A method and system for 3D modeling of ship cabins for intelligent control of ship unloaders

By using neighborhood curvature, density dispersion, and coefficient of variation to filter point clouds in the 3D modeling of the ship's cabin, and combining different reconstruction algorithms and attitude calibration, the problem of accurate layering in the 3D modeling of the ship's cabin under harsh working conditions was solved, achieving high-precision and real-time 3D modeling of the ship's cabin, and meeting the safety and efficiency requirements of intelligent control of the ship unloader.

CN123118902APending Publication Date: 2026-11-17RIZHAO PORT GRP CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611299742.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing 3D modeling methods for ship cabins have poor anti-interference capabilities, low structural differentiation, and insufficient real-time performance under harsh port unloading conditions, and cannot meet the intelligent control requirements of fully automatic, high-precision, and high-safety ship unloaders.

Method used

By acquiring the initial point cloud of the ship's cabin, the cabin structure and the material stack structure are distinguished by filtering based on neighborhood curvature, density dispersion, and coefficient of variation. Different reconstruction algorithms are used to construct a 3D model, including plane fitting and non-uniform surface interpolation. Combined with attitude calibration technology, accurate layered modeling is achieved.

Benefits of technology

It enables precise layered modeling of ship compartments under harsh working conditions, improving modeling accuracy and real-time performance, ensuring the safety and efficiency of the ship unloader, and meeting the requirements of fully automatic intelligent control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118902A_ABST
    Figure CN123118902A_ABST
Patent Text Reader

Abstract

This invention provides a method and system for three-dimensional modeling of a ship's hold for intelligent control of a ship unloader, belonging to the field of three-dimensional modeling technology. The method includes: acquiring an initial point cloud of the ship's hold; acquiring the neighborhood curvature and neighborhood density of each point in the initial point cloud, and obtaining the density dispersion and coefficient of variation based on the neighborhood density; screening the initial point cloud of the ship's hold based on the neighborhood curvature, density dispersion, and coefficient of variation to obtain a final hold point cloud and a final material point cloud; constructing a three-dimensional model of the hold based on the final hold point cloud; and constructing a three-dimensional model of the material based on the final material point cloud. This invention achieves accurate layered modeling of the rigid structure of the ship's hold and the material pile, ensuring modeling accuracy and real-time performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and in particular to a 3D modeling method and system for the cabin of a ship unloader for intelligent control. Background Technology

[0002] Bulk cargo unloading is a core part of port logistics. With the advancement of port intelligence and unmanned upgrades, fully automated intelligent control of ship unloaders has become the mainstream trend in the industry. The 3D model of the ship's hold is the core foundational data for ship unloader trajectory planning, collision avoidance control, material balance monitoring, and fully automated material handling. The accuracy, real-time performance, and completeness of the model directly determine the operational safety and efficiency of the ship unloader's intelligent control system.

[0003] Currently, the main method for 3D modeling of ship cabins is based on real-time point cloud reconstruction modeling using LiDAR and visual cameras. However, this method has inherent technical shortcomings under the harsh working conditions of unloading ships in ports: First, the metal bulkheads of the ship cabins have strong specular reflections. During operations, dust, uneven lighting, and dense local shadows can easily cause LiDAR point cloud feature matching failures, a surge in point cloud noise, and data loss. This results in numerous holes, blurred edges, and distorted contours in the reconstructed model, leading to low model detail fidelity and an inability to support the millimeter-level intelligent control precision requirements. Second, conventional point cloud modeling algorithms do not distinguish between the rigid structure of the ship cabin (bulls, hatches, and bottom) and the flexible material stacking structure. Unified modeling and processing leads to confusion of structural features, making it difficult to accurately identify the ship cabin boundaries and material stacking areas, and failing to provide effective data support for layered material handling and surplus calculation. Third, the existing modeling process has high data processing redundancy, relying on manual intervention for noise reduction, completion, and segmentation. This results in low automation, high modeling latency, and an inability to meet the real-time control requirements of continuous dynamic operations of ship unloaders.

[0004] In summary, existing 3D modeling methods for ship cabins have poor anti-interference capabilities, low structural differentiation, insufficient real-time performance, and poor adaptability to operating conditions, and cannot meet the requirements of fully automatic, high-precision, and high-safety intelligent control operations for ship unloaders. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for three-dimensional modeling of ship compartments for intelligent control of ship unloaders, so as to achieve accurate layered modeling of the rigid structure of the ship compartment and the material pile, and ensure modeling accuracy and real-time performance.

[0006] A method for three-dimensional modeling of the ship's cabin for intelligent control of a ship unloader, comprising:

[0007] Acquire the initial point cloud of the ship's cabin; the initial point cloud of the ship's cabin includes the cabin point cloud and the material point cloud;

[0008] Obtain the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and obtain the density dispersion and coefficient of variation based on the neighborhood density;

[0009] Points in the initial point cloud of the cabin whose neighborhood curvature is less than a set curvature threshold are taken as the first cabin point cloud, and points in the initial point cloud of the cabin whose neighborhood curvature is greater than or equal to the set curvature threshold are taken as the first material point cloud.

[0010] Points in the first cabin point cloud whose density dispersion is less than the dynamic dispersion threshold and whose coefficient of variation is less than the set variation threshold are designated as the second cabin point cloud; points in the first cabin point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold and / or whose coefficient of variation is greater than or equal to the set variation threshold are designated as the second material point cloud.

[0011] Points in the first material point cloud whose density dispersion is less than the dynamic dispersion threshold are used as the third cabin point cloud, and points in the first material point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold are used as the third material point cloud.

[0012] The second cabin point cloud and the third cabin point cloud are merged to obtain the final cabin point cloud; the second material point cloud and the third material point cloud are merged to obtain the final material point cloud;

[0013] A 3D model of the cabin is constructed based on the final cabin point cloud; a 3D model of the materials is constructed based on the final material point cloud.

[0014] Optionally, the neighborhood curvature acquisition process is as follows:

[0015] For point ,by With the center as the circle, draw a circle with a first set radius, and take all points inside the circle as the first neighborhood point set;

[0016] The first neighborhood point set is fitted with a tangent plane using the least squares method to obtain the fitted plane.

[0017] The gradient of the change in the normal vector of the fitted plane is used as... The curvature of the domain.

[0018] Optionally, the process for obtaining the domain density is as follows:

[0019] For point ,by With the center of the sphere as the center, construct a sphere with a second set radius, and take all points inside the sphere as the second neighborhood point set;

[0020] Based on the number of points in the second neighborhood point set The neighborhood density; the expression is:

[0021] ;

[0022] in, for Domain density, for Distance to lidar, for The number of points in the corresponding second neighborhood point set Let be the volume of the sphere.

[0023] Optionally, the density dispersion acquisition process is as follows:

[0024] For point The standard deviation of the neighborhood density of all points within the corresponding sphere is taken as the point. Density dispersion.

[0025] Optionally, the coefficient of variation is expressed as follows:

[0026] ;

[0027] in, For point coefficient of variation, For point density dispersion, For point The mean of the neighborhood density of all points within the corresponding sphere.

[0028] Optionally, the dynamic dispersion threshold is the dispersion threshold corresponding to the current point.

[0029] Optionally, the dispersion threshold of the point corresponding to the reference distance can be used as the reference dispersion threshold;

[0030] Based on the benchmark dispersion threshold and points The distance to the lidar is obtained as a point The corresponding dynamic dispersion threshold.

[0031] Optionally, the expression for the dynamic discreteness threshold is:

[0032] ;

[0033] in, For point The dynamic dispersion threshold, For point Distance to lidar, As the reference distance, As the baseline dispersion threshold, These are the scanning attitude correction coefficients. This is the dust interference correction factor.

[0034] Optionally, the dust disturbance correction factor is negatively correlated with the dust concentration;

[0035] The centerline of the lidar and the point The angle between the planes in which they lie is taken as the scanning incident angle;

[0036] The scanning posture correction coefficient is positively correlated with the scanning incident angle.

[0037] The present invention also provides a three-dimensional modeling system for the cabin of a ship unloader for intelligent control, comprising:

[0038] The point cloud acquisition module is used to acquire the initial point cloud of the ship's cabin; the initial point cloud of the ship's cabin includes the cabin point cloud and the material point cloud;

[0039] The parameter acquisition module is used to acquire the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and to obtain the density dispersion and coefficient of variation based on the neighborhood density.

[0040] The first filtering module is used to select points in the initial point cloud of the cabin whose neighborhood curvature is less than a set curvature threshold as the first cabin point cloud, and select points in the initial point cloud of the cabin whose neighborhood curvature is greater than or equal to the set curvature threshold as the first material point cloud.

[0041] The second filtering module is used to select points in the first cabin point cloud whose density dispersion is less than the dynamic dispersion threshold and whose coefficient of variation is less than the set variation threshold as the second cabin point cloud; and to select points in the first cabin point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold and / or whose coefficient of variation is greater than or equal to the set variation threshold as the second material point cloud.

[0042] The third filtering module is used to select points in the first material point cloud whose density dispersion is less than the dynamic dispersion threshold as the third cabin point cloud, and to select points in the first material point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold as the third material point cloud.

[0043] The point cloud merging module is used to merge the second cabin point cloud and the third cabin point cloud to obtain the final cabin point cloud; and to merge the second material point cloud and the third material point cloud to obtain the final material point cloud;

[0044] The 3D modeling module is used to construct a 3D model of the cabin based on the final cabin point cloud; and to construct a 3D model of the material based on the final material point cloud.

[0045] The effects of this invention are as follows:

[0046] This invention provides a 3D modeling method for the ship compartment of an intelligent control system for ship unloaders. By filtering based on curvature, density dispersion, and coefficient of variation, the method accurately distinguishes between the compartment structure and the material pile structure. This ensures both the regularity and accuracy of the basic structure of the ship compartment and the complete preservation of the irregular and real shape of the material pile. This provides accurate data support for the ship unloader to distinguish the working boundaries, plan the layered material picking trajectory, and calculate the remaining material. Attached Figure Description

[0047] Figure 1 This is a flowchart of the three-dimensional modeling method for the ship compartment used in the intelligent control of the ship unloader according to the present invention. Detailed Implementation

[0048] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart of the 3D modeling method for the ship compartment used in the intelligent control of a ship unloader according to the present invention, as shown below. Figure 1 As shown, this invention provides a method for three-dimensional modeling of the ship's cabin for intelligent control of a ship unloader, comprising:

[0050] S1, Obtain the initial point cloud of the ship's compartment. The initial point cloud of the ship's compartment includes the hull point cloud and the material point cloud.

[0051] Specifically, the original point cloud of the ship's cabin is obtained by deploying multi-line lidar. Before the operation, the internal and external parameters of the lidar are calibrated to unify the three-dimensional coordinate system. The original lidar point cloud data of the entire ship's cabin is collected in real time, covering the entire area of ​​the hatch, bulkhead, bottom and internal material stack.

[0052] The original point cloud of the ship cabin was subjected to noise reduction filtering to obtain the initial point cloud of the ship cabin.

[0053] S2, obtain the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and obtain the density dispersion and coefficient of variation based on the neighborhood density.

[0054] Neighborhood curvature and local curvature are used to characterize the surface smoothness of the area where the measurement point is located.

[0055] Preferably, the neighborhood curvature acquisition process is as follows:

[0056] For point ,by Using the first set radius as the center, draw a circle and take all points inside the circle as the first neighborhood point set.

[0057] The first neighborhood point set is fitted with a tangent plane using the least squares method to obtain the fitted plane.

[0058] The gradient of the change in the normal vector of the fitted plane is used as... The curvature of the domain.

[0059] Neighborhood density is used to characterize the uniformity and density stability of the point cloud distribution around a measurement point. The process of obtaining neighborhood density is as follows:

[0060] For point ,by Using a sphere centered at a certain radius, a second sphere is constructed, and all points within this sphere are designated as the second neighborhood set. A fixed-radius neighborhood ensures a uniform statistical spatial scale for each measurement point, avoiding density comparison failures caused by axial deviations or scale inconsistencies within a rectangular neighborhood.

[0061] Based on the number of points in the second neighborhood point set The density of the domain. The expression is:

[0062] ;

[0063] in, for Domain density, for Distance to lidar, for The number of points in the corresponding second neighborhood point set Let be the volume of the sphere.

[0064] LiDAR scanning point clouds have inherent characteristics: dense point clouds at close range and sparse point clouds at long range. The original number of points on the same structural surface varies greatly under different measurement distances, making it unsuitable for direct structural identification. Therefore, this invention adopts a distance-normalized neighborhood volume density calculation method to uniformly convert the original number of points under different measurement distances and scanning angles into the equivalent point cloud density under standard measurement distances. This makes the density values ​​of all measurement points in the entire field comparable laterally, completely eliminating the calculation distortion problems caused by the inconsistency in density between close and long range scanning and the deviation in scanning angle, and accurately reflecting the true dense characteristics of the structure.

[0065] The process of obtaining density dispersion is as follows:

[0066] For point The standard deviation of the neighborhood density of all points within the corresponding sphere is taken as the point. Density dispersion.

[0067] The expression for the coefficient of variation is as follows:

[0068] ;

[0069] in, For point coefficient of variation, For point density dispersion, For point The mean of the neighborhood density of all points within the corresponding sphere.

[0070] The neighborhood density standard deviation of the rigid structural surface of the cabin is very small, and the density distribution is stable; the neighborhood density standard deviation of the uneven surface of the flexible material is large, and the density jump is obvious.

[0071] S3, points in the initial point cloud of the cabin with a neighborhood curvature less than a set curvature threshold are taken as the first cabin point cloud, and points in the initial point cloud of the cabin with a neighborhood curvature greater than or equal to the set curvature threshold are taken as the first material point cloud.

[0072] For points in the initial point cloud of the ship's cabin whose neighborhood curvature is less than a set curvature threshold, the point cloud density distribution in that area is determined to be stable, uniform, and continuous, consistent with the point cloud distribution characteristics of rigid, flat surfaces such as steel bulkheads, hulls, and hatches. Rigid structural surfaces are physically continuous and free of random irregularities; variations in laser scanning sampling density are solely caused by the equipment's scanning characteristics, without structural abrupt changes in density. Therefore, the normalized density of multiple points in the neighborhood is highly similar with minimal fluctuations.

[0073] For points in the initial point cloud of the ship's cabin whose neighborhood curvature is greater than or equal to a set curvature threshold, the point cloud density distribution in this area is determined to be disordered and exhibits abrupt changes in density, consistent with the point cloud distribution characteristics of bulk material accumulation. The material surface has randomly distributed grooves, protrusions, and slopes. Laser scanning suffers from local occlusion, uneven reflection intensity, and random missing sampling points, resulting in drastic fluctuations in the normalized density within the neighborhood, with relative volatility significantly greater than in rigid structure regions.

[0074] S4, points in the first cabin point cloud with density dispersion less than the dynamic dispersion threshold and coefficient of variation less than the set variation threshold are designated as the second cabin point cloud; points in the first cabin point cloud with density dispersion greater than or equal to the dynamic dispersion threshold and / or coefficient of variation greater than or equal to the set variation threshold are designated as the second material point cloud.

[0075] Specifically, the dynamic dispersion threshold is the dispersion threshold corresponding to the current point.

[0076] Use the dispersion threshold of the point corresponding to the benchmark distance as the benchmark dispersion threshold.

[0077] Based on the benchmark dispersion threshold and points The distance to the lidar is obtained as a point The corresponding dynamic dispersion threshold.

[0078] The expression for the dynamic dispersion threshold is:

[0079] ;

[0080] in, For point The dynamic dispersion threshold, For point Distance to lidar, As the reference distance, As the baseline dispersion threshold, These are the scanning attitude correction coefficients. This is the dust interference correction factor.

[0081] This invention employs an adaptive mechanism for the dispersion threshold: under long-distance, large-angle scanning conditions, the tolerance range of the dispersion threshold is appropriately widened to avoid sparse areas at the edges of rigid structures being misjudged as unstable; under close-range clean conditions, the threshold is tightened to improve the detection capability of unstable features in small uneven areas of materials, ensuring that density stability determination is accurate and effective under all conditions.

[0082] Preferably, the dust interference correction coefficient is negatively correlated with the dust concentration.

[0083] The centerline of the lidar and the point The angle between the planes is taken as the scanning incident angle. The scanning attitude correction coefficient is positively correlated with the scanning incident angle.

[0084] S5, points in the first material point cloud with density dispersion less than the dynamic dispersion threshold are used as the third cabin point cloud, and points in the first material point cloud with density dispersion greater than or equal to the dynamic dispersion threshold are used as the third material point cloud.

[0085] S6, merge the second cabin point cloud and the third cabin point cloud to obtain the final cabin point cloud; merge the second material point cloud and the third material point cloud to obtain the final material point cloud.

[0086] S7: Construct a 3D model of the cabin based on the final cabin point cloud; construct a 3D model of the materials based on the final material point cloud.

[0087] Specifically, for the two independent point cloud datasets after segmentation, reconstruction algorithms adapted to their respective structural features are used for separate modeling. For the final hull point cloud, a planar fitting + mesh normalization reconstruction algorithm is used. Relying on the regular structural characteristics of the bulkhead, hull bottom, and hatch, the missing fine regular contours of the point cloud are normalized and fitted to construct an initial hull 3D model with complete topological structure and dimensional accuracy that closely matches the actual ship hull. This ensures the modeling accuracy of the hull operation boundary, hull size, and hatch contour, providing precise rigid structural constraints for subsequent collision avoidance control of the unloader. For the final material point cloud, the regular structure fitting method is abandoned, and a non-uniform surface interpolation reconstruction algorithm is used. This fully preserves the real details of the material accumulation, such as random concave and convex shapes, local steep slopes, and accumulation grooves, without manual normalization correction. It completely restores the real-time material pile shape, pile height, and local surplus distribution, providing a realistic and effective initial material 3D model for the unloader's layered material handling planning, accurate calculation of material surplus, and residual material cleaning path planning. Through layered independent modeling, differentiated modeling effects are achieved, ensuring the accuracy of the hull structure and the fidelity of the material structure.

[0088] To address the defects of holes and missing edges in the initial 3D model of the hull, a structural constraint interpolation algorithm was used, combined with prior knowledge of the general structure of ship hulls, to complete the regular structure of the hull walls and bottom, thus obtaining the 3D model of the hull. To address the irregular contour defects in the initial 3D model of the material, a neighborhood feature fitting algorithm was used to optimize the edge details of the material pile, preserving the true concave and convex shape of the material accumulation. At the same time, based on the ship attitude detection data, the model coordinate deviations caused by ship yaw, settlement, and tilt were corrected to complete the model attitude calibration, thus obtaining the 3D model of the material.

[0089] The method of the present invention will be further described in detail below with reference to specific engineering embodiments. This embodiment is based entirely on the actual unloading operation conditions in the port, and all equipment parameters, algorithm thresholds, operating conditions and test results are data that can be implemented in actual engineering.

[0090] I. Example Test Conditions and Equipment Parameters

[0091] This embodiment is applied to a fully automated operation scenario of a conventional bridge grab unloader in a port. The object of operation is the hold of a bulk carrier, where coal and bulk materials are piled up. The operating environment is a typical and complex port condition (including dust, uneven natural light, shadows inside the hold, slight tilt of the ship, and non-right berthing). The specific equipment configuration is as follows: two 16-line LiDARs are deployed, with a horizontal angular resolution of 0.2° and a vertical angular resolution of 0.4°, and a maximum effective range of 100m; an attitude sensor is equipped to collect ship attitude data in real time; the controller uses an industrial control computer to ensure that the algorithm iterates every 100ms.

[0092] II. Global Fixed Algorithm Threshold (Engineering Calibration Value)

[0093] After multiple tests and calibrations on multiple ships, this embodiment adopts a steady-state fixed threshold: a local curvature fixed judgment threshold of 0.05; a spherical neighborhood search radius of r=0.3m; a variation setting threshold of 0.12; a baseline dispersion threshold of 0.08; and a baseline distance of 10m.

[0094] III. Specific Implementation Steps

[0095] Equipment calibration and spatiotemporal synchronization calibration. Before operation, the internal and external parameters of the lidar are calibrated, and the coordinate system of all sensor data is unified to the world control coordinate system of the ship unloader. After calibration, the ship unloader is controlled to move and perform multi-view full-coverage scanning of the target ship's hold, collecting the original point cloud of the hatch, bulkhead, bottom and internal coal pile.

[0096] Multidimensional feature quantization calculation of point cloud. The features of the effective point cloud after noise reduction are solved point by point to obtain the neighborhood curvature, neighborhood density, density dispersion and coefficient of variation of each point.

[0097] Dynamic threshold calculation and segmentation in real time. Dynamic judgment thresholds are generated in real time based on the parameters of this operating condition: distance attenuation compensation is completed according to the distance D of the measuring point; there is a slight tilt of the radar in this scan, so the scanning attitude correction coefficient is set to 0.92; the dust interference correction coefficient is 0.85; the final cabin point cloud and the final material point cloud are finally synthesized by filtering according to each characteristic parameter and the set threshold.

[0098] Independent modeling and refined optimization of each structure were performed. For the final segmented cabin point cloud, a planar fitting + mesh normalization reconstruction algorithm was used, combined with prior knowledge of the rectangular structure of the cabin, to fill in missing local holes and edge gaps with structural constraints, restoring accurate physical boundaries, hatch dimensions, and bottom contours of the cabin. For the final material point cloud, a non-uniform surface interpolation algorithm was used to completely preserve the details of the grooves, steep slopes, and undulations of the randomly accumulated materials without manual normalization correction, realistically restoring the real-time stacking shape. Simultaneously, attitude data of the ship at 0.8° heel, 0.5° trim, and slight settlement were collected by attitude sensors, and the model coordinates were corrected point by point to correct model deviations caused by hull deformation and non-righting berthing.

[0099] Data is continuously collected and the model is iteratively updated every 100ms, and the dynamic changes of material grabbing, pile collapse and residual reduction during the operation are synchronized in real time.

[0100] The experimental results of this embodiment:

[0101] 1. Significantly improved anti-interference capability: Under moderate dust, shadow, and metal reflection conditions, the model porosity and edge distortion rate are reduced by more than 92% compared with traditional point cloud modeling methods, and there is no structural distortion problem.

[0102] 2. High structural segmentation accuracy: The segmentation accuracy of rigid structures and flexible materials reaches 98.6%, with no boundary overlap or missegmentation.

[0103] 3. Real-time modeling meets standards: The single-frame modeling and update cycle is stably controlled within 100ms, with no delay or stuttering, making it suitable for continuous high-speed operation of ship unloaders.

[0104] 4. Improved safety and efficiency through intelligent control: Relying on a precise dynamic model, the trajectory planning deviation of the unloader is controlled within ±3mm, eliminating the risk of collision and increasing the material removal rate by 6%, fully meeting the requirements of fully automated intelligent unloading operations in ports.

[0105] The present invention also provides a three-dimensional modeling system for the cabin of a ship unloader for intelligent control, comprising:

[0106] The point cloud acquisition module is used to acquire the initial point cloud of the ship's cabin. The initial point cloud of the ship's cabin includes the cabin point cloud and the material point cloud.

[0107] The parameter acquisition module is used to obtain the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and to obtain the density dispersion and coefficient of variation based on the neighborhood density.

[0108] The first filtering module is used to select points in the initial point cloud of the ship cabin whose neighborhood curvature is less than a set curvature threshold as the first cabin point cloud, and select points in the initial point cloud of the ship cabin whose neighborhood curvature is greater than or equal to the set curvature threshold as the first material point cloud.

[0109] The second filtering module is used to select points in the first cabin point cloud whose density dispersion is less than the dynamic dispersion threshold and whose coefficient of variation is less than the set variation threshold as the second cabin point cloud; and to select points in the first cabin point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold and / or whose coefficient of variation is greater than or equal to the set variation threshold as the second material point cloud.

[0110] The third filtering module is used to select points in the first material point cloud whose density dispersion is less than the dynamic dispersion threshold as the third cabin point cloud, and to select points in the first material point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold as the third material point cloud.

[0111] The point cloud merging module is used to merge the second cabin point cloud and the third cabin point cloud to obtain the final cabin point cloud; and to merge the second material point cloud and the third material point cloud to obtain the final material point cloud.

[0112] The 3D modeling module is used to construct a 3D model of the cabin based on the final cabin point cloud; and to construct a 3D model of the materials based on the final material point cloud.

[0113] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for three-dimensional modeling of the ship's cabin for intelligent control of a ship unloader, characterized in that, It includes: Obtain the initial point cloud of the ship's cabin; The initial point cloud of the ship's compartment includes the hull point cloud and the material point cloud; Obtain the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and obtain the density dispersion and coefficient of variation based on the neighborhood density; Points in the initial point cloud of the cabin whose neighborhood curvature is less than a set curvature threshold are taken as the first cabin point cloud, and points in the initial point cloud of the cabin whose neighborhood curvature is greater than or equal to the set curvature threshold are taken as the first material point cloud. Points in the first cabin point cloud whose density dispersion is less than the dynamic dispersion threshold and whose coefficient of variation is less than the set variation threshold are designated as the second cabin point cloud; points in the first cabin point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold and / or whose coefficient of variation is greater than or equal to the set variation threshold are designated as the second material point cloud. Points in the first material point cloud whose density dispersion is less than the dynamic dispersion threshold are used as the third cabin point cloud, and points in the first material point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold are used as the third material point cloud. The second cabin point cloud and the third cabin point cloud are merged to obtain the final cabin point cloud; The second material point cloud and the third material point cloud are merged to obtain the final material point cloud; A 3D model of the cabin is constructed based on the final cabin point cloud; a 3D model of the materials is constructed based on the final material point cloud.

2. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 1, characterized in that, The process for obtaining the neighborhood curvature is as follows: For point ,by With the center as the circle, draw a circle with a first set radius, and take all points inside the circle as the first neighborhood point set; The first neighborhood point set is fitted with a tangent plane using the least squares method to obtain the fitted plane. The gradient of the change in the normal vector of the fitted plane is used as... The curvature of the domain.

3. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 1, characterized in that, The process for obtaining the domain density is as follows: For point ,by With the center of the sphere as the center, construct a sphere with a second set radius, and take all points inside the sphere as the second neighborhood point set; Based on the number of points in the second neighborhood point set The domain density; the expression is: ; in, for Domain density, for Distance to lidar, for The number of points in the corresponding second neighborhood point set Let be the volume of the sphere.

4. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 3, characterized in that, The density dispersion acquisition process is as follows: For point The standard deviation of the neighborhood density of all points within the corresponding sphere is taken as the point. Density dispersion.

5. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 4, characterized in that, The expression for the coefficient of variation is as follows: ; in, For point coefficient of variation, For point density dispersion, For point The mean of the neighborhood density of all points within the corresponding sphere.

6. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 1, characterized in that, The dynamic dispersion threshold is the dispersion threshold corresponding to the current point.

7. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 6, characterized in that, Use the dispersion threshold of the point corresponding to the benchmark distance as the benchmark dispersion threshold; Based on the benchmark dispersion threshold and points The distance to the lidar is obtained as a point The corresponding dynamic dispersion threshold.

8. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 7, characterized in that, The expression for the dynamic discreteness threshold is: ; in, For point The dynamic dispersion threshold, For point Distance to lidar, As the reference distance, As the baseline dispersion threshold, These are the scanning attitude correction coefficients. This is the dust interference correction factor.

9. The method for three-dimensional modeling of the ship's hold for intelligent control of a ship unloader according to claim 8, characterized in that, The dust disturbance correction factor is negatively correlated with dust concentration; The centerline of the lidar and the point The angle between the planes in which they lie is taken as the scanning incident angle; The scanning posture correction coefficient is positively correlated with the scanning incident angle.

10. A three-dimensional modeling system for the cabin of a ship unloader for intelligent control, characterized in that, It includes: The point cloud acquisition module is used to acquire the initial point cloud of the ship's cabin. The initial point cloud of the ship's compartment includes the hull point cloud and the material point cloud; The parameter acquisition module is used to acquire the neighborhood curvature and neighborhood density of each point in the initial point cloud of the cabin, and to obtain the density dispersion and coefficient of variation based on the neighborhood density. The first filtering module is used to select points in the initial point cloud of the cabin whose neighborhood curvature is less than a set curvature threshold as the first cabin point cloud, and select points in the initial point cloud of the cabin whose neighborhood curvature is greater than or equal to the set curvature threshold as the first material point cloud. The second filtering module is used to select points in the first cabin point cloud whose density dispersion is less than the dynamic dispersion threshold and whose coefficient of variation is less than the set threshold of variation as the second cabin point cloud. Points in the first cabin point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold and / or whose coefficient of variation is greater than or equal to the set variation threshold are used as the second material point cloud. The third filtering module is used to select points in the first material point cloud whose density dispersion is less than the dynamic dispersion threshold as the third cabin point cloud, and select points in the first material point cloud whose density dispersion is greater than or equal to the dynamic dispersion threshold as the third material point cloud. The point cloud merging module is used to merge the second cabin point cloud and the third cabin point cloud to obtain the final cabin point cloud; The second material point cloud and the third material point cloud are merged to obtain the final material point cloud; The 3D modeling module is used to construct a 3D model of the cabin based on the final cabin point cloud; and to construct a 3D model of the material based on the final material point cloud.