Method and system for scanning position of grab bucket for ship unloader

Abstract

The present application relates to the field of crane technology, and discloses a method and system for scanning the position of a grab bucket dedicated to a ship unloader. The method comprises: obtaining the original three-dimensional point cloud data set of the cabin by laser scanning, and forming a cabin structure and cargo surface feature data set based on feature extraction and spatial clustering. The feature data set is gridded to establish a cabin operation area model; the accessibility parameters of each grid unit are calculated to establish a grid unit grab accessibility database. A multi-objective optimization algorithm is used to generate a grid grab operation parameter set based on the accessibility database. Based on the cabin operation area model, the grid unit accessibility database and the grab operation parameter set, strategy training is implemented through deep reinforcement learning to output a dynamic obstacle avoidance grab instruction sequence, including horizontal trajectory compensation, vertical trajectory compensation and grab bucket posture compensation parameters. The present application realizes high-precision scanning and intelligent path planning of the grab bucket position of the ship unloader to improve the grab bucket positioning accuracy and operation safety.

Description

Technical Field

[0001] The present application relates to the technical field of cranes, and in particular to a method and system for scanning the position of a grab bucket dedicated to a ship unloader. Background Art

[0002] At present, the ship unloaders at port terminals generally use grab-type ship unloaders for bulk cargo unloading operations. During the operation, the operator mainly controls the grab to grab the cargo. The traditional grab positioning usually relies on manual operation experience and visual judgment. Some high-end equipment uses a simple auxiliary positioning system based on camera or laser ranging. During the movement of the grab, the movement trajectory of the grab is controlled by manual combined with simple automation devices to achieve cargo grabbing and unloading. At the same time, some advanced ports have also begun to try to use computer vision technology and 3D modeling technology to assist the positioning and path planning of the grab.

[0003] However, the relevant technologies have the following shortcomings: the traditional manual operation method relies heavily on the operator's experience, and it is difficult to ensure the operation efficiency and accuracy; the simple auxiliary positioning system cannot accurately obtain the complex spatial information in the cabin, resulting in low grab positioning accuracy; the existing automatic control system lacks the real-time perception capability of the cabin environment and cannot effectively respond to dynamic changes during the operation process; the existing computer vision technology has poor reliability under complex lighting and weather conditions, and has blind spots, making it difficult to obtain complete spatial information; in addition, the existing technologies generally lack intelligent path planning and obstacle avoidance capabilities, and collision accidents are prone to occur during the operation process, posing a major safety hazard. Summary of the invention

[0004] The present application provides a method and system for scanning the position of a grab bucket dedicated to a ship unloader, which are used to achieve high-precision scanning and intelligent path planning of the grab bucket position of the ship unloader, so as to improve the positioning accuracy of the grab bucket and the safety of operation.

[0005] In a first aspect, the present application provides a method for scanning the position of a grab bucket dedicated to a ship unloader, the method comprising: measuring the cabin space by laser scanning to obtain an original three-dimensional point cloud data set of the cabin; processing the original three-dimensional point cloud data set of the cabin by feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set; gridding the cabin structure feature data set and the cargo surface feature data set to establish a grid model of the cabin operation area, wherein each grid unit includes the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters; calculating the accessibility parameters for each grid unit in the grid model of the cabin operation area, and constructing a grid unit grab accessibility data set. The grid unit grasping accessibility database includes the entry angle, rope length and horizontal position of the trolley of each grid unit; a multi-objective optimization algorithm is used to calculate the motion trajectory parameters based on the grid unit grasping accessibility database to generate a grid grasping operation parameter set, and the grid grasping operation parameter set includes the trolley acceleration, deceleration, uniform speed and rope retraction and release speed; based on the grid model of the cabin operating area, the grid unit grasping accessibility database and the grid grasping operation parameter set, a deep reinforcement learning algorithm is used to perform strategy training, and a dynamic obstacle avoidance grasping instruction sequence is output, and the dynamic obstacle avoidance grasping instruction sequence includes timing-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab bucket posture compensation parameters.

[0006] In a second aspect, the present application provides a grab position scanning system for a ship unloader, the grab position scanning system for a ship unloader comprising:

[0007] A scanning module is used to measure the cabin space through laser scanning and obtain the original three-dimensional point cloud data set of the cabin;

[0008] A clustering module, used for processing the original three-dimensional point cloud data set of the cabin through feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set;

[0009] A processing module, used for gridding the cabin structure feature data set and the cargo surface feature data set to establish a grid model of the cabin operation area, wherein each grid unit includes the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters;

[0010] A calculation module, for calculating accessibility parameters for each grid unit in the grid model of the cabin operation area, and constructing a grid unit grabbing accessibility database, wherein the grid unit grabbing accessibility database includes an entry angle, a rope length, and a horizontal position of a trolley for each grid unit;

[0011] An analysis module, for calculating motion trajectory parameters based on the grid unit grabbing accessibility database using a multi-objective optimization algorithm, and generating a grid grabbing operation parameter set, wherein the grid grabbing operation parameter set includes trolley acceleration, deceleration, uniform speed, and rope retraction and release speed;

[0012] A decision module is used to perform strategy training through a deep reinforcement learning algorithm based on the grid model of the cabin operation area, the grid unit grasping reachability database and the grid grasping operation parameter set, and output a dynamic obstacle avoidance grasping instruction sequence, wherein the dynamic obstacle avoidance grasping instruction sequence includes a timing-based horizontal motion trajectory compensation parameter, a vertical motion trajectory compensation parameter and a grab posture compensation parameter.

[0013] In the technical solution provided by the present application, by performing laser scanning processing on the cabin space, a high-precision original three-dimensional point cloud data set of the cabin can be obtained, which provides a reliable data basis for subsequent feature extraction and spatial analysis, and avoids the recognition errors of traditional visual systems in complex environments; by performing feature extraction and spatial clustering processing on the original three-dimensional point cloud data set, a cabin structure feature data set and a cargo surface feature data set are obtained respectively, which realizes the effective separation of cabin space information and improves the recognition accuracy of spatial features; the cabin structure feature data set and the cargo surface feature data set are gridded to obtain a grid model of the cabin operation area including the grid center point coordinates, average height value, highest point and lowest point parameters, so that the complex spatial information is expressed in a standardized manner, which is convenient for subsequent path planning and control; by performing accessibility calculation processing on each grid unit in the grid model of the cabin operation area, the grid unit grasping accessibility including the entry angle, rope length and horizontal position of the trolley is obtained. The database provides comprehensive spatial constraints for the grab movement and ensures the feasibility of the motion trajectory. Based on the grid unit grab accessibility database, a multi-objective optimization algorithm is used to calculate the parameters of the motion trajectory, and a grid grab operation parameter set including the trolley acceleration, deceleration, uniform speed and rope retraction and release speed is obtained, which realizes the optimal design of the motion trajectory and improves the stability and efficiency of the motion process. Finally, the deep reinforcement learning algorithm is used to perform strategy training on the grid model of the cabin operation area, the grid unit grab accessibility database and the grid grab operation parameter set, and a dynamic obstacle avoidance grab instruction sequence including time-series-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters is obtained, which realizes the intelligent control and dynamic obstacle avoidance of the grab movement process, significantly improves the safety and reliability of the operation process, and at the same time, through the real-time adjustment of various compensation parameters, the stability and accuracy of the grab process are guaranteed, which greatly improves the automation level and operation efficiency of the ship unloading operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of a method for scanning the position of a grab bucket dedicated to a ship unloader in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a grab position scanning system for a ship unloader in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide a method and system for scanning the position of a grab bucket dedicated to a ship unloader. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for scanning the position of the grab bucket dedicated to the ship unloader includes:

[0019] Step S101, measuring the cabin space by laser scanning to obtain an original three-dimensional point cloud data set of the cabin;

[0020] Step S102, processing the original three-dimensional point cloud data set of the cabin by feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set;

[0021] Step S103, gridding the cabin structure feature data set and the cargo surface feature data set to establish a grid model of the cabin operation area, wherein each grid unit includes the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters;

[0022] Step S104: Calculate accessibility parameters for each grid unit in the grid model of the cabin operation area, and construct a grid unit grabbing accessibility database, where the grid unit grabbing accessibility database includes an entry angle, a rope length, and a horizontal position of a trolley for each grid unit;

[0023] Step S105, using a multi-objective optimization algorithm to calculate motion trajectory parameters based on a grid unit grabbing accessibility database, and generating a grid grabbing operation parameter set, the grid grabbing operation parameter set including trolley acceleration, deceleration, uniform speed, and rope retraction and release speed;

[0024] Step S106: Based on the grid model of the cabin operation area, the grid unit grasping accessibility database and the grid grasping operation parameter set, strategy training is performed through a deep reinforcement learning algorithm to output a dynamic obstacle avoidance grasping instruction sequence, which includes timing-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab bucket posture compensation parameters.

[0025] It is understandable that the execution subject of the present application may be a grab position scanning system for ship unloaders, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject.

[0026] Specifically, a laser scanner is used to collect all-round data of the cabin space. The laser scanner is installed on the trolley of the ship unloader and scans the cabin space in segments as the trolley moves. The scanner emits a laser beam and receives the reflected signal to record the spatial coordinate information of each sampling point. After time synchronization processing of the scanned data, multiple scanned data segments are converted into spatial coordinates and merged to eliminate the influence of environmental interference factors such as dust and temperature on the measurement accuracy. The outliers are removed by the RANSAC algorithm, and then down-sampling is performed by voxel filtering to generate the original three-dimensional point cloud data set of the cabin. After obtaining the original three-dimensional point cloud data, the data is subjected to feature extraction and spatial clustering processing. First, the point cloud data is surface fitted to identify structural feature points such as cabin boundaries and bulkheads. These feature points are grouped by connected domain analysis, and shape descriptors are extracted for structural feature classification. For the surface features of the cargo, the height gradient of the point cloud data is calculated, the density distribution characteristics of the cargo accumulation are analyzed, and the contour features of the cargo accumulation are extracted. Through these processes, the cabin structure feature data set and the cargo surface feature data set are obtained.

[0027] Based on the cabin structure and cargo surface feature data, grid processing is performed. First, the cabin boundary coordinates are calculated, the operation area is divided into basic grid units, and each grid unit is assigned a unique spatial coordinate. The cargo surface feature data in the grid unit is spatially mapped, the height data in each grid unit is counted, the key feature points are extracted, and the characteristic parameters of the grid unit are calculated, including the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters. The spatial relationship between the grid units is established to generate a complete grid model of the cabin operation area. The accessibility analysis is performed on each grid unit, the spatial position constraints of the grab to reach the grid unit are analyzed, the kinematic parameters of the grab are calculated, collision detection is performed, and a safe approach path is determined. The optimal angle for the grab to enter each grid unit is calculated, the force on the rope is analyzed, the required rope length is determined, and the horizontal position of the trolley is planned. These parameters are integrated into the grid unit grab accessibility database.

[0028] Based on the accessibility database, the motion trajectory parameters are calculated through a multi-objective optimization algorithm. The motion space constraints of the trolley are analyzed, the acceleration and deceleration parameters that meet the stability requirements are calculated, and the uniform speed of the stable operation is determined. The dynamic force characteristics of the rope are analyzed to calculate the appropriate retraction and release speed. The trolley motion parameters and rope parameters are dynamically coupled and the motion constraints are comprehensively considered to generate a grid grab operation parameter set. Finally, a deep reinforcement learning algorithm is used to train the dynamic obstacle avoidance strategy. The parameters of the grid model are constructed as the environment state space, the operation parameter set is constructed as the action space, and the reward function is designed to evaluate the effect of each action. The basic control strategy is obtained through reinforcement learning training, the trajectory compensation parameters of horizontal and vertical motion are calculated, the timing correlation is analyzed, and the grab posture compensation parameters are calculated. These compensation parameters are integrated to generate a dynamic obstacle avoidance grab instruction sequence.

[0029] For example, in a certain ship unloader grabbing operation, the laser scanner scans the cabin and obtains 1 million original sampling points. After downsampling and filtering, 200,000 valid point cloud data are retained. Cluster analysis of the point cloud data identifies 12 key structural feature areas and 8 cargo accumulation areas. The working space is divided into 10×10 grid cells, each of which is 1m×1m in size. The grid cell numbered (5, 5) is analyzed, and its center point coordinates are (5.5m, 5.5m), the average height is 3.2m, the highest point is 3.8m, and the lowest point is 2.6m. The accessibility analysis shows that the optimal entry angle of this grid cell is 75 degrees, the required rope length is 8.5m, and the horizontal position of the trolley is 6.2m. The motion parameter optimization results show that the trolley acceleration is 0.3m / s², the deceleration is 0.25m / s², the uniform speed is 1.2m / s, and the rope retraction speed is 0.8m / s. After deep reinforcement learning training, the horizontal trajectory compensation parameters are ±0.2m, the vertical trajectory compensation parameters are ±0.15m, and the grab posture compensation parameters are ±5 degrees.

[0030] In the embodiment of the present application, by performing laser scanning processing on the cabin space, a high-precision original three-dimensional point cloud data set of the cabin can be obtained, which provides a reliable data basis for subsequent feature extraction and spatial analysis, and avoids the recognition errors of traditional visual systems in complex environments; by performing feature extraction and spatial clustering processing on the original three-dimensional point cloud data set, a cabin structure feature data set and a cargo surface feature data set are obtained respectively, which realizes the effective separation of cabin space information and improves the recognition accuracy of spatial features; the cabin structure feature data set and the cargo surface feature data set are gridded to obtain a grid model of the cabin operation area including the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters, so that the complex spatial information is expressed in a standardized manner, which is convenient for subsequent path planning and control; by performing accessibility calculation processing on each grid unit in the grid model of the cabin operation area, the grid unit grab accessibility data including the entry angle, the rope length and the horizontal position of the trolley is obtained. The database provides comprehensive spatial constraints for the grab movement and ensures the feasibility of the motion trajectory. Based on the grid unit grab accessibility database, a multi-objective optimization algorithm is used to calculate the parameters of the motion trajectory, and a grid grab operation parameter set including the trolley acceleration, deceleration, uniform speed and rope retraction and release speed is obtained, which realizes the optimal design of the motion trajectory and improves the stability and efficiency of the motion process. Finally, a deep reinforcement learning algorithm is used to train the grid model of the cabin operation area, the grid unit grab accessibility database and the grid grab operation parameter set, and a dynamic obstacle avoidance grab instruction sequence including time-series-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters is obtained, which realizes the intelligent control and dynamic obstacle avoidance of the grab movement process, significantly improves the safety and reliability of the operation process, and at the same time, through the real-time adjustment of various compensation parameters, the stability and accuracy of the grab process are guaranteed, which greatly improves the automation level and operation efficiency of the ship unloading operation.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) Scanning the cabin space in segments according to a predetermined scanning trajectory to obtain multiple cabin space scanning data segments, and processing the multiple cabin space scanning data segments through time synchronization to generate time-aligned cabin scanning data;

[0033] (2) Processing the time-aligned cabin scan data through spatial coordinate conversion to obtain scan data in a unified coordinate system, and processing the scan data in the unified coordinate system after eliminating environmental interference to generate reference cabin scan data;

[0034] (3) The RANSAC algorithm is used to process the benchmark cabin scanning data to remove outliers, obtain the cabin point cloud basic data, and process the cabin point cloud basic data through density equalization to generate uniform cabin point cloud data;

[0035] (4) Using the voxel filtering algorithm to downsample the uniform cabin point cloud data, obtain the target cabin point cloud data, and then perform coordinate smoothing on the target cabin point cloud data to generate cabin smoothed point cloud data;

[0036] (5) The Euclidean clustering algorithm is used to identify the boundaries of the cabin smooth point cloud data, obtain the cabin contour point cloud data, and generate the cabin original three-dimensional point cloud data set by merging the cabin contour point cloud data and the cabin smooth point cloud data.

[0037] Specifically, the cabin space is segmented and scanned using a predetermined scanning trajectory, which includes scanning paths in the horizontal and vertical directions. In the horizontal direction, the scanning is performed along the moving trajectory of the ship unloader trolley, and a scan is performed every 0.5 meters. In the vertical direction, the scanning is performed from the top of the cabin downward, and a scan is performed every 0.3 meters to generate a grid-shaped scanning path. The laser scanner obtains a cabin space scanning data segment at each scanning position, and each data segment contains scanning time, laser emission angle, reflection signal intensity and distance information. When time synchronization is performed on multiple cabin space scanning data segments, the scanning time at the center of the cabin is used as the reference time point, the time difference between other scanning data segments and the reference time point is calculated, and each data segment is time-corrected to obtain time-aligned cabin scanning data. When the time-aligned cabin scanning data is subjected to spatial coordinate conversion processing, a unified coordinate system with the center of the cabin as the origin is established, with the X axis along the length direction of the cabin, the Y axis along the width direction of the cabin, and the Z axis vertically upward. According to the spatial position and attitude angle of the laser scanner at each scanning position, the scanning data is converted from the local coordinate system of the scanner to the unified coordinate system. The converted scanning data is interfered by environmental factors such as dust, temperature, and humidity. By analyzing the changing law of the reflected signal intensity, an environmental interference model is established, the scanning data is corrected, the influence of environmental interference is eliminated, and the benchmark cabin scanning data is obtained.

[0038] The RANSAC algorithm is used to remove outliers from the scanned data of the reference cabin. A plane model is established by repeated random sampling, the distance from the point to the plane is calculated, and the points whose distance exceeds the threshold are identified as outliers and removed. The specific process is: randomly select three non-collinear points to determine a plane, calculate the distance from other points to the plane, count the number of inliers whose distance is less than the threshold, repeat this process many times, select the plane model with the most inliers as the optimal model, and remove outliers that do not conform to the model. After the outliers are removed, the basic data of the cabin point cloud still has the problem of uneven local density. By calculating the local density of each point, the points in the area with too high density are deleted, and the points in the area with too low density are interpolated to achieve density balancing of the point cloud data. When voxel filtering and downsampling are performed on the uniform cabin point cloud data, the space is divided into equal-sized cubic grids (voxels), and the average value of the points in each voxel is taken as the representative point of the voxel. The size of the voxel is set according to actual needs. A larger voxel size will reduce the amount of data but lose detail information, while a smaller voxel size will retain more details but the amount of data is larger. The downsampled target cabin point cloud data is smoothed by using the local weighted average method, considering the influence of other points in the neighborhood of each point, calculating the smoothed coordinate value, reducing the noise in the data, and obtaining the cabin smooth point cloud data.

[0039] When the Euclidean clustering algorithm performs boundary recognition on the cabin smooth point cloud data, the Euclidean distance between each point and the adjacent points is first calculated, and the distance threshold is set to classify the points with a distance less than the threshold into the same class. Starting from an unclassified point, the points in its neighborhood are continuously searched until no new points that meet the conditions can be found, and the clustering of a class is completed. This process is repeated until all points are classified, and the boundary points are extracted from the clustering results to generate the cabin contour point cloud data. Finally, the cabin contour point cloud data is merged with the cabin smooth point cloud data to generate a complete cabin original 3D point cloud data set.

[0040] For example, the predetermined scanning trajectory covers a cabin space of 20 meters × 15 meters × 10 meters, and a scan is performed every 0.5 meters in the horizontal direction and every 0.3 meters in the vertical direction, and a total of 800 cabin space scan data segments are obtained. During time synchronization processing, the scanning time of the center position is selected as the reference time point, and the time deviation of other data segments is within ±0.5 seconds. A unified coordinate system is established during coordinate conversion. By analyzing the intensity of the reflected signal, it is found that the signal attenuation caused by dust is about 15%. After environmental interference elimination, the reference cabin scan data containing 2 million points is obtained. The RANSAC algorithm is used for 100 iterations, and the distance threshold is set to 0.1 meters, eliminating about 5% of outliers. After density equalization processing, the point cloud density remains at about 100 points per cubic meter. Voxels with a side length of 0.2 meters are used for downsampling, and the data volume is reduced to 500,000 points. The influence of neighboring points within a range of 0.5 meters is considered during coordinate smoothing. The distance threshold was set to 0.3 meters during Euclidean clustering, and 4,000 boundary points were identified, ultimately resulting in an original 3D point cloud dataset of the cabin containing 52,000 points.

[0041] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0042] (1) The original 3D point cloud data set of the cabin is processed by surface fitting to obtain the initial feature point set of the cabin, and the initial feature point set of the cabin is processed by geometric boundary segmentation to generate the cabin boundary feature data set;

[0043] (2) Processing the cabin boundary feature dataset through connected domain analysis to obtain the cabin area feature dataset, and processing the cabin area feature dataset through shape descriptor extraction to obtain the cabin shape feature dataset;

[0044] (3) Processing the cabin shape feature dataset using feature similarity calculation to obtain a cabin structure candidate dataset, and processing the cabin structure candidate dataset by feature region merging to generate a cabin structure feature dataset;

[0045] (4) Based on the height gradient calculation, the original three-dimensional point cloud data set of the cabin is processed to obtain the cargo surface height data set, and the cargo surface height data set is processed through density distribution analysis to generate a cargo distribution feature data set;

[0046] (5) The cargo distribution feature dataset is processed by stacking contour extraction to obtain the cargo contour feature dataset, and the cargo contour feature dataset is processed by region marking to generate the cargo surface feature dataset.

[0047] Specifically, when the original three-dimensional point cloud data set of the cabin is subjected to surface fitting processing, the local plane fitting method is adopted to select the point set within the neighborhood range of each point in the point cloud data, and calculate the best fitting plane. The normal vector and curvature features of the fitting plane are used to characterize the local surface characteristics, and the points with curvature values ​​greater than the preset threshold are marked as feature points to form the initial feature point set of the cabin. The initial feature point set of the cabin is subjected to geometric boundary segmentation processing. Based on the spatial distribution and normal vector consistency of the feature points, the adjacent feature points with similar geometric characteristics are grouped, and the main structural boundaries such as the wall panels and bulkheads of the cabin are identified to generate the cabin boundary feature data set. The connected domain analysis processing is used to perform regional division on the boundary features in the cabin boundary feature data set. Starting from any boundary feature point, the regional growing method is adopted to classify the points with adjacent spatial positions and similar geometric features into the same connected region until no new feature points that meet the conditions can be found, and the extraction of a connected domain is completed. This process is repeated until all boundary feature points are assigned to the corresponding connected domains to obtain the cabin region feature data set. Shape descriptors are extracted for each connected domain in the cabin area feature dataset, and geometric parameters such as area, perimeter, aspect ratio, as well as shape feature parameters such as main direction, roundness, rectangularity, etc. are calculated to form the cabin shape feature dataset.

[0048] The cabin shape feature dataset is processed by feature similarity calculation, and a multidimensional feature vector is used to represent the shape feature of each region. The feature vector distance between regions is calculated as a similarity measure. According to the preset similarity threshold, the regions with high feature similarity are clustered to generate a cabin structure candidate dataset. Feature region merging is performed on the candidate regions in the cabin structure candidate dataset, and the spatial relationship and shape feature continuity of adjacent regions are analyzed. Regions that belong to the same structure but are mis-segmented are merged to obtain a complete cabin structure feature dataset. The height gradient of the original three-dimensional point cloud dataset of the cabin is calculated. The height difference between adjacent points is calculated based on the horizontal plane to obtain the height gradient data representing the undulation of the cargo surface. According to the distribution characteristics of the height gradient value, the platform, slope and other areas where cargo is piled are identified to generate a cargo surface height dataset. The density distribution of the cargo surface height dataset is analyzed to calculate the number distribution of point clouds per unit area. The density and distribution law of cargo accumulation are analyzed in combination with the height information to obtain a cargo distribution feature dataset.

[0049] The cargo distribution feature dataset is subjected to accumulation contour extraction. Based on the height gradient and density distribution characteristics, the boundary points and contour lines of cargo accumulation are identified. The areas with significant height changes or density changes are marked as contour features to generate cargo contour feature datasets. The cargo contour feature dataset is subjected to region marking processing. Different areas divided by contour features are numbered and identified. The characteristic parameters such as the average height and area of ​​each area are calculated to obtain the cargo surface feature dataset.

[0050] For example, when the original three-dimensional point cloud data of the cabin with 52,000 points was surface fitted, a local neighborhood with a radius of 0.5 meters was selected, the average curvature threshold was calculated to be 0.15, and 3,000 initial feature points were extracted. When the geometric boundary was segmented, the normal vector angle threshold was set to 15 degrees, and the spatial distance threshold was set to 0.3 meters, and 800 boundary feature points were obtained. The connected domain analysis divided these boundary feature points into 12 main areas, and the shape descriptor extraction calculated 8 geometric parameters and 6 shape feature parameters of each area. The feature similarity calculation used a 14-dimensional feature vector, set a similarity threshold of 0.85, aggregated the 12 areas into 6 structural candidate areas, and finally obtained 4 complete cabin structure feature areas after feature area merging. For cargo surface feature extraction, the height gradient calculation used a sampling interval of 0.2 meters to identify 5 main cargo accumulation areas, and the density distribution analysis showed that the average point density was 80 points per square meter. The stacking contour extraction obtained a contour line feature with a total length of about 60 meters. The area marking divided the cargo surface into 8 characteristic areas. The average height of each area was calculated to be between 2.5 meters and 4.8 meters, and the area of ​​a single area ranged from 4 square meters to 12 square meters.

[0051] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0052] (1) Processing the cabin structure feature data set through boundary coordinate calculation to obtain the cabin boundary coordinate data set, and processing the cabin boundary coordinate data set through region segmentation to generate the cabin operation area coordinate data set;

[0053] (2) Processing the coordinate data set of the cabin operation area by grid unit division to obtain the basic grid unit data set, and processing the basic grid unit data set by coordinate position allocation to generate a grid unit position data set;

[0054] (3) Using spatial mapping to process the grid cell location dataset and the cargo surface feature dataset to obtain the grid cell surface dataset, and then processing the grid cell surface dataset through height statistics to generate a grid cell height dataset;

[0055] (4) Processing the grid cell height data set through key point extraction to obtain a grid cell feature point data set, and processing the grid cell feature point data set through parameter calculation to generate a grid cell parameter data set. The grid cell parameter data set includes the grid center point coordinates, average height value, highest point and lowest point parameters of each grid cell;

[0056] (5) Based on the spatial relationship, a grid unit parameter data set is established and processed, a grid unit associated data set is obtained, and the grid unit associated data set is processed after data organization to construct a grid model of the cabin operation area.

[0057] Specifically, when calculating the boundary coordinates of the cabin structure feature dataset, the boundary point sequence in the structural features is extracted, the spatial coordinate distribution of the boundary is calculated, and a closed boundary contour line is generated. By analyzing the continuity and curvature changes of the boundary points, the corner points and feature points of the cabin structure are identified, and the cabin boundary coordinate dataset is generated. The cabin boundary coordinate dataset is segmented, and the main operating areas of the cabin are divided based on the geometric shape and spatial distribution of the boundary contour, including the loading and unloading operation area, the avoidance area and the buffer zone, to obtain the cabin operation area coordinate dataset.

[0058] The coordinate data set of the cabin operation area is processed by grid unit division, and the operation space is divided into regular grid units. When dividing, the size of the grab and the operation characteristics are considered to determine the appropriate grid unit size to ensure that a single grid unit can accommodate the operation space of the grab. Each basic grid unit is numbered and identified, and a grid index system is established. The position of the grid unit is uniquely determined by the row and column numbers to generate a basic grid unit data set. The coordinate position of the basic grid unit data set is assigned, and the spatial range and boundary coordinates of each grid unit are calculated to obtain a grid unit position data set. The grid unit position data set is spatially mapped with the cargo surface feature data set, and the cargo surface feature points are projected into the corresponding grid units. The feature points in each grid unit are statistically analyzed, and the spatial distribution characteristics of the feature points are calculated to generate a grid unit surface data set. The grid unit surface data set is highly statistically processed to analyze the height distribution law of the cargo surface in each grid unit to obtain a grid unit height data set.

[0059] The key point extraction process in the grid unit height data set mainly includes the identification of local extreme points, inflection points and feature points in each grid unit. The extracted key points are used to characterize the important feature positions of the cargo surface and constitute the grid unit feature point data set. The grid unit feature point data set is parameterized to obtain the grid unit parameter data set, where:

[0060] 1. Grid center point coordinate calculation formula:

[0061]

[0062] in, , is the center point coordinate, , is the initial coordinate of the i-th point in the grid, , , , is the position correction coefficient, , is the coordinate offset, is the boundary correction factor, and N is the total number of points in the grid.

[0063] 2. Average height calculation formula:

[0064]

[0065] in, is the average height value, is the height value of the i-th point, is the weight coefficient, is the local correction coefficient, is the height compensation factor, is the global correction factor.

[0066] 3. The highest point calculation formula:

[0067]

[0068] in, is the highest point height, is the height value of the i-th point, is the altitude correction factor, is the outlier processing coefficient, is the height compensation value, is the boundary effect correction factor.

[0069] 4. The lowest point parameter calculation formula:

[0070]

[0071] in, is the lowest point height, is the height value of the i-th point, is the lower bound correction coefficient, is the low point screening coefficient, is the height compensation value, is the bottom effect correction factor.

[0072] The spatial relationship of the grid unit parameter data set is established, the position relationship and parameter change law between adjacent grid units are analyzed, the connection relationship and topological structure between grid units are established, and the grid unit association data set is generated. The grid unit association data set is organized and processed, and the spatial information, parameter information and association information of the grid unit are integrated into a unified data structure to build a complete grid model of the cabin operation area.

[0073] For example, the size of the cabin operation area is 20 meters × 15 meters, divided into 400 grid units, each unit size is 1 meter × 1 meter. Through boundary coordinate calculation, 12 characteristic corner points are identified and 3 main operation areas are determined. The grid unit numbered (8, 6) is analyzed. This unit contains 57 characteristic points of the cargo surface. The coordinates of the grid center point are calculated to be (8.52m, 6.35m), the average height is 3.75m, the highest point is 4.12m, and the lowest point is 3.28m. When establishing the spatial relationship, the grid unit is connected with the surrounding 8 adjacent units to generate a local grid structure, which is finally integrated into the grid model of the cabin operation area.

[0074] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0075] (1) Each grid cell in the grid model of the cabin operation area is processed through spatial position analysis to obtain a grid cell spatial position data set, and the grid cell spatial position data set is processed through grab kinematic calculation to generate a grab motion reference data set;

[0076] (2) The grab motion reference dataset is processed using collision detection to obtain a grid unit obstacle distribution dataset, which is then processed through safety distance calculation to generate a grid unit safety path dataset.

[0077] (3) Calculating and processing the grid unit safe path dataset based on the entry angle to obtain the grid unit angle dataset, and then calculating and processing the grid unit angle dataset through the rope length to generate the grid unit rope length dataset;

[0078] (4) The grid unit rope length data set is processed by grab balance analysis to obtain the grid unit force data set, and the grid unit force data set is processed by trolley horizontal position calculation to generate the grid unit trolley horizontal position data set;

[0079] (5) The grid unit angle dataset, the grid unit rope length dataset, and the grid unit trolley horizontal position dataset are integrated to obtain the grid unit accessibility parameter dataset. The grid unit accessibility parameter dataset is processed through database construction to generate a grid unit grasping accessibility database. The grid unit grasping accessibility database includes the entry angle, rope length, and trolley horizontal position of each grid unit.

[0080] Specifically, the spatial position analysis is performed on each grid unit in the grid model of the cabin operation area. The relative position and posture of the grid unit in the cabin space, including the center position, normal vector and boundary characteristics of the grid unit, are calculated to obtain the grid unit spatial position data set. The grab kinematics calculation is performed on this data set to analyze the motion parameters required for the grab to move from the initial position to the target grid unit, including motion path, speed and acceleration, and generate a grab motion reference data set. Collision detection is performed on the grab motion reference data set to analyze the positional relationship between the grab and the cabin structure and other obstacles during the movement, establish a collision boundary model, and generate a grid unit obstacle distribution data set. The safety distance calculation is performed on this data set to determine the minimum safety distance that needs to be maintained between the grab and the obstacle, plan the obstacle avoidance path, and generate a grid unit safety path data set.

[0081] The entry angle is calculated using the following formula:

[0082]

[0083] in, is the entry angle, , are the velocity components of the grab bucket in the horizontal and vertical directions, is the attitude correction factor, is the angle compensation increment, is the stability coefficient, is the swing amplitude of the grab bucket.

[0084] Rope length is calculated using the following formula:

[0085]

[0086] in, is the required rope length, ( , ) is the position coordinate of the winch, ( , ) is the grab bucket target position coordinate, is the relaxation coefficient, is the length compensation factor, is the elongation of the rope.

[0087] The calculation formula for mesh unit force analysis is:

[0088]

[0089] in, is the total force, M is the grab mass, is the gravitational acceleration vector, is the torque distribution coefficient, are the force vectors, is the dynamic compensation coefficient, is the tension vector.

[0090] The calculation formula of the horizontal position of the car is:

[0091]

[0092] in, is the horizontal position of the trolley, is the position weight factor, is the reference position coordinate, is the displacement compensation coefficient, is the position offset, is the orbit correction factor, is the surplus coefficient.

[0093] The entry angle of the grid unit safe path dataset is calculated. Based on the motion characteristics of the grab bucket and the spatial relationship of the target position, the optimal entry direction is determined to obtain the grid unit angle dataset. The required rope length is calculated based on the position and height information of the grid unit to generate the grid unit rope length dataset. The grab bucket balance analysis is performed on the grid unit rope length dataset. Considering the effects of rope tension, grab bucket gravity and inertia force, the force state of the grab bucket at different positions is calculated to generate the grid unit force dataset. Based on the force analysis results, the horizontal position of the trolley that meets the balance condition is calculated to obtain the grid unit trolley horizontal position dataset. The grid unit angle dataset, rope length dataset and trolley horizontal position dataset are integrated to establish a unified parameter index system to generate the grid unit accessibility parameter dataset. A database structure is established for the dataset, and the parameters such as the entry angle, rope length and trolley horizontal position of each grid unit are stored in the grid unit grab accessibility database.

[0094] For example, the grid unit numbered (5, 7) is analyzed, and the spatial position coordinates of the unit are first determined (5.5m, 7.5m, 4.2m). The motion trajectory of the grab bucket is obtained through kinematic calculation, and the design speed is 1.2m / s. Collision detection finds that there is an obstacle near the unit, and a safety distance of 0.8m needs to be maintained. The entry angle calculation obtains the optimal angle of 65 degrees, considering the swing amplitude of 0.2rad. The rope length calculation result is 9.3m, including a margin of 0.3m. The force analysis shows that the grab bucket is subjected to a combined force of 12kN at this position, and the horizontal position of the trolley is determined to be 6.8m. These parameters are integrated into the accessibility database to guide subsequent grabbing operations.

[0095] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0096] (1) The grid cell reachability database is processed through the spatial constraint analysis of the vehicle motion to obtain the vehicle motion constraint data set, and the vehicle motion constraint data set is processed through acceleration performance calculation to generate the vehicle acceleration parameter data set;

[0097] (2) The acceleration parameter data set of the trolley is processed by using stability analysis to obtain the deceleration parameter data set of the trolley. The deceleration parameter data set of the trolley is processed by speed stability calculation to generate the speed data set of the trolley in the uniform speed segment.

[0098] (3) Based on the analysis of rope force, the grid unit grab accessibility database is processed to obtain the rope dynamic parameter data set, and the rope dynamic parameter data set is processed by tension calculation to generate the rope retraction and release speed data set;

[0099] (4) The trolley motion correlation analysis is used to process the trolley acceleration parameter data set, the trolley deceleration parameter data set, and the trolley uniform speed data set to obtain the trolley motion comprehensive parameter data set. The trolley motion comprehensive parameter data set and the rope retraction and release speed data set are then processed through dynamic coupling to generate a motion parameter coordination data set.

[0100] (5) The motion parameter coordination data set is comprehensively processed through the operation constraints to obtain the motion trajectory parameter data set, and the motion trajectory parameter data set is processed through parameter organization to generate a grid grasping operation parameter set. The grid grasping operation parameter set includes the trolley acceleration, deceleration, uniform speed, and rope retraction and release speed.

[0101] Specifically, the spatial constraint analysis of the trolley motion is performed on the entry angle, rope length and trolley horizontal position data in the grid unit grab accessibility database to determine the spatial boundary conditions of the trolley motion on the track. The constraint conditions include track length limit, deceleration buffer zone requirements, safety distance, etc., and the influence of the grab bucket swing on the trolley motion is considered. The trolley motion constraint data set is generated according to the spatial constraint conditions, and then the acceleration performance analysis of the trolley motion constraint data set is performed to calculate the maximum allowable acceleration and minimum required acceleration under different working conditions. The factors such as the trolley deadweight, load weight, and motor performance are considered to generate the trolley acceleration parameter data set. The stability analysis of the trolley acceleration parameter data set is performed to analyze the swing state of the grab bucket during the acceleration process, calculate the critical deceleration and safe deceleration range, and generate the trolley deceleration parameter data set. The speed stability analysis of the deceleration parameter is performed to calculate the uniform speed requirement between the acceleration section and the deceleration section to ensure the stability and safety of the operation process, and obtain the uniform speed data set of the trolley.

[0102] The rope force analysis is based on the rope length and position information in the grid unit grab accessibility database. The force state of the rope during the movement is calculated, including the influence of gravity, inertia and swing force, and the rope dynamic parameter data set is generated. The tension of the rope dynamic parameters is calculated, the risk of rope slack and over-tension is analyzed, the appropriate range of retraction and release speed is determined, and the rope retraction and release speed data set is generated. The trolley acceleration parameter data set, deceleration parameter data set and uniform speed data set are correlated and analyzed. Considering the mutual influence and constraint relationship between each motion parameter, the continuity and stability of the motion process are evaluated, and the trolley motion comprehensive parameter data set is obtained. The dynamic coupling analysis of the trolley motion comprehensive parameters and the rope retraction and release speed is carried out to study the coordination of the trolley motion and the rope motion, calculate the optimal motion parameter combination, and generate the motion parameter coordination data set.

[0103] Based on the motion parameter coordination data set, the operation constraints are comprehensively processed, and multiple factors such as equipment performance limitations, safety requirements and operating efficiency are considered to generate a motion trajectory parameter data set that meets various constraints. The motion trajectory parameters are uniformly organized and managed to generate a grid grabbing operation parameter set, which includes key parameters such as trolley acceleration, deceleration, uniform speed, and rope retraction and release speed.

[0104] For example, a grid unit located in the middle of the cabin is selected, the horizontal position of the trolley is 12 meters, the entry angle is 55 degrees, and the required rope length is 15 meters. First, the trolley motion space constraint analysis is carried out. Considering that the total length of the track is 25 meters, a 2-meter buffer zone is required at both ends, and the actual available motion space is determined to be 21 meters. Combined with the rated lifting capacity of 5 tons, the performance of the trolley and the motor is analyzed, and the maximum allowable acceleration is calculated to be 0.8m / s², and the safe acceleration value is 0.6m / s². The stability analysis shows that in order to suppress the swing of the grab bucket, the deceleration should not exceed 0.5m / s², and the actual value is 0.4m / s². The speed stability calculation shows that the uniform speed is 1.5m / s, which can ensure a smooth and controllable operation process. The rope force analysis shows that the maximum dynamic load borne by the rope during the lifting process is 65kN. Considering the safety factor of 1.5, the rope tension needs to be kept at no less than 40kN. Based on these parameters, the rope retraction speed range is calculated to be 0.3-0.8m / s, and the actual operating speed is set to 0.6m / s. A correlation analysis of all motion parameters shows that when the trolley accelerates to 1.5m / s, the rope retraction speed needs to be adjusted to 0.6m / s synchronously to maintain the stability of the grab. The dynamic coupling analysis determines the optimal combination of parameters: the trolley accelerates at 0.6m / s² when starting, reaches a uniform speed of 1.5m / s for 2.5 seconds, decelerates at 0.4m / s² when approaching the target position, and the rope retraction speed is smoothly adjusted to 0.6m / s during the entire process. This set of parameters has been comprehensively evaluated through constraints to meet the equipment performance requirements and ensure operational safety and efficiency.

[0105] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0106] (1) Perform state space construction processing on the parameters of the grid model of the cabin operation area to generate an environmental state data set, and perform action space construction processing on the grid capture operation parameter set to obtain an action state data set;

[0107] (2) The environment state data set and the action state data set are processed through the reward function to generate a reward evaluation data set, and the reward evaluation data set is processed by reinforcement learning training to obtain the basic strategy data set;

[0108] (3) Based on the basic strategy data set, perform horizontal motion compensation calculation processing to obtain a horizontal motion trajectory compensation data set, and perform time series correlation analysis on the horizontal motion trajectory compensation data set to obtain a time series horizontal trajectory data set;

[0109] (4) Perform vertical motion compensation calculation processing on the basic strategy data set to obtain a vertical motion trajectory compensation data set, and perform grab posture calculation processing on the vertical motion trajectory compensation data set to generate a grab posture compensation data set;

[0110] (5) Based on the grid model of the cabin operation area, the grid unit grasping reachability database and the grid grasping operation parameter set, the strategy is trained through the deep reinforcement learning algorithm to output a dynamic obstacle avoidance grasping instruction sequence, which contains the timing-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters.

[0111] Specifically, when constructing the state space of the parameters in the grid model of the cabin operation area, it is necessary to extract the geometric features, location information, cargo distribution status and other parameters of the grid unit. The state space includes static parameters such as the center coordinates, height value, boundary features of the grid unit, as well as dynamic parameters such as the grab position, speed, and posture, to generate an environmental state data set. The action space is constructed for the parameters such as the acceleration, deceleration, uniform speed, and rope retraction speed of the trolley in the grid grab operation parameter set, and the value range and discretization interval of each parameter are defined to generate an action state data set. The reward function is constructed by combining the environmental state data set and the action state data set to evaluate the execution effect of each action. The reward function considers multiple evaluation indicators, including the accuracy of reaching the target position, the smoothness of the movement process, energy consumption, and work efficiency. The reward evaluation data set is trained by reinforcement learning. Through repeated trial and error and optimization, the optimal action strategy is learned to generate a basic strategy data set.

[0112] The horizontal motion compensation calculation uses the following formula:

[0113]

[0114] in, is the horizontal compensation amount, is the compensation coefficient, is the displacement deviation, is the time integration weight, is the speed function, is the gradient correction coefficient, is the objective function gradient.

[0115] Vertical motion compensation calculation formula:

[0116]

[0117] in, is the vertical compensation amount, is the height compensation coefficient, is the height deviation, is the dynamic response coefficient, is the acceleration function, is the energy compensation coefficient, is the change in potential energy.

[0118] The formula for calculating the force of the grid unit is:

[0119]

[0120] in, For the joint efforts, is the force coefficient, is the force vector, is the rotational damping coefficient, is the rotation torque, is the air resistance coefficient, is the resistance vector.

[0121] Grab bucket posture calculation formula:

[0122]

[0123] in, is the attitude adjustment amount, is the attitude coefficient, is the angle deviation, is the angular velocity integral coefficient, is the angular velocity function, is the torque coefficient, is the external torque.

[0124] Calculate the horizontal motion compensation parameters through the basic strategy data set, perform time-series correlation analysis on the compensated motion trajectory, establish the relationship between the compensation parameters at different time points, and generate a time-series horizontal trajectory data set. Perform vertical motion compensation calculation on the basic strategy data set, combine the height change and speed requirements to obtain the vertical motion trajectory compensation data set, then calculate the grab posture, determine the posture adjustment strategy, and generate the grab posture compensation data set. Fusion the time-series horizontal trajectory data set, vertical motion trajectory compensation data set, and grab posture compensation data set to generate a complete grab instruction sequence. Perform dynamic obstacle avoidance optimization on the grab instruction, adjust the motion trajectory and posture parameters in real time, and finally generate a dynamic obstacle avoidance grab instruction sequence.

[0125] For example, the target grid unit is located in the middle of the cabin, and the environmental state includes information such as the center position of the grid (8m, 6m), the cargo height of 3.5m, and the distribution of surrounding obstacles. The action space defines the speed range of the trolley as 0-2m / s and the acceleration range as 0-1m / s². In the reward function setting, the accuracy weight of reaching the target is 0.5, the motion smoothness weight is 0.3, and the energy consumption weight is 0.2. After 1000 training iterations, the basic control strategy is obtained. The horizontal motion compensation calculation results show that it is necessary to increase the displacement compensation of 0.2m in the x-direction and the vertical compensation amount is 0.15m. Timing analysis shows that the compensation amount needs to be gradually adjusted within 5m of the target point. The grab posture calculation results show that a posture compensation angle of 5 degrees is required. The final generated instruction sequence contains compensation parameters for 50 time points, achieving a smooth and accurate grasping action.

[0126] The above describes the method for scanning the position of the grab bucket for the ship unloader in the embodiment of the present application. The following describes the system for scanning the position of the grab bucket for the ship unloader in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the grab position scanning system for ship unloader includes:

[0127] A scanning module is used to measure the cabin space through laser scanning and obtain the original three-dimensional point cloud data set of the cabin;

[0128] A clustering module, used for processing the original three-dimensional point cloud data set of the cabin through feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set;

[0129] A processing module, used for gridding the cabin structure feature data set and the cargo surface feature data set to establish a grid model of the cabin operation area, wherein each grid unit includes the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters;

[0130] A calculation module, for calculating accessibility parameters for each grid unit in the grid model of the cabin operation area, and constructing a grid unit grabbing accessibility database, wherein the grid unit grabbing accessibility database includes an entry angle, a rope length, and a horizontal position of a trolley for each grid unit;

[0131] An analysis module, for calculating motion trajectory parameters based on the grid unit grabbing accessibility database using a multi-objective optimization algorithm, and generating a grid grabbing operation parameter set, wherein the grid grabbing operation parameter set includes trolley acceleration, deceleration, uniform speed, and rope retraction and release speed;

[0132] A decision module is used to perform strategy training through a deep reinforcement learning algorithm based on the grid model of the cabin operation area, the grid unit grasping reachability database and the grid grasping operation parameter set, and output a dynamic obstacle avoidance grasping instruction sequence, wherein the dynamic obstacle avoidance grasping instruction sequence includes a timing-based horizontal motion trajectory compensation parameter, a vertical motion trajectory compensation parameter and a grab posture compensation parameter.

[0133] Through the coordinated cooperation of the above-mentioned components, by laser scanning the cabin space, a high-precision original three-dimensional point cloud data set of the cabin can be obtained, which provides a reliable data basis for subsequent feature extraction and spatial analysis, and avoids the recognition errors of traditional visual systems in complex environments; by performing feature extraction and spatial clustering processing on the original three-dimensional point cloud data set, the cabin structure feature data set and the cargo surface feature data set are obtained respectively, which realizes the effective separation of cabin space information and improves the recognition accuracy of spatial features; the cabin structure feature data set and the cargo surface feature data set are gridded to obtain a grid model of the cabin operation area containing the grid center point coordinates, average height value, highest point and lowest point parameters, so that the complex spatial information is expressed in a standardized manner, which is convenient for subsequent path planning and control; by performing reachability calculation processing on each grid unit in the grid model of the cabin operation area, the grid unit grasping accessibility including the entry angle, rope length and horizontal position of the trolley is obtained. The accessibility database provides comprehensive spatial constraints for the grab movement and ensures the feasibility of the motion trajectory. Based on the grid unit grab accessibility database, a multi-objective optimization algorithm is used to calculate the parameters of the motion trajectory, and a grid grab operation parameter set including the trolley acceleration, deceleration, uniform speed and rope retraction and release speed is obtained, which realizes the optimal design of the motion trajectory and improves the stability and efficiency of the motion process. Finally, the deep reinforcement learning algorithm is used to perform strategy training on the grid model of the cabin operation area, the grid unit grab accessibility database and the grid grab operation parameter set, and a dynamic obstacle avoidance grab instruction sequence including time-series-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters is obtained, which realizes the intelligent control and dynamic obstacle avoidance of the grab movement process, significantly improves the safety and reliability of the operation process, and at the same time, through the real-time adjustment of various compensation parameters, the stability and accuracy of the grab process are guaranteed, which greatly improves the automation level and operation efficiency of the ship unloading operation.

[0134] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for scanning the position of a grab bucket for a ship unloader, characterized in that: The method for scanning the position of the grab bucket dedicated to the ship unloader comprises: The cabin space is measured by laser scanning to obtain the original 3D point cloud data set of the cabin; Processing the original three-dimensional point cloud data set of the cabin by feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set; Gridding the cabin structure feature data set and the cargo surface feature data set to establish a cabin operation area grid model, wherein each grid unit includes grid center point coordinates, average height value, highest point and lowest point parameters; The accessibility parameters are calculated for each grid unit in the grid model of the cabin operation area, and a grid unit grab accessibility database is constructed, including: processing each grid unit in the grid model of the cabin operation area through spatial position analysis to obtain a grid unit spatial position data set, and processing the grid unit spatial position data set through grab kinematic calculation to generate a grab motion reference data set; processing the grab motion reference data set through collision detection to obtain a grid unit obstacle distribution data set, and processing the grid unit obstacle distribution data set through safety distance calculation to generate a grid unit safety path data set; processing the grid unit safety path data set based on entry angle calculation to obtain a grid unit angle data set, and processing the grid unit angle data set through rope length meter. The grid unit angle data set is processed by calculation to generate a grid unit rope length data set; the grid unit rope length data set is processed by grab balance analysis to obtain a grid unit force data set, and the grid unit force data set is processed by trolley horizontal position calculation to generate a grid unit trolley horizontal position data set; the grid unit angle data set, the grid unit rope length data set and the grid unit trolley horizontal position data set are integrated to obtain a grid unit accessibility parameter data set, and the grid unit accessibility parameter data set is processed by database construction to generate a grid unit grab accessibility database, wherein the grid unit grab accessibility database includes an entry angle, a rope length and a trolley horizontal position of each grid unit; A multi-objective optimization algorithm is used to calculate motion trajectory parameters based on the grid unit grasping accessibility database to generate a grid grasping operation parameter set, including: processing the grid unit grasping accessibility database through trolley motion space constraint analysis to obtain a trolley motion constraint data set, and processing the trolley motion constraint data set through acceleration performance calculation to generate a trolley acceleration parameter data set; processing the trolley acceleration parameter data set through stability analysis to obtain a trolley deceleration parameter data set, and processing the trolley deceleration parameter data set through speed stability calculation to generate a trolley uniform speed data set; processing the grid unit grasping accessibility database based on rope force analysis to obtain a rope dynamic parameter data set, and processing the rope dynamic parameter data set through a tension meter The rope dynamic parameter data set is processed by calculation to generate a rope retracting and releasing speed data set; the trolley motion association analysis is used to process the trolley acceleration parameter data set, the trolley deceleration parameter data set, and the trolley uniform speed data set to obtain a trolley motion comprehensive parameter data set, and the trolley motion comprehensive parameter data set and the rope retracting and releasing speed data set are processed by dynamic coupling to generate a motion parameter coordination data set; the motion parameter coordination data set is processed by operation constraints to obtain a motion trajectory parameter data set, and the motion trajectory parameter data set is processed by parameter organization to generate a grid grabbing operation parameter set, and the grid grabbing operation parameter set includes the trolley acceleration, deceleration, uniform speed and rope retracting and releasing speed; Based on the grid model of the cabin operation area, the grid unit grasping reachability database and the grid grasping operation parameter set, a strategy training is performed through a deep reinforcement learning algorithm to output a dynamic obstacle avoidance grasping instruction sequence, including: performing state space construction processing on the parameters of the grid model of the cabin operation area to generate an environmental state data set, performing action space construction processing on the grid grasping operation parameter set to obtain an action state data set; constructing the environmental state data set and the action state data set through a reward function to generate a reward evaluation data set, performing reinforcement learning training processing on the reward evaluation data set to obtain a basic strategy data set; performing horizontal motion compensation calculation processing based on the basic strategy data set to obtain a horizontal motion compensation data set. A horizontal motion trajectory compensation data set is prepared, and a time-series correlation analysis is performed on the horizontal motion trajectory compensation data set to obtain a time-series horizontal trajectory data set; for the basic strategy data set, vertical motion compensation calculation processing is performed to obtain a vertical motion trajectory compensation data set, and grab posture calculation processing is performed on the vertical motion trajectory compensation data set to generate a grab posture compensation data set; based on the grid model of the cabin operating area, the grid unit grab accessibility database and the grid grab operation parameter set, strategy training is performed through a deep reinforcement learning algorithm to output a dynamic obstacle avoidance grabbing instruction sequence, wherein the dynamic obstacle avoidance grabbing instruction sequence includes time-series-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters.

2. The method for scanning the position of a grab bucket for a ship unloader according to claim 1, characterized in that: The method of measuring the cabin space by laser scanning to obtain the original three-dimensional point cloud data set of the cabin includes: Scanning the cabin space in segments according to a predetermined scanning trajectory to obtain a plurality of cabin space scanning data segments, and processing the plurality of cabin space scanning data segments by time synchronization to generate cabin scanning data with time alignment; Processing the time-series aligned cabin scan data by spatial coordinate conversion to obtain scan data in a unified coordinate system, and processing the scan data in the unified coordinate system by eliminating environmental interference to generate reference cabin scan data; The reference cabin scanning data is processed by using the RANSAC algorithm to remove outliers, obtain cabin point cloud basic data, and the cabin point cloud basic data is processed by density equalization to generate uniform cabin point cloud data; Downsampling the uniform cabin point cloud data by using a voxel filtering algorithm to obtain target cabin point cloud data, and performing coordinate smoothing on the target cabin point cloud data to generate cabin smoothed point cloud data; A Euclidean clustering algorithm is applied to identify the boundary of the cabin smooth point cloud data, to obtain cabin contour point cloud data, and to generate an original three-dimensional point cloud data set of the cabin by merging the cabin contour point cloud data and the cabin smooth point cloud data.

3. The method for scanning the position of a grab bucket for a ship unloader according to claim 1, characterized in that: The process of processing the original three-dimensional point cloud data set of the cabin by feature extraction and spatial clustering to obtain the cabin structure feature data set and the cargo surface feature data set includes: Processing the original three-dimensional point cloud data set of the cabin by surface fitting to obtain an initial feature point set of the cabin, and processing the initial feature point set of the cabin by geometric boundary segmentation to generate a cabin boundary feature data set; Processing the cabin boundary feature data set by connected domain analysis to obtain a cabin region feature data set, and extracting and processing the cabin region feature data set by shape descriptor to obtain a cabin shape feature data set; Processing the cabin shape feature dataset by feature similarity calculation to obtain a cabin structure candidate dataset, and processing the cabin structure candidate dataset by feature region merging to generate a cabin structure feature dataset; Processing the original three-dimensional point cloud data set of the cabin based on height gradient calculation to obtain a cargo surface height data set, and processing the cargo surface height data set through density distribution analysis to generate a cargo distribution feature data set; The cargo distribution feature data set is processed by stacking contour extraction to obtain a cargo contour feature data set, and the cargo contour feature data set is processed by region marking to generate a cargo surface feature data set.

4. The method for scanning the position of a grab bucket for a ship unloader according to claim 1, characterized in that: The gridding process of the cabin structure feature data set and the cargo surface feature data set is performed to establish a gridding model of the cabin operation area, wherein each grid unit contains the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters, including: Processing the cabin structure feature data set by boundary coordinate calculation to obtain a cabin boundary coordinate data set, and processing the cabin boundary coordinate data set by region segmentation to generate a cabin operation area coordinate data set; Processing the cabin operation area coordinate data set by grid unit division to obtain a basic grid unit data set, and processing the basic grid unit data set by coordinate position allocation to generate a grid unit position data set; Processing the grid unit position dataset and the cargo surface feature dataset by spatial mapping to obtain a grid unit surface dataset, and processing the grid unit surface dataset by height statistics to generate a grid unit height dataset; Processing the grid unit height data set through key point extraction to obtain a grid unit feature point data set, and processing the grid unit feature point data set through parameter calculation to generate a grid unit parameter data set, wherein the grid unit parameter data set includes the grid center point coordinates, average height value, highest point and lowest point parameters of each grid unit; The grid unit parameter data set is established and processed based on the spatial relationship, a grid unit associated data set is obtained, and the grid unit associated data set is processed through data organization to construct a grid model of the cabin operation area.

5. A special grab position scanning system for ship unloader, used to implement the special grab position scanning method for ship unloader as claimed in any one of claims 1 to 4, characterized in that: The grab position scanning system for ship unloader comprises: A scanning module is used to measure the cabin space through laser scanning and obtain the original three-dimensional point cloud data set of the cabin; A clustering module, used for processing the original three-dimensional point cloud data set of the cabin through feature extraction and spatial clustering to obtain a cabin structure feature data set and a cargo surface feature data set; A processing module, used for gridding the cabin structure feature data set and the cargo surface feature data set to establish a grid model of the cabin operation area, wherein each grid unit includes the coordinates of the grid center point, the average height value, the highest point and the lowest point parameters; The calculation module is used to calculate the accessibility parameters for each grid unit in the grid model of the cabin operation area and construct a grid unit grab accessibility database, including: processing each grid unit in the grid model of the cabin operation area through spatial position analysis to obtain a grid unit spatial position data set, and processing the grid unit spatial position data set through grab kinematic calculation to generate a grab motion reference data set; processing the grab motion reference data set through collision detection to obtain a grid unit obstacle distribution data set, and processing the grid unit obstacle distribution data set through safety distance calculation to generate a grid unit safety path data set; processing the grid unit safety path data set based on entry angle calculation to obtain a grid unit angle data set, and processing the grid unit safety path data set through rope The grid unit angle data set is processed by length calculation to generate a grid unit rope length data set; the grid unit rope length data set is processed by grab balance analysis to obtain a grid unit force data set, and the grid unit force data set is processed by trolley horizontal position calculation to generate a grid unit trolley horizontal position data set; the grid unit angle data set, the grid unit rope length data set and the grid unit trolley horizontal position data set are integrated to obtain a grid unit accessibility parameter data set, and the grid unit accessibility parameter data set is processed by database construction to generate a grid unit grab accessibility database, wherein the grid unit grab accessibility database includes an entry angle, a rope length and a trolley horizontal position of each grid unit; The analysis module is used to calculate the motion trajectory parameters based on the grid unit grasping accessibility database by adopting a multi-objective optimization algorithm to generate a grid grasping operation parameter set, including: processing the grid unit grasping accessibility database through the trolley motion space constraint analysis to obtain the trolley motion constraint data set, and processing the trolley motion constraint data set through acceleration performance calculation to generate a trolley acceleration parameter data set; processing the trolley acceleration parameter data set through stability analysis to obtain the trolley deceleration parameter data set, and processing the trolley deceleration parameter data set through speed stability calculation to generate a trolley uniform speed segment speed data set; processing the grid unit grasping accessibility database based on rope force analysis to obtain the rope dynamic parameter data set, and The rope dynamic parameter data set is processed by tightness calculation to generate a rope retraction and release speed data set; the trolley motion association analysis is used to process the trolley acceleration parameter data set, the trolley deceleration parameter data set, and the trolley uniform speed data set to obtain a trolley motion comprehensive parameter data set, and the trolley motion comprehensive parameter data set and the rope retraction and release speed data set are processed by dynamic coupling to generate a motion parameter coordination data set; the motion parameter coordination data set is comprehensively processed by operation constraints to obtain a motion trajectory parameter data set, and the motion trajectory parameter data set is processed by parameter organization to generate a grid grabbing operation parameter set, and the grid grabbing operation parameter set includes the trolley acceleration, deceleration, uniform speed and rope retraction and release speed; The decision module is used to perform strategy training through a deep reinforcement learning algorithm based on the grid model of the cabin operation area, the grid unit grasping reachability database and the grid grasping operation parameter set, and output a dynamic obstacle avoidance grasping instruction sequence, including: performing state space construction processing on the parameters of the grid model of the cabin operation area to generate an environmental state data set, performing action space construction processing on the grid grasping operation parameter set to obtain an action state data set; constructing the environmental state data set and the action state data set through a reward function to generate a reward evaluation data set, performing reinforcement learning training processing on the reward evaluation data set to obtain a basic strategy data set; performing horizontal motion compensation calculation processing based on the basic strategy data set, A horizontal motion trajectory compensation data set is obtained, and a time-series correlation analysis is performed on the horizontal motion trajectory compensation data set to obtain a time-series horizontal trajectory data set; vertical motion compensation calculation processing is performed on the basic strategy data set to obtain a vertical motion trajectory compensation data set, and grab posture calculation processing is performed on the vertical motion trajectory compensation data set to generate a grab posture compensation data set; based on the grid model of the cabin operating area, the grid unit grab accessibility database and the grid grab operation parameter set, strategy training is performed through a deep reinforcement learning algorithm to output a dynamic obstacle avoidance grabbing instruction sequence, wherein the dynamic obstacle avoidance grabbing instruction sequence includes time-series-based horizontal motion trajectory compensation parameters, vertical motion trajectory compensation parameters and grab posture compensation parameters.

Images