A laser radar system for obstacle avoidance of a bucket wheel machine

Abstract

This invention belongs to the field of obstacle avoidance technology for bucket wheel excavators, specifically a lidar system for obstacle avoidance in bucket wheel excavators. It includes a ground point cloud recognition module, a working material pile recognition module, a candidate target clustering module, an obstacle attribute determination module, and a safety control module. The system identifies the ground point cloud by performing height gradient analysis on the point cloud in the bucket wheel excavator's working area. Then, it predicts the material pile envelope based on the cantilever attitude and defines the point cloud on the material pile surface by combining surface continuity and the angle of repose, thereby structurally separating the ground, material pile, and residual point cloud. After eliminating the former two, the residual point cloud is three-dimensionally clustered to obtain candidate target clusters. By comparing the difference in reflection intensity and spatial relationship between the candidate clusters and the material pile surface, real obstacles are identified through multi-dimensional logic determination. Finally, kinematic collision detection is performed on the obstacles, and avoidance, deceleration, or stopping commands are generated according to the collision level, ensuring that alarms are only triggered for real, stable obstacles.

Description

Technical Field

[0001] This invention belongs to the field of obstacle avoidance technology for bucket wheel excavators, specifically a laser radar system for obstacle avoidance in bucket wheel excavators. Background Technology

[0002] Bucket wheel excavators are key pieces of equipment in bulk material storage yards such as ports and thermal power plants. During stacking and reclaiming operations, the bucket wheel at the front end of its cantilever needs to approach the dynamically changing stockpile at close range in a complex three-dimensional space. To ensure operational safety and prevent collisions with foreign objects mixed in with the stockpile or unexpected intrusions, real-time and precise obstacle avoidance capabilities are crucial.

[0003] In existing technologies, lidar is used for environmental perception to achieve collision avoidance. For example, Chinese Patent Publication No. CN119716901A discloses a lidar-based collision avoidance method and system for bucket wheel excavators. This method calibrates the lidar in the stockpile coordinate system, acquires a three-dimensional point cloud of the collision avoidance area, and determines whether obstacles exist. If obstacles exist, the minimum straight-line distance between the obstacle and the protected area is calculated and converted into a control signal. This provides intelligent collision avoidance protection for bucket wheel excavator operations based on lidar.

[0004] However, such existing technologies still have obvious shortcomings when dealing with actual operating scenarios of bucket wheel excavators. Specifically: 1. Existing technologies usually regard all point clouds obtained by scanning as homogeneous obstacle detection targets, and cannot structurally separate the working ground, the dynamically changing material pile surface and independent foreign objects in three-dimensional space, which can easily lead to false alarms or missed alarms.

[0005] 2. Existing obstacle detection technologies mostly rely on fixed distances or simple reflection intensity thresholds, and have not constructed a multi-dimensional judgment logic that integrates geometry, reflection intensity, spatial topology, and temporal motion consistency. They also lack effective filtering and verification mechanisms, resulting in insufficient obstacle avoidance robustness of bucket wheel excavators. Summary of the Invention

[0006] To overcome the shortcomings in the prior art, this invention provides a lidar system for obstacle avoidance in bucket wheel excavators, which can effectively solve the problems mentioned in the prior art.

[0007] The objective of this invention can be achieved through the following technical solution: a lidar system for obstacle avoidance of a bucket wheel excavator, comprising: a ground point cloud recognition module, a work material pile recognition module, a candidate target clustering module, an obstacle attribute determination module, and a safety control module.

[0008] The ground point cloud recognition module is connected to the work material pile recognition module, the work material pile recognition module is connected to the candidate target clustering module, the candidate target clustering module is connected to the obstacle attribute determination module, and the obstacle attribute determination module is connected to the safety control module.

[0009] The ground point cloud recognition module performs height distribution gradient analysis on the point cloud data of the bucket wheel excavator's operating area and dynamically identifies the ground point cloud within it.

[0010] The working material pile identification module predicts the envelope region of the material pile based on the real-time cantilever attitude parameters of the bucket wheel excavator. Within the envelope region, the module identifies the point cloud on the surface of the current material pile of the bucket wheel excavator and determines its reflection intensity characteristics by determining the continuity of the curved surface and defining the material repose angle.

[0011] The candidate target clustering module removes ground point clouds and current stockpile surface point clouds from the full-frame point cloud, and performs three-dimensional clustering on the remaining point clouds to form candidate target clusters.

[0012] The obstacle attribute determination module measures the difference in reflection intensity and calculates the relative spatial relationship between each candidate target cluster and the current point cloud on the surface of the material pile. It then identifies the real obstacle target in the candidate target cluster through multi-dimensional logical determination.

[0013] The safety control module performs kinematic collision detection on real obstacle targets and generates avoidance, deceleration, or stopping commands for the bucket wheel excavator based on the collision level.

[0014] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention dynamically identifies ground point clouds through height distribution gradient analysis, predicts the envelope region based on cantilever attitude, and defines the current material pile surface point cloud by combining surface continuity and material repose angle. The ground, dynamic material pile and residual point cloud in the current operation scenario are structurally separated, and an adaptive safety background is established for subsequent obstacle detection to avoid the problem of misreporting the material pile outline as an obstacle.

[0015] (2) The present invention performs three-level judgments on candidate target clusters in sequence: homogeneity of reflection intensity, local geometric change and spatial connection relationship. At the same time, it performs time-series tracking of the target and ensures that alarms are only issued for real and stable obstacles by analyzing the consistency of its displacement with the kinematics of the bucket wheel machine and its morphological stability. Attached Figure Description

[0016] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the module connection of the present invention.

[0018] Figure 2 This is a schematic diagram of the logic flow for identifying the point cloud on the surface of the current material pile of the bucket wheel excavator according to the present invention.

[0019] Figure 3This is a schematic diagram of the logic flow for kinematic conflict detection in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 As shown, the present invention provides a lidar system for obstacle avoidance of bucket wheel excavators, including: a ground point cloud recognition module, a work material pile recognition module, a candidate target clustering module, an obstacle attribute determination module, and a safety control module.

[0022] The ground point cloud recognition module is connected to the work material pile recognition module, the work material pile recognition module is connected to the candidate target clustering module, the candidate target clustering module is connected to the obstacle attribute determination module, and the obstacle attribute determination module is connected to the safety control module.

[0023] Given that the 3D point cloud of the bucket wheel excavator's operating area contains ground, dynamic material piles, and potential foreign objects, this invention begins with an understanding of the operating environment. To establish a stable spatial reference, the ground point cloud recognition module performs height distribution gradient analysis on the point cloud data of the bucket wheel excavator's operating area, dynamically identifying the ground point cloud within it. The specific implementation process is as follows: Through coordinate transformation, the point cloud data of the bucket wheel excavator's operating area is unified to the bucket wheel excavator's body operating coordinate system. The bucket wheel excavator's body operating coordinate system has the bucket wheel excavator's rotation center as its origin, where the X-axis points to the horizontal extension direction when the cantilever is at a zero-degree rotation angle, the Y-axis is perpendicular to the X-axis in the horizontal plane, and the Z-axis is vertically upward.

[0024] On the horizontal plane of the bucket wheel excavator's working coordinate system, the point cloud data is divided into equally spaced grids. The arithmetic mean of the elevation of all point cloud data in each grid is calculated as the average height of the grid. All grids are then sorted in ascending order based on the average height to generate a grid sequence.

[0025] Calculate the absolute difference between the average height of each grid cell in the grid sequence and the median of the grid sequence. Then, integrate and calculate the median of all absolute differences, and record it as the median absolute difference. Set twice the median absolute difference as the height tolerance limit.

[0026] Initialize an empty ground grid set. Perform region growth starting from the first grid cell, checking the four-connected or eight-connected grid cells adjacent to the first grid cell. If the difference between the average height of the adjacent grid cells and the average height of the grid cells in the current ground grid set is less than a set height tolerance limit, then merge the adjacent grid cells into the ground grid set. Iterate through the region growth process until no new grid cells meet the merging condition.

[0027] Remove the raster grids that have been incorporated into the ground raster set from the grid sequence, and repeat the region growth operation until the grid sequence is empty or the preset number of iterations is reached, finally obtaining the complete ground raster set.

[0028] For a point cloud in a ground grid, principal component analysis is used to calculate the local surface normal vector of each point cloud in its local neighborhood. The specific process of principal component analysis is as follows: select the k points that are closest to the point cloud to be calculated in Euclidean distance to form the local neighborhood point set of the point cloud to be calculated. The parameter k is preset according to the point cloud density, for example, the value range is 15 to 50.

[0029] Calculate the arithmetic mean of the three-dimensional coordinates of all point clouds within the local neighborhood point set to obtain the coordinates of the neighborhood centroid.

[0030] Using centroid coordinates, a 3×3 covariance matrix is ​​constructed to characterize the spatial distribution of the neighborhood point set. The covariance matrix is ​​then decomposed into eigenvalues, and the unit eigenvector corresponding to the smallest eigenvalue is selected as the local surface normal vector of the point cloud to be calculated.

[0031] In addition, to ensure that the normal vectors of the entire ground point cloud are consistent, global direction consistency processing is required. The dot product of the local surface normal vector and the globally specified direction (such as the vertical upward unit vector) is calculated; if the dot product is less than 0, the original local surface normal vector is inverted.

[0032] Based on the spatial distribution of all point clouds within the ground grid set, an initial terrain reference plane is fitted, and the average slope angle of the terrain reference plane is calculated.

[0033] The sum of the average slope angle and the fixed safety margin angle is defined as the deviation limit of the normal vector, where the fixed safety margin angle ranges from [5° to 15°], and can be exemplarily set to 10°.

[0034] Calculate the angle between the local surface normal vector of each point cloud in the ground grid set and the absolute vertical direction (0, 0, 1). If the angle is greater than the deviation limit of the normal vector, it is determined that it does not belong to the ground point and is removed from the candidate points of the ground grid set. The point cloud retained after filtering the ground grid set is taken as the ground point cloud.

[0035] A ground reference plane is fitted to the ground point cloud using the least squares method, and its plane equation is obtained. The plane equation obtained from this fitting is used as a new reference for processing newly acquired point cloud data, realizing iterative updating and optimization of the reference plane, thereby dynamically identifying the ground point cloud within it.

[0036] After acquiring the ground point cloud, the point cloud data still contains dynamically changing material pile surfaces and potential obstacles. The next goal of the operational material pile identification module is to accurately separate the surface of the currently operational material pile from the remaining point cloud. This process faces two challenges: the shape of the material pile changes in real time with the operation, and its surface must conform to physical laws such as the angle of repose of the material. The operational material pile identification module addresses these challenges through the following two collaborative steps: S1. Predicting the envelope region of the material pile based on the real-time cantilever attitude parameters of the bucket wheel excavator. Specifically, this involves acquiring the real-time pitch and rotation angles of the bucket wheel excavator cantilever, where the pitch angle is defined as the angle between the cantilever axis and the horizontal plane, and the rotation angle is defined as the angle between the projection of the cantilever axis onto the horizontal plane and the X-axis.

[0037] Based on the fixed length of the cantilever mechanism, the real-time position coordinates of the bucket wheel center in the bucket wheel excavator's working coordinate system are calculated using the following forward kinematics formula. : .

[0038] In the formula, These represent the real-time pitch angle and real-time slewing angle of the bucket wheel excavator boom, respectively. This indicates the length of the cantilever fixed machine.

[0039] This represents the projected length of the cantilever fixed mechanical length onto the horizontal plane. As the cantilever pitch angle changes, the length of the shadow cast on the horizontal plane by the line connecting the bucket wheel center to the slewing center also changes.

[0040] This indicates that the horizontal projection length is then projected onto the X-axis, which is used to determine the position of the bucket wheel center in the X-axis direction.

[0041] This means that the horizontal projection length is then projected onto the Y-axis to determine the position of the bucket wheel center in the Y-axis direction.

[0042] This represents the projected height of the cantilever fixed mechanical length in the vertical direction.

[0043] This forward kinematics formula is used for real-time positioning of the bucket wheel center. Essentially, it is a converter for the bucket wheel machine's working coordinate system, transforming easily measurable angular quantities into coordinate quantities that are easier to calculate and control in space.

[0044] Based on the actual physical radius and thickness of the bucket wheel, an instantaneous three-dimensional geometric envelope of the bucket wheel centered on the real-time position coordinates is constructed. This three-dimensional geometric envelope can be exemplarily a cylindrical shape, with the axis of the cylinder parallel to the current cantilever axis, the radius of the cylinder being the actual physical radius of the bucket wheel, and the height being the thickness of the bucket wheel.

[0045] The material extraction depth is determined based on the material extraction process parameters. The instantaneous three-dimensional geometric envelope of the bucket wheel is translated to the material extraction depth in the opposite direction of the cantilever pitch direction (i.e., the direction in which the cantilever axis points to the inside of the material pile) to form an extended envelope. This operation simulates the theoretical range occupied in space by the mechanical structure and cutting part of the bucket wheel when it is performing material extraction operations at the current position.

[0046] Obtain the planned range of motion of the bucket wheel excavator in the current working cycle, including the rotation and pitch angles. With the center of the bucket wheel excavator as the axis, rotate and sweep the extended envelope around the Z-axis on the horizontal plane from the planned range of motion of the rotation angle in the current working cycle. This sweeping operation generates a fan-shaped columnar body in three-dimensional space and defines it as the horizontal predicted boundary.

[0047] Calculate the extreme Z-coordinates reached by the highest and lowest points of the extended envelope when the cantilever pitch angle varies within the planned range of pitch angle motion, denoted as follows: and .

[0048] Measure the projected profile of the extended envelope on the horizontal plane, determine the maximum horizontal radius from its outer boundary to the center of rotation of the bucket wheel excavator, consider the inherent natural angle of repose of the material pile due to gravity, and multiply the tangent value of the inherent natural angle of repose by the maximum horizontal radius to obtain the height margin. , by interval Define the vertical prediction boundary. It should be noted that the inherent natural angle of repose is a known physical property parameter of the material, stored in the material property database of the bucket wheel excavator control system, and retrieved according to the current material type being processed.

[0049] The intersection of the horizontal and vertical prediction boundaries in three-dimensional space is used to define the three-dimensional spatial region of the overlapping portion, which is the final predicted material pile envelope region. This material pile envelope region is a conservative spatial estimate of the portion of the material pile that the bucket wheel excavator may contact during the current operating cycle.

[0050] Reference Figure 2 As shown, S2. Within the envelope region, the point cloud on the surface of the current material pile of the bucket wheel is identified by determining the continuity of the surface and defining the material's angle of repose. Specifically, the local surface normal vector of each point cloud within the envelope region of the material pile is calculated by principal component analysis.

[0051] Based on the inherent angle of repose physical parameters of the materials processed by the bucket wheel excavator, a normal vector screening range is set, and point clouds whose angle between the local surface normal vector and the vertical direction is within the normal vector screening range are marked as points within the angle of repose.

[0052] Calculate the ratio of the local point cloud density of each point within the repose angle to its adjacent points, and logically compare it with the overall coefficient of variation of the local point cloud density of all points within the repose angle. Specifically, calculate the difference between the local point cloud density ratio and 1. If this difference is less than the overall coefficient of variation, it is determined that the local point cloud density of the point within the repose angle and its adjacent points are statistically similar and belong to the same continuous surface, and the clusters of the two points are identified.

[0053] Identify and merge pairs of points within the angle of repose belonging to the same continuous surface, and traverse all points within the angle of repose to generate an initial surface point cloud cluster.

[0054] Perform surface fitting on the initial surface point cloud cluster, calculate the set of distances from all points in the cluster to the fitted surface. If the kurtosis and skewness of the distance set are both less than the preset constants of the normal distribution statistical characteristics, for example, the kurtosis is less than 2 and the skewness is less than 1, then the initial surface point cloud cluster is taken as the current material pile surface point cloud.

[0055] Otherwise, distance-based clustering is used to split the initial surface point cloud cluster into multiple sub-clusters, and surface fitting is performed again. Sub-clusters that satisfy the statistical characteristics of normal distribution and are interconnected are retained as the current surface point cloud of the material pile.

[0056] It should be noted that the execution process of the distance-based clustering method is as follows: First, calculate the average nearest neighbor distance between point clouds within the initial surface point cloud cluster, and multiply it by a preset coefficient (such as 1.5) to obtain the neighborhood radius. Multiply the total number of point clouds within the cluster by a preset ratio (a constant greater than 0 and less than 0.5) and round down to obtain the minimum number of points.

[0057] Next, traverse each point within the cluster, construct a spherical neighborhood centered on that point and with the radius of its neighborhood as the radius, and count the number of points within that neighborhood. If the number of points is greater than or equal to the minimum number of points, mark it as a core point. If the number of points is less than the minimum number of points but falls within the spherical neighborhood of any core point, mark it as a boundary point. Otherwise, mark it as a noise point and discard it.

[0058] Next, initialize the sub-cluster list and the unvisited point set. Select a core point from the unvisited point set as a seed, create a new sub-cluster, and add all points in its neighborhood (core point and boundary points) to the sub-cluster and mark them as visited.

[0059] For each core point newly added to the sub-cluster, recursively add all unvisited points in its neighborhood to the current sub-cluster until it can no longer be expanded.

[0060] After constructing a subcluster, the next core point is searched for as a seed in the set of unvisited points, and the above process is repeated until all core points have been visited and processed.

[0061] Ultimately, the initial cluster is split into multiple density-connected subsets of the largest points, which are the desired subclusters.

[0062] After identifying the point cloud on the surface of the current stockpile, it is necessary to further determine its reflection intensity characteristics. The reflection intensity characteristics include the reference reflection intensity and the upper limit of the homogeneity allowable deviation, so as to provide the basis for analysis data for the subsequent obstacle attribute determination module. Specifically, the grid corresponding to the point cloud on the surface of the current stockpile is calibrated, the median of the point cloud reflection intensity in each calibrated grid is calculated, and then the second median is taken from the medians of all calibrated grids, and the second median is established as the reference reflection intensity.

[0063] Calculate the standard deviation of the median reflection intensity of the point cloud relative to the reference reflection intensity within each calibration grid. Based on the number of calibration grids and the preset confidence level (95%), obtain the corresponding statistical critical value by querying the t-distribution table, which is used as the statistical confidence level of the median. Multiply it by the standard deviation to obtain the upper limit of the homogeneity allowable deviation.

[0064] After accurately removing the ground point cloud and the dynamic material pile surface point cloud, the remaining point cloud contains both potential real obstacles and discrete noise and local interference. To organize the disordered and scattered point cloud into candidate targets with independent spatial meaning for subsequent attribute analysis and judgment, the candidate target clustering module removes the ground point cloud and the current material pile surface point cloud from the full-frame point cloud and performs three-dimensional clustering on the remaining point cloud to form candidate target clusters.

[0065] Specifically, a spatial index data structure, such as an octree, is established for the union of the ground point cloud and the current material pile surface point cloud.

[0066] For the spatially aligned full-frame point cloud set, perform a nearest neighbor search in the spatial indexing process described above. If the distance between the full-frame point cloud and its nearest neighbor in the union set is less than the spatial sampling resolution, the point is considered to belong to the ground or the surface of the material pile and is excluded from the set to be output. The point cloud in the set after the exclusion is taken as the residual point cloud.

[0067] Noise filtering is performed on the residual point cloud. A density-based clustering method is used to aggregate several independent candidate target clusters. A geometric descriptor containing centroid coordinates and bounding boxes is established for each target cluster. The centroid coordinates specifically refer to the arithmetic mean of the coordinate components of all points in the target cluster. The bounding box is specifically obtained by traversing all points in the target cluster and finding the maximum and minimum values ​​of each point on the X, Y, and Z coordinate axes. The cuboid defined by these six extreme values ​​is the bounding box of the target cluster.

[0068] It should be explained that the density-based clustering method and the distance-based clustering method mentioned above are logically consistent, both belonging to spatial clustering applications. Their process is still to traverse all points, identify core points, boundary points, and noise points, and merge density-connected core points and their boundary points into the same candidate target cluster. The difference lies in the setting of the neighborhood radius and the minimum number of points. The neighborhood radius can be set to 1.5 to 3.0 times the average point spacing of the residual point cloud. The value of the minimum number of points is related to the overall density of the point cloud and the minimum target size to be detected, and can be set to 0.5% to 2% of the total number of residual point clouds, and not less than 3.

[0069] To ensure the feasibility of this method in the specific scenario of this invention, the neighborhood radius needs to be adapted. The execution process of the density-based clustering method will not be described in detail here.

[0070] After removing obstacles from the ground and material pile surfaces, the remaining point cloud may contain real obstacles, but it also contains noise. Therefore, the obstacle attribute determination module measures the difference in reflection intensity and calculates the relative spatial relationship between each candidate target cluster and the current material pile surface point cloud. Through multi-dimensional logical judgment, it identifies the real obstacle targets within the candidate target clusters.

[0071] Specifically, the average reflection intensity of the candidate target cluster is subtracted from the reference reflection intensity of the current point cloud on the surface of the stockpile. If the absolute value of the difference is greater than the upper limit of the homogeneity allowable deviation, it is marked as a first-level target cluster.

[0072] The point cloud with the minimum Z-coordinate in the primary target cluster is selected as the lowest point of the cluster. The mean Z-coordinate of all point clouds within a circular area of ​​a set radius centered on the lowest point of the cluster is calculated as the reference height of the fitted material pile surface around the lowest point of the cluster. The difference between the Z-coordinate of the lowest point of the cluster and the reference height is taken as the relative height difference between the primary target cluster and the fitted material pile surface around it.

[0073] For a primary target cluster, extract all points in its point cloud that are adjacent to the current stockpile surface point cloud on the horizontal projection, forming a boundary point set. For each point in the boundary point set, utilize the local surface normal vector of its neighboring point clouds within the primary target cluster. Calculate the complementary angle between the local surface normal vector and the vertical direction as the local slope angle. Calculate the average local slope angle of all points in the boundary point set as the local slope of the boundary point cloud.

[0074] If the relative height difference and local slope meet the local geometric change detection conditions, it is marked as a secondary target cluster. The specific local geometric change detection conditions are: the relative height difference is greater than the spatial resolution of the lidar point cloud, and the local slope of the boundary point cloud is greater than the inherent angle of repose of the material being processed by the bucket wheel excavator.

[0075] Calculate the nearest neighbor distance set between the secondary target cluster and the current material pile surface point cloud. If the median of the distance set is greater than the average spacing of the point cloud, or the standard deviation is greater than a preset multiple of the standard deviation of the spacing of the current material pile surface point cloud itself, then it is determined to be an obstacle target.

[0076] Data association tracking of real obstacle targets in multiple consecutive frames is performed based on the 3D centroid and the bounding box.

[0077] Calculate the absolute displacement vector of the centroid of the tracked target between adjacent frames, and set dynamic motion constraints based on the real-time attitude parameters of the bucket wheel excavator boom: obtain the changes in the slewing angle, pitch angle, and trolley travel displacement of the bucket wheel excavator boom between two adjacent frames. From the previous frame to the current frame, the boom first rotates around the Z-axis to obtain the change in the slewing angle, and then rotates around the Y-axis to obtain the change in the pitch angle, thus obtaining the rotation matrix from the previous frame to the current frame. Represent the change in the trolley travel displacement as a column vector in the global coordinate system. Based on the attitude rotation matrix of the previous frame coordinate system relative to the global coordinate system, transform the column vector to the previous frame coordinate system.

[0078] Combine the rotation matrix from the previous frame to the current frame and the column vector transformed to the coordinate system of the previous frame. The rigid body transformation matrix of the bucket wheel excavator's working coordinate system from the previous frame to the current frame is obtained. , which is represented as , This is the transpose of the rotation matrix from the previous frame to the current frame. It is a zero vector with 1 row and 3 columns.

[0079] Using the rigid body transformation matrix, calculate the expected centroid coordinates of the target in the current frame coordinate system based on the centroid coordinates of the target in the previous frame. Use the difference between the actual centroid coordinates and the expected centroid coordinates of the target in the current frame as the expected displacement vector.

[0080] The deviation vector is obtained by subtracting the expected displacement vector from the absolute displacement vector, and is decomposed into a longitudinal deviation component along the current main direction of movement of the bucket wheel machine and a vertical deviation component perpendicular to the main direction of movement.

[0081] For the currently tracked real obstacle target, cache its longitudinal deviation components and bucket wheel machine speed scalar of the most recent at least 5 consecutive frames. The bucket wheel machine speed scalar is calculated from the composite vector magnitude of the bucket wheel machine cantilever rotation angular velocity, pitch angular velocity and trolley travel speed.

[0082] Calculate the Pearson correlation coefficient between the longitudinal deviation component sequence and the bucket wheel excavator speed sequence. If the Pearson correlation coefficient is greater than 0.7, it is determined that there is a correlation between the longitudinal deviation component and the bucket wheel excavator speed; otherwise, it is determined that there is no correlation.

[0083] When the longitudinal deviation component is correlated with the speed of the bucket wheel excavator, and the vertical deviation component is smaller than the longitudinal deviation component, it indicates that the motion constraint requirements are met.

[0084] Therefore, if the target centroid displacement vector meets the motion constraints for multiple sampling periods, the target motion state is determined to be reasonable.

[0085] The temporal rate of change of the target point cloud volume is calculated. If the temporal rate of change is consistently lower than the stability threshold determined by the point cloud registration error statistics, the target morphology is considered stable. The stability threshold is determined as follows: Under system idle and static conditions, multiple frames of point cloud data are collected and paired for registration. The rate of change of the point cloud volume after each registration is calculated, and the upper statistical limit of this rate of change sequence (e.g., the 95th percentile) is taken as the stability threshold.

[0086] Targets whose motion is deemed reasonable and whose form is stable are confirmed to be real obstacle targets.

[0087] After accurately identifying real obstacle targets, to ensure the safe operation of the bucket wheel excavator, it is necessary to quantitatively assess the potential collision risk between the obstacle and the moving parts of the equipment (mainly the bucket wheel), and drive the actuators to implement corresponding protective operations based on the assessment results. To this end, the safety control module performs kinematic collision detection on the real obstacle targets and generates avoidance, deceleration, or shutdown commands for the bucket wheel excavator based on the collision level.

[0088] Reference Figure 3 As shown, specifically, based on the current operating speed of the bucket wheel excavator Maximum deceleration and response delay parameters The kinematic braking distance of the bucket wheel excavator within a preset time period is calculated using the following formula: .

[0089] In the formula, The reaction distance represents the distance traveled at the original speed during the bucket wheel excavator's response delay.

[0090] This represents the deceleration distance, indicating the theoretical distance required to come to a stop with uniform deceleration at maximum deceleration.

[0091] This calculation formula is used to predict the minimum braking distance required for bucket wheel excavators during emergency braking or normal shutdown, providing a basis for safety protection, collision warning, and shutdown control.

[0092] Centered on the predicted bucket wheel trajectory, a dynamic safety envelope is extended outward, defined by the bucket wheel's mechanical dimensions and kinematic braking distance. This dynamic safety envelope is typically modeled as a capsule or an stretched boundary cylinder with the predicted bucket wheel trajectory as its centerline, and its cross-sectional radius is the sum of the bucket wheel's mechanical radius and the kinematic braking distance.

[0093] Calculate the shortest distance between the real obstacle target and the dynamic safety envelope in three-dimensional space.

[0094] If the shortest distance is negative or zero, it indicates an intrusion and a shutdown command will be triggered immediately.

[0095] If positive, the ratio of the shortest distance to the magnitude of the relative motion vector is calculated based on the relative motion vector between the bucket wheel excavator and the actual obstacle target, in order to obtain the expected first intrusion time of the obstacle.

[0096] The expected first intrusion time is compared with the response time required for different actions of the bucket wheel excavator, which is preset from the system's prior equipment debugging or type test. When the expected first intrusion time is greater than the long-term early warning response time, it is determined to be at the safe level, and only status monitoring is performed.

[0097] When the response time is less than or equal to the long-term warning response time, but greater than the standard avoidance operation response time of the bucket wheel excavator, it is determined to be an avoidance level and an avoidance command is generated.

[0098] When the response time is less than or equal to the standard obstacle avoidance operation time of the bucket wheel excavator, but greater than the safe braking time of the bucket wheel excavator, it is determined to be a deceleration stage and a deceleration command is generated.

[0099] When the time is less than or equal to the safe braking time of the bucket wheel excavator, it is determined to be at the shutdown level and an emergency shutdown command is generated.

[0100] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A lidar system for obstacle avoidance in bucket wheel excavators, characterized in that, include: The ground point cloud recognition module performs height distribution gradient analysis on the point cloud data of the bucket wheel excavator's operating area and dynamically identifies the ground point cloud within it. The working material pile identification module predicts the envelope region of the material pile based on the real-time cantilever attitude parameters of the bucket wheel excavator. Within the envelope region, the module identifies the point cloud on the surface of the current material pile of the bucket wheel excavator and determines its reflection intensity characteristics by determining the continuity of the curved surface and defining the material repose angle. The identification of the current point cloud on the surface of the bucket wheel excavator includes: The local surface normal vectors of each point cloud within the envelope region of the material pile are calculated using principal component analysis. Based on the inherent angle of repose physical parameters of the materials processed by the bucket wheel excavator, a normal vector screening range is set, and point clouds whose angle between the local surface normal vector and the vertical direction is within the normal vector screening range are marked as points inside the angle of repose. Calculate the ratio of the local point cloud density of each point within a repose angle to its adjacent points, and logically compare it with the overall coefficient of variation of the local point cloud density of all points within a repose angle. Identify and merge pairs of points within a repose angle belonging to the same continuous surface, and traverse all points within a repose angle to generate an initial surface point cloud cluster. Perform surface fitting on the initial surface point cloud cluster, calculate the set of distances from all points in the cluster to the fitted surface, and if the kurtosis and skewness of the distance set are both less than the preset constant of the normal distribution statistical characteristics, then the initial surface point cloud cluster is taken as the current material pile surface point cloud. Otherwise, distance-based clustering is used to split the initial surface point cloud cluster into multiple sub-clusters, and surface fitting is performed again. Sub-clusters that satisfy the statistical characteristics of normal distribution and are interconnected are retained as the current surface point cloud of the material pile. The candidate target clustering module removes ground point clouds and current stockpile surface point clouds from the full-frame point cloud, and performs three-dimensional clustering on the remaining point clouds to form candidate target clusters; The obstacle attribute determination module measures the difference in reflection intensity and calculates the relative spatial relationship between each candidate target cluster and the current point cloud on the surface of the stockpile. It then identifies the real obstacle targets in the candidate target clusters through multi-dimensional logical determination. The safety control module performs kinematic collision detection on real obstacle targets and generates avoidance, deceleration, or stopping commands for the bucket wheel excavator based on the collision level.

2. The lidar system for obstacle avoidance of a bucket wheel excavator according to claim 1, characterized in that, The height distribution gradient analysis of the point cloud data for the bucket wheel excavator's operating area includes: Unify the point cloud data of the bucket wheel excavator's operating area into the bucket wheel excavator's main operating coordinate system and perform horizontal grid division; Calculate the average height of each grid cell and sort them in ascending order; Based on the point cloud height distribution statistics, the height tolerance limit is dynamically set. Region growth is performed starting from the first grid cell. Grid cells that are adjacent to the current grid cell and whose height difference is less than the height tolerance limit are merged into the ground grid cell set. Based on the dynamic determination of the normal vector deviation limit of the terrain slope of the ground grid set, the local surface normal vector of the point cloud in the ground grid set is calculated. The point cloud with the angle between the local surface normal vector and the vertical direction less than the normal vector deviation limit is taken as the ground point cloud.

3. The lidar system for obstacle avoidance of a bucket wheel excavator according to claim 1, characterized in that, The envelope region of the material pile predicted based on the real-time cantilever attitude parameters of the bucket wheel excavator includes: Obtain the real-time pitch and slewing angles of the bucket wheel excavator boom, and calculate the real-time position coordinates of the bucket wheel center by combining the fixed mechanical length of the boom. Based on the actual physical radius and thickness of the bucket wheel, an instantaneous three-dimensional geometric envelope of the bucket wheel centered on its real-time position coordinates is constructed. The material extraction depth is determined based on the material extraction process parameters. The instantaneous three-dimensional geometric envelope of the bucket wheel is translated to the material extraction depth in the opposite direction of the cantilever pitch direction to form an extended envelope. Obtain the planned range of motion of the bucket wheel excavator in the current work cycle, including the rotation and pitch angles. Then, perform horizontal and vertical expansion on the extended envelope to generate horizontal and vertical prediction boundaries. The envelope region of the stockpile is defined by the horizontal and vertical prediction boundaries.

4. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 2, characterized in that, The reflection intensity characteristics are determined in the following way: The grid corresponding to the point cloud on the surface of the current stockpile is calibrated, the median of the point cloud reflection intensity in each calibrated grid is calculated, and then the second median of all calibrated grid medians is taken and established as the reference reflection intensity. The standard deviation of the median reflection intensity of the point cloud within each calibration grid relative to the reference reflection intensity is calculated. The median is then linearly amplified using statistical confidence to obtain the upper limit of the permissible deviation for homogeneity.

5. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 1, characterized in that, The step of performing three-dimensional clustering of the residual point cloud to form candidate target clusters includes: Boolean subtraction is used to remove the ground point cloud and the current stockpile surface point cloud from the spatially aligned full-frame point cloud set; Noise filtering is performed on the residual point cloud. A density-based clustering method is used to aggregate several independent candidate target clusters, and a geometric descriptor containing centroid coordinates and bounding boxes is established for each target cluster.

6. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 4, characterized in that, The method of identifying real obstacle targets in the candidate target cluster through multi-dimensional logical judgment includes: The average reflection intensity of the candidate target cluster is subtracted from the reference reflection intensity of the current point cloud on the surface of the stockpile. If the absolute value of the difference is greater than the upper limit of the homogeneity allowable deviation, it is marked as a first-level target cluster. Calculate the relative height difference between the primary target cluster and the surface of the surrounding fitted material pile, and analyze the local slope of the boundary point cloud between the two. If the relative height difference and local slope meet the local geometric change detection conditions, then mark it as a secondary target cluster. Calculate the nearest neighbor distance set between the secondary target cluster and the current material pile surface point cloud. If the median of the distance set is greater than the average spacing of the point cloud, or the standard deviation is greater than a preset multiple of the standard deviation of the spacing of the current material pile surface point cloud itself, then it is determined to be an obstacle target.

7. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 6, characterized in that, The method of identifying real obstacle targets in the candidate target cluster through multi-dimensional logical judgment also includes: Data association tracking of real obstacle targets in multiple consecutive frames based on 3D centroid and bounding box; Calculate the absolute displacement vector of the centroid of the tracked target between adjacent frames, set dynamic motion constraints based on the real-time attitude parameters of the bucket wheel excavator boom, and determine that the target's motion state is reasonable if the target centroid displacement vector continuously meets the motion constraints. Calculate the temporal rate of change of the target point cloud volume. If the temporal rate of change is consistently lower than the stability threshold determined by the point cloud registration error statistics, the target shape is determined to be stable. Targets whose motion is deemed reasonable and whose form is stable are confirmed to be real obstacle targets.

8. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 7, characterized in that, The dynamic motion constraints based on the real-time attitude parameters of the bucket wheel excavator boom include: Obtain the changes in the slewing angle, pitch angle, and trolley travel displacement of the bucket wheel excavator boom between two adjacent frames, and calculate the rigid body transformation matrix of the bucket wheel excavator's working coordinate system from the previous frame to the current frame. Using the rigid body transformation matrix, calculate the expected centroid coordinates of the target in the current frame coordinate system based on the centroid coordinates of the target in the previous frame. Use the difference between the actual centroid coordinates and the expected centroid coordinates of the target in the current frame as the expected displacement vector. The deviation vector is obtained by subtracting the expected displacement vector from the absolute displacement vector, and decomposed into a longitudinal deviation component along the current main direction of movement of the bucket wheel machine and a vertical deviation component perpendicular to the main direction of movement. The motion constraints require that the longitudinal deviation component is correlated with the speed of the bucket wheel excavator, and that the vertical deviation component is smaller than the longitudinal deviation component.

9. A lidar system for obstacle avoidance in a bucket wheel excavator according to claim 1, characterized in that, The kinematic collision detection of real obstacle targets includes: Based on the current operating speed, maximum deceleration, and response delay parameters of the bucket wheel excavator, calculate the kinematic braking distance of the bucket wheel excavator within a preset future time period; Centered on the predicted bucket wheel trajectory, a dynamic safety envelope is extended outward, defined by the bucket wheel's mechanical dimensions and kinematic braking distance. Calculate the shortest distance between the real obstacle target and the dynamic safety envelope in three-dimensional space; If the shortest distance is negative or zero, it indicates that an intrusion has occurred, and a shutdown command will be triggered immediately. If positive, the estimated first intrusion time of the obstacle is calculated based on the relative motion vector between the bucket wheel excavator and the actual obstacle target; Based on the comparison between the expected first intrusion time and the response time of the bucket wheel excavator under different motion states, the degree of conflict is divided into four levels: safe, avoidance, deceleration, and shutdown.

Images