , representing the credibility weight of each segment, and the overall road surface slope. Used as a macroscopic road surface slope baseline for global risk assessment; S46. Structured Joint Output: Indexed by Arc Length To unify the primary key, construct a structured output result. This ensures that the retaining wall and road surface indicators correspond one-to-one in the same structural coordinate system; S47. Preservation of traceable evidence: For each segment Preserve the segmented point set Segmented ground point set Point cloud index and planar parameters , Statistics such as in-point ratio and residual scale provide a basis for auditing, visualization and anomaly playback of measurement results.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a joint estimation method for retaining wall height and road surface slope in spoil heaps. Through point cloud preprocessing, adaptive candidate extraction, unified benchmark construction, piecewise robust estimation and cross-frame fusion, and joint estimation of road surface slope, a joint estimation framework for two core indicators under complex point cloud conditions in spoil heaps is constructed. This achieves accurate and stable measurement of retaining wall height and road surface slope, aligning these two indicators under the same arc length index. This provides a reliable measurement basis for risk classification and hierarchical push for intelligent safety supervision of spoil heaps, effectively solving problems such as candidate point set contamination, benchmark inconsistency, and cross-frame jumps in existing methods, and adapting to the actual business needs of intelligent monitoring of safety production in spoil heaps.

[0013] The multi-scale point cloud preprocessing and ground and non-ground initial segmentation mechanism constructed in this invention effectively suppresses outlier noise in the spoil heap point cloud through finite screening, multi-scale downsampling, statistical outlier suppression, and spatial block robust plane fitting. It reduces the interference of non-ground structures on ground seed selection and achieves robust segmentation of ground and non-ground areas. This provides a high-quality point cloud foundation for subsequent retaining wall candidate extraction and road slope estimation, and alleviates the problem of poor processing effect of single preprocessing methods under non-ideal point cloud conditions in spoil heaps.

[0014] Based on this, the structural confidence introduced in this invention enables adaptive extraction of retaining wall candidate points. It integrates multi-regional statistical features such as planar consistency, residual scale, normal geometry, and point density to construct confidence, replacing the traditional fixed threshold screening method. This allows candidate extraction to adapt to the characteristics of non-stationary point cloud density, noise scale variation, and high structural similarity in spoil heaps. It can effectively distinguish between retaining walls and similar structures such as slopes and piles, significantly reducing the probability of candidate point set contamination and avoiding systematic bias in measurement results from the source. At the same time, it provides high-purity candidate point input for subsequent segmented estimation.

[0015] The unified structural benchmark of the central path arc length index constructed in this invention transforms the measurement problem of retaining walls and road surfaces into a measurement problem under a one-dimensional ordered index, enabling the retaining wall height and road surface slope to be accurately aligned at the same arc length position. This solves the benchmark inconsistency and spatial mismatch problems caused by independent estimation in existing methods, allowing the two indicators of retaining wall height and road surface slope to be directly used together for risk classification of spoil heaps, significantly reducing the post-processing complexity of subsequent risk assessment, and improving the interpretability and practicality of measurement results.

[0016] The piecewise robust estimation and cross-frame confidence-weighted fusion mechanism proposed in this invention improves the accuracy of local measurements by piecewise fitting and analytical refinement of retaining wall candidate points. Simultaneously, it constructs an adaptive uncertainty metric and employs gated updates and confidence-weighted fusion to achieve cross-frame fusion. This effectively suppresses abnormal estimations caused by point cloud occlusion, echo loss, or local plane degradation, and allows for a smooth regression to the true value when observation quality is restored. It reduces estimation variance in the time dimension and maintains piecewise analytical performance in the spatial dimension, significantly improving the cross-frame stability of estimation results in online monitoring and effectively reducing false alarms and missed alarms.

[0017] This invention reuses the unified arc length benchmark of the retaining wall in the road slope estimation process to achieve precise alignment with the retaining wall segments. It also suppresses pollution from the bottom echo of the retaining wall through nearest neighbor exclusion, introduces correction factors and compensation terms to improve the robustness of the road slope estimation, and outputs both segmented slope and overall slope. This satisfies the needs of local risk assessment and provides a baseline for global risk assessment. It constructs a joint estimation and output framework for retaining wall height and road slope, which fully considers the synergy between the two in spatial location and risk assessment, and is adapted to the actual business needs of intelligent safety supervision of spoil heaps.

[0018] Overall, the present invention has the following significant advantages over the prior art: The structure confidence of multi-feature fusion is used to achieve adaptive extraction of candidate points for retaining walls, replacing the traditional fixed threshold method. This effectively solves the problem of candidate point set contamination caused by the non-ideal characteristics of the point cloud of the spoil heap, and reduces the systematic bias of the measurement results. By constructing a center path arc length index as a unified structural benchmark, the retaining wall height and road slope are accurately aligned under the same index position, which solves the spatial mismatch problem caused by inconsistent benchmarks. It can be directly used for risk classification, improving the practicality of measurement results. By introducing a cross-frame confidence weighted fusion and gated update strategy, and combining it with an adaptive uncertainty metric to modulate the update intensity, the cross-frame estimation jump caused by instantaneous misfitting is effectively suppressed, thereby improving the stability and reliability of online monitoring. A joint estimation framework for retaining wall height and road slope is constructed to achieve coordinated measurement and structured output of the two indicators, while retaining traceable geometric evidence, thus meeting the coordination and traceability requirements of spoil heap risk assessment. The invention adopts a modular, end-to-end design, with each link working together and the computational complexity being controllable. It has good engineering feasibility, can be adapted to lidar point cloud monitoring systems for different open-pit mine spoil heaps, and has strong generalization ability. This invention reuses the unified benchmark of retaining walls for road slope estimation. It effectively reduces the interference of retaining wall structures through nearest neighbor exclusion and robust fitting, while outputting segmented and overall slopes to meet the needs of both local and global risk assessment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the process of ternary adaptive candidate extraction and structure confidence construction in this invention; Figure 3 This is a schematic diagram illustrating the principles of retaining wall path estimation, arc length parameterization, and piecewise mapping in this invention. Figure 4 This is a structured schematic diagram illustrating the combined output of retaining wall height, wall orientation, and road slope under the unified path reference of this invention. Detailed Implementation

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] This invention aims to achieve accurate and stable joint estimation of retaining wall height and road surface slope in complex point cloud conditions at spoil heaps. It addresses the problems of candidate point set contamination, inconsistent benchmarks, and cross-frame jumps in existing methods. A unified technical framework is constructed, encompassing point cloud preprocessing and initial segmentation, adaptive extraction of retaining wall candidates and construction of a unified benchmark, piecewise robust estimation and cross-frame fusion, and joint estimation and structured output of road surface slope. The system uses 3D point clouds of the spoil heap acquired by LiDAR as input. Through multi-stage collaborative design, it ultimately outputs retaining wall height, wall orientation, road surface slope, and corresponding reliability metrics under the same arc length index, while retaining traceable geometric evidence. This provides reliable measurement data for intelligent safety supervision of spoil heaps. The overall structure of this invention is as follows: Figure 1 As shown.

[0022] This invention proposes a method for jointly estimating the height of retaining walls and the slope of road surfaces in a spoil heap, comprising the following steps: S1. Point cloud preprocessing and initial segmentation of ground and non-ground surfaces; Input point cloud A finite screening was performed to obtain an effective point set, in which Representing the coordinate vectors of 3D points, multi-scale downsampling and robust denoising are performed to form a stable point set. Robust estimation of the local plane is then performed under spatial partitioning conditions. Based on deviations and consistency checks from points to the local plane, the ground point set is completed. Non-ground point set The initial segmentation; Step S1 is as follows: S11. Let the input point cloud be... , Given a three-dimensional coordinate vector, if the point cloud carries intensity information, then... , As an intensity component, it can assist in the identification of outlier noise; S12, Limited Screening: Elimination , , For points with invalid values ​​(such as NaN, Inf), obtain the set of valid points. To avoid the impact of numerical instability on subsequent processing; S13. Multi-scale voxel downsampling: Set voxel side length (Based on the scale of the spoil heap scenario, take 0.2~0.5m), for the effective point set Perform voxel mapping, selecting the point closest to the voxel center for each voxel as a representative point to obtain a set of downsampling points. This reduces computational complexity while maintaining the macroscopic geometric structure. S14. Statistical Outlier Suppression: Set the neighborhood size (Take a value of 20~30) and the standard deviation threshold (Take a value of 1.5~2.0), for Construct each point Nearest neighbor set: Calculate the mean and standard deviation of the Euclidean distances within the neighborhood, and remove neighborhoods with distances greater than the mean. The outliers, multiplied by the standard deviation, yield the set of denoised points. This is used to construct the neighborhood set of each point and statistically analyze the neighborhood distance distribution to eliminate outlier noise points; S15. Spatial Blocking: Set block dimensions in the XY plane. (Take a value of 5~10m) and set the noise points. Divided into region subsets , The number of blocks is used to divide the grid in the XY plane according to the preset block size, and the stable point set is divided into multiple region subsets. S16. Ground Seed Selection: Set Elevation Level. (Take a value of 0.2~0.3), for each <s> Region subset< / s> According to elevation Sort in ascending order and take the first few. Each point serves as a ground seed. , for The points are used to sort each region subset in ascending order of elevation, and the low-quantile subset is taken as the ground seed to reduce interference from non-ground structures. S17, Local Ground Plane Estimation: For ground seeds Perform least-squares plane fitting, the plane equation is: ,in It is a unit normal vector and , For the plane offset term, the algebraic distance from the point to the plane. ,because Normalization is used, and algebraic distance is equivalent to geometric distance. Based on this, least-squares plane fitting is performed on the ground seed to obtain plane parameters. S18. Ground and Non-Ground Segmentation: Set Adaptive Distance Threshold (Obtained from regional point cloud roughness statistics, taking values ​​of 0.1~0.3m), if ,but ,otherwise Complete the ground point set Non-ground point set The initial segmentation is then performed, based on the geometric distance from each point to the local plane, and combined with an adaptive threshold to divide the point cloud into ground point sets. Non-ground point set .

[0023] S2. Retaining wall candidate extraction and path baseline construction; non-ground point set In this process, a regionalized robust plane estimation is performed, and a structure confidence score is constructed through multiple features to adaptively filter candidate regions for retaining walls, which are then aggregated to obtain a set of candidate points for retaining walls. In the candidate point set of the retaining wall Estimate the main direction of the structure and construct the central path. A unified index is obtained through arc length parameterization, which serves as a common benchmark for the joint estimation of retaining walls and road surfaces; Step S2 is as follows: S21. Non-ground regionization: Transforming non-ground point sets... Block Scale in the XY Plane Divide the region into blocks of 2-3m to obtain a subset of the region. The number of points to be removed is less than the minimum threshold. The subset of (50) points is used to obtain the subset of region points that meets the minimum point count threshold, which is the effective subset of region points. This ensures the statistical sufficiency of a robust fit; S22, Robust Plane Fitting of Regions: For each subset of valid region points... Perform Random Consistent Sampling (RANSAC) estimation and set a distance threshold. (Take a value of 0.15~0.25m) and the number of iterations (Take a value of 200~300) to obtain the plane parameters. With interior point set ; S23. Multi-feature calculation: Calculate the planar consistency measure, residual scale measure, normal geometric measure, and point density measure for each region. Specifically: Planar consistency characteristics: Interior point ratio , for The number of points describes the planar consistency of the region; Residual scaling characteristics: Quantile intervals of inlier residuals , , for Quantile operator to suppress non-Gaussian tail effects; Normal geometric metric features: , for The vertical component, The smaller the value, the closer the plane is to vertical, which conforms to the geometric priors of the retaining wall; Point density metric features: , for The minimum bounding box area in the XY plane represents the sufficiency of local sampling, which is used to comprehensively characterize the regional structural features. S24. Construction of structural confidence: Constructing regional structural confidence by fusing multiple features including planar consistency measure, residual scale measure, normal geometric measure, and point density measure. in: , These are the weighting coefficients; This invention , , ; This is the adaptive residual scaling normalization term; The coefficient of variation is the residual. As a global adaptive density normalization benchmark; S25. Adaptive Candidate Filtering: Setting Confidence Thresholds (Take a value of 0.4~0.5) and the normal threshold (Take a value of 0.3~0.4), if and ,but For each candidate point of the retaining wall, a set of candidate retaining wall points is obtained by aggregating all candidate retaining wall points. This is used to filter candidate regions for retaining walls based on structural confidence and normal threshold, and then aggregate them to obtain a set of candidate points for retaining walls. ; S26. Center Path Construction: Calculate the candidate point set for the retaining wall. 3D mean From the covariance matrix The principal direction of the structure is obtained from the eigenvector corresponding to the largest eigenvalue. For each Calculate the projected scalar ,according to The sequence is obtained by sorting in ascending order. The Ramer–Douglas–Peucker criterion (error threshold) is adopted. The path is simplified by taking a distance of 0.5~1.0m to obtain the center path control point. Connect the control points to obtain the central path ; S27, Arc Length Parametric: Calculate Center Path arc length sequence , Total path length Each control point corresponds to an arc length index. This forms a unified structural coordinate system.

[0024] S3, segmented robust estimation and cross-frame fusion; Candidate point set for retaining wall under arc length index The system is divided into multiple segmented point sets. Robust plane estimation and analytical refinement are performed on each segment to obtain local attitude parameters of the wall and estimates of the retaining wall height. Uncertainty metrics are calculated, and a fusion strategy of gated update and confidence weighting is adopted on the time series to suppress anomalous jumps and maintain the ability to respond to real structural changes. Step S3 is as follows: S31. Valid Candidate Constraints: Valid candidate constraints: For each Calculate its path to the center lateral distance in the XY plane Construct distance set Take quantiles As the width of the corridor Remove The points are used to obtain the set of effective retaining wall candidate points. This allows for adaptive estimation of corridor width, eliminating points far from the main retaining wall structure, and improving the purity of the candidate point set. S32, Arc Length Segment Mapping: Sets segment length (Take a length of 3-5m), and mark the center path. Divided into There are segments, and the segment index is... For each Calculate its position on the central path Arc length corresponding to the projection point on Mapped to segments The segmented point set is obtained. , Segment numbering; S33, Piecewise Robust Plane Fitting: For each piecewise point set... Perform RANSAC robust plane fitting to obtain the initial plane parameters. With interior point set ; S34. Analytical Refinement: For the set of interior points Constructing the covariance matrix The eigenvector corresponding to the smallest eigenvalue is taken as the refinement normal. If one diligently cultivates the Dharma direction vertical component Then let This ensures that the normal orientation faces the inside of the spoil heap, thereby unifying the normal orientation. S35. Wall attitude and height estimation: based on refined normal. Generate wall pose related quantities Wall posture related quantities , The retaining wall height estimate is generated by using the elevation statistics of points within segments. Retaining wall height estimation ,in for mean elevation The average ground elevation of the area surrounding the retaining wall. Spatial correlation weights are introduced, along with a temporal smoothing correction term. This is to improve the robustness of the estimation; S36. Uncertainty Measure Construction: An adaptive uncertainty measure is constructed by integrating the global residual mean, in-point residual skewness, and in-point elevation distribution entropy. ,in In this invention , , , The mean of the global residual scale. For interior point residual skewness, The information entropy of the elevation distribution of interior points; S37. Cross-frame confidence weighted fusion: Setting a gate threshold (Take a value of 0.2~0.3), if Then the adaptive smoothing coefficient ,otherwise A confidence-weighted adaptive gating fusion formula is used to fuse the current frame estimate with the historical state. The cross-frame fusion formula is as follows: Quantities related to wall posture Using the same fusion strategy, time-smoothed results were obtained. ;in The adaptive smoothing coefficient is modulated by the uncertainty and consistency test results to achieve a balance between stability and responsiveness in time series estimation.

[0025] S4. Road slope estimation and joint output; Extract the boundary zone ground point set of the retaining wall neighborhood from the ground point set. The arc length index of the center path is reused to achieve accurate alignment with the retaining wall segments. The local plane of the road surface is fitted for each segment to obtain the road slope estimate. Finally, the retaining wall height, wall attitude, road slope and corresponding confidence measure are jointly output under the same arc length reference. Step S4 is as follows: S41. Boundary Zone Ground Extraction: Set Neighborhood Radius (Take a height of 1~2m), and use the points inside the retaining wall as the meeting point. To query the set, on the ground point set Perform a K-nearest neighbor search to obtain the set of ground points on the boundary zone of the retaining wall's neighborhood. ; S42. Neighbor exclusion and pollution suppression: Set exclusion radius. (Take a depth of 0.3~0.5m), remove the ground point set in the boundary zone. Set of points inside the retaining wall Distance less than exclusion radius The points are used to obtain a pure set of ground points. This is to eliminate ground points whose distance from the retaining wall is less than a preset threshold, thereby reducing the bias of the bottom echo of the retaining wall on the road surface measurement. S43. Segment Alignment: Reuse the arc length index of the center path. Set up a clean ground point set The points in the graph are mapped to the corresponding segments to obtain a set of segmented ground points that are consistent with the retaining wall segments. ; S44. Road segment slope estimation: For each segment's ground point set... A robust plane fitting is performed to obtain the road surface normal vector. Road surface segment slope Introducing pollution correction factors With the spatial smoothing compensation term, a robust estimate is obtained. This is to improve the accuracy of the estimation; S45. Overall road surface slope estimation: A weighted robust global estimation combined with regional reliability weighting is used to generate the overall road surface slope. Overall road surface slope ,in , representing the credibility weight of each segment, and the overall road surface slope. Used as a macroscopic road surface slope baseline for global risk assessment; S46. Structured Joint Output: Indexed by Arc Length To unify the primary key, construct a structured output result. This ensures that the retaining wall and road surface indicators correspond one-to-one in the same structural coordinate system; S47. Preservation of traceable evidence: For each segment Preserve the segmented point set Segmented ground point set Point cloud index and planar parameters , Statistics such as in-point ratio and residual scale provide a basis for auditing, visualization and anomaly playback of measurement results.

[0026] Example To verify the feasibility, effectiveness, and engineering applicability of the proposed method for jointly estimating retaining wall height and road slope for spoil heaps, a field application system was built based on an intelligent safety supervision project for a spoil heap in an open-pit coal mine. Based on the actual lidar point cloud data collected on site, full-condition functional verification and on-site performance evaluation were carried out. The verification process fully conformed to the actual operation scenario of the spoil heap and the application requirements of the project.

[0027] Both experimental and field application data were collected using real point cloud data from spoil heaps acquired by a 64-line lidar system. The equipment's acquisition frequency was 10Hz, with a point cloud density of 500-1500 points / m². The acquisition range covered the core operating area of ​​the spoil heap, 160m × 200m. The data included typical spoil heap structures such as retaining walls, operating roads, stockpiles, and operating vehicles. It also fully captured the non-ideal characteristics of point clouds encountered in actual spoil heap operations, such as dust and noise, sparse obstruction, and undulating terrain. The data conditions were highly consistent with the actual usage scenario at the project site. This field evaluation set key indicators based on the core application requirements of the project, including Measurement Error (MAE), Cross-Frame Standard Deviation (STD), Spatial Alignment Accuracy (Acc), and Candidate Point Extraction Purity. These indicators respectively characterize the measurement accuracy, cross-frame stability, indicator spatial alignment, and candidate point set effectiveness of the method in field use, all of which are core focus indicators for the intelligent safety supervision project of spoil heaps.

[0028] Field application verification results show that the method of this invention exhibits excellent application performance under actual working conditions of spoil heaps: the MAE for retaining wall height measurement is 0.08m, and the MAE for road slope measurement is 0.6°, fully meeting the actual requirements of the project for measurement accuracy; the STD across frames for retaining wall height is 0.05m, and the STD across frames for road slope is 0.3°, demonstrating good cross-frame stability and effectively avoiding interference from measurement result jumps to on-site supervision; the spatial alignment accuracy reaches 98.2%, achieving accurate matching of retaining wall height, wall orientation, and road slope under a unified spatial benchmark, meeting the project's requirements for structured data output. Simultaneously, the method of this invention achieves a candidate purity of 92.5% in the candidate point extraction stage, effectively filtering out interference factors such as stockpiles, vehicles, dust, and noise, solving the problem of candidate point set contamination under complex working conditions of spoil heaps from the source, and significantly improving the efficiency and accuracy of subsequent estimation stages.

[0029] In practical applications under various working conditions at project sites, the method of this invention can be adapted to different operating areas, different weather conditions (dust, light rain) and different equipment installation locations at spoil heaps. It can maintain stable estimation performance under non-ideal conditions such as point cloud obstruction, uneven density, and undulating terrain, without significant performance degradation, and is fully adaptable to the complex operating environment at project sites.

[0030] In summary, the method of this invention, applied in a field project for intelligent safety monitoring of an open-pit coal mine spoil heap, fully verified its engineering feasibility and practical application value in jointly estimating retaining wall height and road slope under complex point cloud conditions at the spoil heap. This invention, through a collaborative design of adaptive candidate extraction based on structural confidence, unified arc length benchmark construction, cross-frame confidence weighted fusion, and multi-indicator joint estimation, fundamentally solves practical problems existing in field applications of spoil heaps, such as candidate point set contamination, inconsistency of multi-indicator benchmarks, and cross-frame jumps in measurement results. The output structured results of arc length, retaining wall height, wall attitude, and road slope can be directly connected to the risk classification and graded push module of intelligent safety monitoring of spoil heaps in the project, achieving seamless integration of monitoring data to field applications. This method requires no complex parameter debugging, is adaptable to various actual operating scenarios of open-pit mine spoil heaps, and possesses good engineering feasibility, field adaptability, and promotional application value. It can be directly applied to the engineering practice of intelligent monitoring of safety production at open-pit mine spoil heaps.

[0031] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for jointly estimating the height of retaining walls and the slope of road surfaces in a spoil heap, characterized in that, Includes the following steps: S1. Point cloud preprocessing and initial segmentation of ground and non-ground surfaces; Input point cloud A finite screening was performed to obtain an effective point set, in which Representing the coordinate vectors of 3D points, multi-scale downsampling and robust denoising are performed to form a stable point set. Robust estimation of the local plane is then performed under spatial partitioning conditions. Based on deviations and consistency checks from points to the local plane, the ground point set is completed. Non-ground point set The initial segmentation; S2. Retaining wall candidate extraction and path baseline construction; non-ground point set In this process, a regionalized robust plane estimation is performed, and a structure confidence score is constructed through multiple features to adaptively filter candidate regions for retaining walls, which are then aggregated to obtain a set of candidate points for retaining walls. In the candidate point set of the retaining wall Estimate the main direction of the structure and construct the central path. A unified index is obtained through arc length parameterization, which serves as a common benchmark for the joint estimation of retaining walls and road surfaces; S3, segmented robust estimation and cross-frame fusion; Candidate point set for retaining wall under arc length index The system is divided into multiple segmented point sets. Robust plane estimation and analytical refinement are performed on each segment to obtain local attitude parameters of the wall and estimates of the retaining wall height. Uncertainty metrics are calculated, and a fusion strategy of gated update and confidence weighting is adopted on the time series to suppress anomalous jumps and maintain the ability to respond to real structural changes. S4. Road slope estimation and joint output; Extract the boundary zone ground point set of the retaining wall neighborhood from the ground point set. The arc length index of the center path is reused to achieve accurate alignment with the retaining wall segments. The local plane of the road surface is fitted for each segment to obtain the road slope estimate. Finally, the retaining wall height, wall attitude, road slope and corresponding confidence metric are jointly output under the same arc length reference.

2. The method for jointly estimating the height of retaining walls and the slope of road surfaces in a work area facing a spoil heap as described in claim 1, characterized in that, Step S1 is as follows: S11. Let the input point cloud be... , Given a three-dimensional coordinate vector, if the point cloud carries intensity information, then... , As an intensity component, it can assist in the identification of outlier noise; S12, Limited Screening: Elimination , , For points with invalid values, obtain the set of valid points. ; S13. Multi-scale voxel downsampling: Set voxel side length For the effective point set Perform voxel mapping, selecting the point closest to the voxel center for each voxel as a representative point to obtain a set of downsampling points. ; S14. Statistical Outlier Suppression: Set the neighborhood size With standard deviation threshold ,right Construct each point Nearest neighbor set: Calculate the mean and standard deviation of the Euclidean distances within the neighborhood, and remove neighborhoods with distances greater than the mean. The outliers, multiplied by the standard deviation, yield the set of denoised points. ; S15. Spatial Blocking: Set block dimensions in the XY plane. , denoise set Divided into region subsets , Number of blocks; S16. Ground Seed Selection: Set Elevation Level. For each region subset According to elevation Sort in ascending order and take the first few. Each point serves as a ground seed. , for Points; S17, Local Ground Plane Estimation: For ground seeds Perform least-squares plane fitting, the plane equation is: ,in It is a unit normal vector and , For the plane offset term, the algebraic distance from the point to the plane. ,because Normalization: algebraic distance is equivalent to geometric distance; S18. Ground and Non-Ground Segmentation: Set Adaptive Distance Threshold ,like ,but ,otherwise Complete the ground point set Non-ground point set The initial segmentation.

3. The method for jointly estimating the height of retaining walls and the slope of road surfaces in a work area facing a spoil heap as described in claim 2, characterized in that, Step S2 is as follows: S21. Non-ground regionization: Transforming non-ground point sets... Block Scale in the XY Plane Divide into blocks to obtain region subsets The number of points to be removed is less than the minimum threshold. This subset is used to obtain the subset of region points that satisfies the minimum number of points threshold, i.e., the effective subset of region points. ; S22, Robust Plane Fitting of Regions: For each subset of valid region points... Perform Random Consistent Sampling (RANSAC) estimation and set a distance threshold. With the number of iterations To obtain the plane parameters With interior point set ; S23. Multi-feature calculation: Calculate the planar consistency measure, residual scale measure, normal geometric measure, and point density measure for each region. Specifically: Planar consistency quantity characteristics: Interior point ratio , for The number of points describes the planar consistency of the region; Residual scaling characteristics: Quantile intervals of inlier residuals , , for Quantile operator to suppress non-Gaussian tail effects; Normal geometric metric features: , for The vertical component, The smaller the value, the closer the plane is to vertical, which conforms to the geometric priors of the retaining wall; Point density metric features: , for The minimum bounding box area in the XY plane characterizes the sufficiency of local sampling; S24. Construction of structural confidence: Constructing regional structural confidence by fusing multiple features including planar consistency measure, residual scale measure, normal geometric measure, and point density measure. in: , These are the weighting coefficients; This is the adaptive residual scaling normalization term; The coefficient of variation is the residual. As a global adaptive density normalization benchmark; S25. Adaptive Candidate Filtering: Setting Confidence Thresholds With normal threshold ,like and ,but For each candidate point of the retaining wall, a set of candidate retaining wall points is obtained by aggregating all candidate retaining wall points. ; S26. Center Path Construction: Calculate the candidate point set for the retaining wall. 3D mean From the covariance matrix The principal direction of the structure is obtained from the eigenvector corresponding to the largest eigenvalue. For each Calculate the projected scalar ,according to The sequence is obtained by sorting in ascending order. The Ramer–Douglas–Peucker criterion was used to simplify the path and obtain the central path control point. Connect the control points to obtain the central path ; S27, Arc Length Parametric: Calculate Center Path arc length sequence , Total path length Each control point corresponds to an arc length index. This forms a unified structural coordinate system.

4. The method for jointly estimating the height of retaining walls and the slope of road surfaces in a work area facing a spoil heap as described in claim 3, characterized in that, Step S3 is as follows: S31. Valid Candidate Constraints: Valid candidate constraints: For each Calculate its path to the center lateral distance in the XY plane Construct distance set Take quantiles As the width of the corridor Remove The points are used to obtain the set of effective retaining wall candidate points. ; S32, Arc Length Segment Mapping: Sets segment length , central path Divided into There are segments, and the segment index is... For each Calculate its position on the central path Arc length corresponding to the projection point on Mapped to segments The segmented point set is obtained. ; S33, Piecewise Robust Plane Fitting: For each piecewise point set... Perform RANSAC robust plane fitting to obtain the initial plane parameters. With interior point set ; S34. Analytical Refinement: For the set of interior points Constructing the covariance matrix The eigenvector corresponding to the smallest eigenvalue is taken as the refinement normal. If one diligently cultivates the Dharma direction vertical component Then let Ensure that the normal orientation faces the inside of the spoil heap; S35. Wall attitude and height estimation: based on refined normal. Generate wall pose related quantities Wall posture related quantities , The retaining wall height estimate is generated by using the elevation statistics of points within segments. Retaining wall height estimation ,in for mean elevation The average ground elevation of the area surrounding the retaining wall. Spatial correlation weights are introduced, along with a temporal smoothing correction term. ; S36. Uncertainty Measure Construction: An adaptive uncertainty measure is constructed by integrating the global residual mean, in-point residual skewness, and in-point elevation distribution entropy. ,in , The mean of the global residual scale. For interior point residual skewness, The information entropy of the elevation distribution of interior points; S37. Cross-frame confidence weighted fusion: Setting a gate threshold ,like Then the adaptive smoothing coefficient ,otherwise A confidence-weighted adaptive gating fusion formula is used to fuse the current frame estimate with the historical state. The cross-frame fusion formula is as follows: Quantities related to wall posture Using the same fusion strategy, time-smoothed results were obtained. ;in The adaptive smoothing coefficient is modulated by the uncertainty and consistency test results to achieve a balance between stability and responsiveness in time series estimation.

5. The method for jointly estimating the height of retaining walls and the slope of road surfaces in a work area facing a spoil heap as described in claim 4, characterized in that, Step S4 is as follows: S41. Boundary Zone Ground Extraction: Set Neighborhood Radius Set up with points inside the retaining wall To query the set, on the ground point set Perform a K-nearest neighbor search to obtain the set of ground points on the boundary zone of the retaining wall's neighborhood. ; S42. Neighbor exclusion and pollution suppression: Set exclusion radius. Remove the boundary zone ground point set Set of points inside the retaining wall Distance less than exclusion radius The points are used to obtain a pure set of ground points. ; S43. Segment Alignment: Reuse the arc length index of the center path. Set up a clean ground point set The points in the graph are mapped to the corresponding segments to obtain a set of segmented ground points that are consistent with the retaining wall segments. ; S44. Road segment slope estimation: For each segment's ground point set... A robust plane fitting is performed to obtain the road surface normal vector. Road surface segment slope Introducing pollution correction factors With the spatial smoothing compensation term, a robust estimate is obtained. ; S45. Overall road surface slope estimation: A weighted robust global estimation combined with regional reliability weighting is used to generate the overall road surface slope. Overall road surface slope ,in , where represents the credibility weight of each segment; S46. Structured Joint Output: Indexed by Arc Length To unify the primary key, construct a structured output result. This ensures that the retaining wall and road surface indicators correspond one-to-one in the same structural coordinate system; S47. Preservation of traceable evidence: For each segment Preserve the segmented point set Segmented ground point set Point cloud index and planar parameters , Interior point ratio, residual scale <s> wait< / s> Statistics provide a basis for auditing, visualizing, and replaying anomalies in measurement results.

Images