Bulk material scanning method based on lidar sensor

Abstract

The present application belongs to the technical field of bulk material scanning, and relates to a bulk material scanning method based on a laser radar sensor, which discards the inefficient mode of continuous full-power operation of all laser radar sensors in the prior art, sets up a master scanning node for global monitoring, and places a subordinate scanning node in a low-power mode, registers point cloud data collected by the master scanning node in adjacent scanning periods, generates a change heat map and retrieves a high-heat change area in the map through voxelization difference calculation and spatial clustering, considers the geometric relationship between the sensor and the area and the complementary nature of the multi-sensor view angle for the high-heat change area, selects a sensor subset from all subordinate scanning nodes to perform cooperative enhanced scanning, fuses high-density local point clouds obtained by the enhanced scanning with global point clouds of the master scanning node to generate a three-dimensional surface model of the bulk material pile, and realizes dynamic on-demand scheduling of scanning resources and energy efficiency optimization.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of bulk material scanning, and relates to a bulk material scanning method based on a laser radar sensor. BACKGROUND

[0002] In the storage and logistics management of bulk materials, it is crucial to accurately and efficiently obtain the three-dimensional form and volume information thereof. Laser radar scanning technology has become a mainstream technical solution for realizing the digital inventory of bulk materials due to its high precision and non-contact characteristics.

[0003] Currently, in order to achieve comprehensive coverage of a large range of material piles, the industry generally adopts a fixed multi-station scanning network, that is, multiple laser radar sensors are fixedly installed around the material warehouse or yard. These sensors usually work continuously all day long at a preset fixed frequency and power, uninterruptedly collecting point cloud data to reconstruct a three-dimensional model of the material pile and then calculate the volume.

[0004] However, this mode has obvious defects. In actual production scenarios, the stacking and taking operations of materials are often intermittent, and the material pile is in a static or slowly changing state for most of the time. The continuous full-power scanning of the fixed network generates a large amount of redundant data, which not only increases the burden of data transmission and processing, but also causes unnecessary wear and tear of the sensors. More importantly, when the stacking and taking operations are performed, especially when the material is quickly taken, the local material pile may form steep sections or depressions. At this time, the fixed scanning frequency and resource allocation mode are difficult to capture the geometric details of these rapidly changing areas in a timely and focused manner, resulting in distortion of the local model and seriously affecting the calculation accuracy of the volume change in this area, which cannot meet the demand for high-precision measurement in a dynamic operation environment. SUMMARY

[0005] In view of this, in order to solve the problems raised in the background art, a bulk material scanning method based on a laser radar sensor is proposed.

[0006] The purpose of the present application can be achieved by the following technical solutions: The present application provides a bulk material scanning method based on a laser radar sensor, comprising: S1, in a plurality of laser radar sensors deployed around a bulk material pile, a master scanning node is specified to continuously monitor the bulk material pile, and the remaining sensors are set as slave scanning nodes and placed in a low-power mode.

[0007] S2, registering the point cloud data collected by the master scanning node in adjacent scanning periods, identifying the dynamic changes of the surface of the bulk material pile through voxelization difference calculation, generating a change heat map for representing the form change intensity of different spatial regions in unit time through spatial clustering, and retrieving the high-heat change region in the map.

[0008] S3, for the high heat change area, according to the geometric relationship between the sensor and the area and the complementary view angle of the multi-sensor, a sensor subset is selected from all slave scanning nodes.

[0009] S4, the sensor subset is activated and the scanning angle resolution is improved, and the cooperative enhanced scanning is performed on the high heat change area.

[0010] S5, the high-density local point cloud obtained by the enhanced scanning is fused with the global point cloud of the main scanning node to generate a three-dimensional surface model of the bulk material pile.

[0011] Compared with the prior art, the beneficial effects of the present application are as follows: (1) the present application realizes dynamic on-demand scheduling and energy efficiency optimization of scanning resources, discards the inefficient mode of all laser radar sensors working at full power in the prior art, and sets up a main scanning node for global monitoring, while the slave scanning nodes are placed in a low-power mode. Only when the high heat change area caused by the pile and taking material operation is identified, the optimal slave scanning node subset is dynamically activated and scheduled for focused scanning. This mechanism fundamentally avoids generating redundant data and unnecessary energy consumption during material static or slow change period, significantly reduces the scanning data processing load, and prolongs the overall service life of the sensor network.

[0012] (2) The present application improves the scanning accuracy and model fidelity of the key area under dynamic operation. By introducing a change heat map to quantitatively perceive and locate the surface changes of the material pile, the area of rapid change such as steep section, pit and other forms formed by taking material can be accurately identified. Then, for the high heat change area, multiple slave scanning nodes are dispatched for cooperative enhanced scanning with higher angle resolution, ensuring that the geometric details of the operation disturbance area are captured in time, completely and with high precision, effectively solving the problem of local model distortion caused by fixed scanning frequency.

[0013] (3) The present application constructs an intelligent sensor cooperative strategy based on geometric constraints and coverage optimization. When scheduling slave scanning nodes to form a sensor subset, not only the geometric relationship between a single node and the target area is considered, but also the complementary view angle between multiple sensors is emphasized. Through redundant node identification and elimination based on the intersection of coverage range, and fine-tuning optimization of the attitude of part of the redundant nodes without affecting full coverage, the finally selected sensor subset can realize efficient and non-missing coverage of all high heat change areas with the least number of nodes and the optimal combination of observation view angles. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed for the description of the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0015] Figure 1 The method of the present application is implemented by the step flowchart.

[0016] Figure 2 The process flowchart is specified for the main scanning node of the present application.

[0017] Figure 3 The point cloud data registration flowchart of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Please refer to Figure 1 As shown in the drawings, the present application provides a bulk material scanning method based on a laser radar sensor, which comprises the following steps: S1, among a plurality of laser radar sensors deployed around a bulk material pile, a main scanning node is specified to continuously monitor the bulk material pile, and the remaining sensors are set as slave scanning nodes and placed in a low-power consumption mode.

[0020] Among them, the low-power consumption mode refers to a mode in which the slave scanning node is controlled to work at a scanning frequency lower than that of the main scanning node or at a scanning angle resolution lower than that of the main scanning node. The specific parameters can be set by the implementers according to the actual energy consumption and monitoring requirements.

[0021] Please refer to Figure 2As shown, the designation process of the main scanning node is based on the theoretical observation integrity of the bulk material pile by each laser radar sensor, and specifically includes the following steps: obtaining the fixed installation position and spatial attitude parameters of all laser radar sensors in the scanning network, combining the pre-input typical spatial profile model of the bulk material pile, performing line-of-sight simulation analysis, which can be completed in software or environment with three-dimensional spatial analysis function, such as in the development interface of computer-aided design software or professional point cloud processing software, by calling the view cone analysis algorithm. The specific process is: regarding the installation position of each laser radar sensor as a view point, constructing its three-dimensional observation view cone according to its calibrated horizontal and vertical field of view angle parameters, obtaining the visible area of the laser radar sensor on the model surface by calculating the intersection part of the view cone and the surface model of the typical spatial profile model, and taking the ratio of the visible area to the total volume of the model surface as the visible area ratio to obtain the visible area ratio of each laser radar sensor on the surface of the typical spatial profile model.

[0022] According to the visible area ratio, if there is a sensor whose visible area ratio on the surface of the typical spatial profile model reaches 100%, i.e., it can independently realize non-blind area coverage, it is directly designated as the main scanning node.

[0023] If there is no sensor that can independently realize full coverage, a group of sensors need to be selected as the main scanning node through iterative construction to realize complete observation of the material pile. The iterative construction process is as follows: a. Select the laser radar sensor with the highest visible area ratio as the first main scanning node.

[0024] b. Calibrate the blind area of the typical spatial profile model surface corresponding to the first main scanning node, and select the sensors that have observation angles for the blind area from the remaining sensors. Here, the selection condition is: if the visible area of a laser radar sensor on the current blind area of the typical spatial profile model surface is greater than zero in the line-of-sight simulation analysis, it is determined that it has an observation angle. Among the sensors that meet the condition, select the sensor with the largest visible area for the blind area and supplement it as a new main scanning node.

[0025] c. Repeat steps a and b until the blind area of the typical spatial profile model surface is eliminated.

[0026] In the designation process, if multiple sensors that meet the current designation conditions and are in parallel, such as having the same visible area proportion or the same contribution to reducing the blind area, appear, the sensor with higher theoretical average point cloud density is preferentially selected, and the technical principle is that: even for the same type of lidar sensor, due to the different layout distances in the scanning network and the different laser beam incident angles relative to the points on the surface of the material pile, there is a theoretical difference in the number of effective measurement points that can be obtained per unit surface area. Under the premise of meeting the basic coverage requirement, higher theoretical point cloud density means that more detailed surface information can be provided in subsequent modeling, which helps to generate more accurate three-dimensional models and achieve high-precision scanning.

[0027] The theoretical average point cloud density can be completed simultaneously in the aforementioned line-of-sight simulation analysis, specifically: obtaining the inherent parameters of the lidar sensor to be evaluated, including its horizontal angle resolution and vertical angle resolution, which refers to the number of laser beams that the sensor can emit per unit angle.

[0028] In the line-of-sight simulation analysis, when it is determined that a certain laser beam emitted by the sensor intersects with a certain triangular facet of the typical spatial profile model, the intersection point is recorded, and the geometric parameters of the point are calculated: the straight-line distance between the intersection point and the optical center of the sensor is calculated as the measurement distance.

[0029] The angle between the emission direction vector of the laser beam and the normal vector of the triangular facet is calculated as the laser incidence angle.

[0030] For each intersection point, the local theoretical point cloud density is estimated based on the obtained geometric parameters. The local theoretical point cloud density is proportional to the angle resolution of the sensor, inversely proportional to the square of the measurement distance, and proportional to the cosine value of the laser incidence angle. The angle resolution, the reciprocal of the square of the measurement distance, and the cosine value of the laser incidence angle can be multiplied to obtain the estimated local theoretical point cloud density, and the calculation process is dimensionless multiplication.

[0031] The local theoretical point cloud densities of all intersection points are arithmetically averaged to obtain the theoretical average point cloud density of the sensor for the typical spatial profile model.

[0032] S2, registering the point cloud data collected by the main scanning node in adjacent scanning periods, identifying the dynamic changes of the surface of the bulk material pile through voxelization difference calculation, generating a change heat map for representing the morphological change intensity of different spatial regions per unit time through spatial clustering, and retrieving high-heat change regions in the map.

[0033] The registration of the point cloud data aims to eliminate the spatial pose difference between the two-phase point clouds caused by the slight vibration of the sensor or the deviation of the scanning starting angle, so that they are in the same coordinate system for comparison. Referring to Figure 3 As shown, the process specifically includes: defining the two-phase point cloud data of adjacent scanning periods as source point cloud and target point cloud respectively, and constructing the spatial index of the target point cloud.

[0034] For each point in the source point cloud, the spatial index is used to find its spatial nearest neighbor point in the target point cloud, which specifically refers to the point in the target point cloud with the shortest spatial position Euclidean distance compared with the source point cloud, forming a temporary point pair correspondence.

[0035] The normal vector included angle of the two points in all temporary corresponding point pairs is counted, and a normal vector included angle sequence is generated in ascending order to select the high percentile as a calibration threshold to remove the outliers that obviously deviate from the main distribution, for example, the 90th percentile. If the normal vector included angle of the two points in a temporary corresponding point pair exceeds the calibration threshold, it is considered as a point pair that does not meet the geometric consistency and is removed, and an optimized corresponding point pair set is obtained.

[0036] According to the optimized corresponding point pair set, a rigid transformation matrix is solved by least square optimization, which aims to find a rotation matrix and a translation vector, so that the Euclidean distance square sum between the corresponding points in the transformed source point cloud and the target point cloud is minimized. The solution steps are as follows: the corresponding point set in the source point cloud and the corresponding point set in the target point cloud are extracted from the optimized corresponding point pair set respectively, the centroids of the two point sets are calculated, the covariance matrix composed of the coordinates of the corresponding points after decentralization is calculated, and the singular value decomposition of the covariance matrix is performed to obtain the optimal rotation matrix. The optimal translation vector is obtained by subtracting the product of the optimal rotation matrix and the centroid coordinates of the source point cloud from the centroid coordinates of the target point cloud. The final rigid transformation matrix is usually represented as a 4×4 matrix in homogeneous coordinates, which is expressed as: .

[0037] Wherein, is the optimal rotation matrix, which is a 3×3 orthogonal matrix, representing the rotation transformation, is the optimal translation vector, which is a 3×1 column vector, representing the translation transformation. Multiplying the matrix by the homogeneous coordinates of a source point cloud can obtain its coordinates in the target coordinate system.

[0038] In addition, the least square optimization to solve the rigid transformation matrix is a prior art, and the implementation details are not described.

[0039] The spatial position of the source point cloud is updated by applying the rigid transformation matrix, and the average distance of the transformed optimized corresponding point pair set is calculated synchronously as the registration residual of this iteration.

[0040] Taking the source point cloud after previous iteration transformation as new input, iteratively performing the process of finding corresponding points and calculating rigid transformation matrix until satisfying the termination condition, outputting the registered point cloud data of two periods in the same coordinate system.

[0041] The termination condition includes: the variation of transformation matrix parameters calculated by two consecutive iterations is less than the preset convergence threshold, or the iteration number reaches the preset maximum value.

[0042] After obtaining the registered point cloud, the dynamic change of the bulk material pile surface is identified by voxelization difference calculation, including: determining the space bounding box of the two period point cloud data in the target coordinate system, calculating the maximum and minimum coordinate values of the bounding box in X, Y and Z directions, and uniformly dividing each axis coordinate range according to the predefined spatial resolution, so as to divide the coordinate system in which the two period point cloud data is registered into a three-dimensional voxel grid.

[0043] The effective point number and point cloud centroid position of each voxel in two scanning periods are counted, and the effective point number difference and point cloud centroid displacement of the same voxel between two scanning periods are calculated.

[0044] In order to distinguish between real material change and measurement noise, the permissible threshold value needs to be set for the two change values. Specifically, the permissible threshold value of the point cloud centroid displacement is mainly set according to the alignment error introduced in the registration process, and can usually take a preset multiple of the registration residual of the last round of iteration, for example, 2 times or 3 times, to tolerate the position fluctuation caused by imperfect registration.

[0045] The permissible threshold value of the effective point number difference needs to consider the natural fluctuation of point cloud density caused by scanning distance, incident angle, etc., and can take the standard deviation of the effective point number of each voxel in the historical scanning period, and round up the standard deviation to set.

[0046] If the effective point number difference or the point cloud centroid displacement of a voxel between two scanning periods exceeds the corresponding permissible threshold value, the voxel is marked as a change voxel.

[0047] All change voxels are searched to identify the dynamic change of the bulk material pile surface.

[0048] In order to extract continuous change regions with monitoring significance from discrete change voxels, spatial clustering is needed to generate a change heat map, including: performing three-dimensional spatial connected component clustering on all marked change voxels, aggregating change voxels that are adjacent in space, such as face adjacent or edge adjacent, to form a number of change clusters, each change cluster corresponding to an independent deformation region on the surface of the material pile.

[0049] According to the proportion of the total volume of the change voxel contained in the change cluster and the average change rate of the point cloud feature in the scanning interval period, the difference between the two is calculated to calculate the morphological change intensity of each change cluster, wherein the average change rate of the point cloud feature in the scanning interval period is specifically the average point cloud centroid displacement amount and the average effective point difference of all voxels in the change cluster, respectively divided by the time interval between the adjacent two scanning periods, and the maximum value of the two calculation results is taken as the average change rate of the point cloud feature in the scanning interval period. In addition, when participating in the calculation of the morphological change intensity, the average change rate is only substituted into the numerical calculation, which is a dimensionless parameter.

[0050] The change heat map is generated based on the spatial distribution of each change cluster and the corresponding morphological change intensity.

[0051] Based on the change heat map, further retrieve the high-heat change area that needs to start enhanced scanning, and the specific process includes: obtaining the change heat map of the current and previous several continuous monitoring periods to form a time sequence.

[0052] For each spatial position in the change heat map, the morphological change intensity can be referred to as a change cluster, and the change trend of the morphological change intensity on the time sequence is analyzed, so as to divide the spatial position into a continuous change area, an intermittent change area and a stable area, and the specific division process is: calculating the average morphological change intensity, the coefficient of variation and the high position persistence of the time sequence, wherein the high position persistence is the proportion of the number of periods in which the morphological change intensity is greater than the median of the global intensity of all regions in the same period.

[0053] If the average morphological change intensity and the coefficient of variation of a certain spatial position corresponding to the time sequence are lower than the overall average level of all regions, the spatial position is marked as a stable area.

[0054] If the average morphological change intensity of a certain spatial position corresponding to the time sequence is higher than the overall average level of all regions, and the high position persistence exceeds the preset proportion, the spatial position is marked as a continuous change area, and the preset proportion can be exemplarily 0.8.

[0055] If a certain spatial position does not belong to the continuous change area or the stable area, it is directly classified as an intermittent change area.

[0056] For different types of regions, adaptive threshold values are used for retrieval. First, the median and standard deviation of all morphological change intensities in the current period change heat map are calculated.

[0057] For the regions classified as stable areas, because the background change is weak, a higher threshold value is required to avoid false detection, and the corresponding high-heat change definition value can be set to the sum of the median and 1 to 1.5 times the standard deviation.

[0058] For the continuously changing area, a relatively low threshold can be used to maintain sensitivity due to its active state itself, and the corresponding high heat change boundary value can be set to the sum of the median and 0.3 to 0.6 times the standard deviation.

[0059] For the intermittently changing area, its characteristics are between the two, and the corresponding high heat change boundary value can be set to the sum of the median and 0.7 to 0.9 times the standard deviation.

[0060] Finally, all the spatial regions covered by the change clusters that meet the corresponding search conditions are determined as the high heat change regions that need to be focused on in the current scanning period.

[0061] It should be noted that the above examples of standard deviation multiples are mainly based on the following principles: (1) Statistical principle: such as using standard deviation multiples to define the tail or significant deviation part in the distribution, which is a common method of anomaly detection.

[0062] (2) Engineering experience: In the monitoring scene, using a stricter threshold for the low active area can effectively reduce false positives, and using a more sensitive threshold for the high active area can help track its dynamics and reflect the conventional considerations of balancing sensitivity and specificity by those skilled in the art.

[0063] In addition, those skilled in the art can follow the core principle of using a higher threshold for the stable area, a lower threshold for the continuously changing area, and a medium threshold for the intermittently changing area, and can customize the standard deviation multiple value according to the point cloud data quality or monitoring accuracy requirements in actual applications through a limited number of routine experiments or historical data statistical analysis.

[0064] The embodiments of the present application improve the scanning accuracy and model fidelity of key areas under dynamic operation. By introducing the change heat map to quantitatively perceive and locate the surface changes of the material pile, the steep section, pit and other rapidly changing regions formed by material taking can be accurately identified. Furthermore, for the high heat change region, multiple slave scanning nodes are dispatched to perform collaborative enhanced scanning with higher angular resolution, ensuring that the geometric details of the disturbed region are captured in time, completely and with high accuracy, effectively solving the problem of local model distortion caused by fixed scanning frequency.

[0065] S3, for the high heat change region, according to the geometric relationship between the sensor and the region and the complementarity of the multi-sensor view angle, a subset of sensors is selected from all slave scanning nodes.

[0066] The selection of the sensor subset aims to achieve effective coverage of all high-heat-change regions with the least sensor resources and optimize the comprehensive observation perspective as much as possible. The selection process is a step-by-step optimization algorithm, which specifically includes: for each high-heat-change region, selecting the subordinate scanning node with the smallest angle between the laser beam incident direction and the surface normal direction as the associated node. The selection is based on the fact that the smallest angle relative to the surface normal direction directly represents the optimal observation geometry of the sensor for the region.

[0067] All associated nodes are merged to form an initial sensor subset. Based on the theoretical coverage range of each node in the initial sensor subset, the coverage range intersection between nodes is identified, and the completely redundant nodes and partially redundant nodes are determined based on the coverage range intersection. The completely redundant node is a node whose theoretical coverage range is completely contained in the theoretical coverage range of at least one other node in the sequence. The partially redundant node is a node whose theoretical coverage range overlaps with the theoretical coverage range of other nodes in the sequence, and after the node is removed from the subset, the remaining nodes cannot jointly cover the high-heat-change region originally associated with the node.

[0068] For the completely redundant node, it is removed from the initial sensor subset.

[0069] For the partially redundant node, the ratio of its overlapping area to the theoretical coverage range is defined as the redundancy rate, and the nodes are sorted according to the redundancy rate. The node with the highest current redundancy rate is selected as the node to be adjusted.

[0070] Within the allowed attitude adjustment range of the node to be adjusted, the laser beam incident direction is optimized to reduce the coverage of the redundant region in the coverage range intersection.

[0071] After verifying the direction optimization, the node to be adjusted is verified to see if it still maintains effective coverage of the high-heat-change region it is originally associated with. The effective coverage means that the angle between the laser beam incident direction and the surface normal direction of the high-heat-change region it is originally associated with is less than a preset effective scanning angle. The preset effective scanning angle is determined according to the technical specifications of the selected laser radar sensor. Specifically, according to the correspondence between the ranging error and the echo intensity attenuation of the sensor at different incident angles in the calibration data, the maximum allowed incident angle under the condition of maintaining scanning accuracy is determined.

[0072] If the verification is passed, the attitude information of the node and the theoretical coverage range and coverage range intersection of all nodes in the initial sensor subset are updated.

[0073] If the verification fails, the original attitude of the node is restored, and the node is marked as an unadjustable node.

[0074] The partial redundant node posture adjustment process is repeated until a convergence condition is met, the convergence condition including any one of the following: (i) the current redundancy rate of all partial redundant nodes is lower than a preset redundancy threshold.

[0075] (ii) a preset maximum number of iterations is reached.

[0076] (iii) there is no partial redundant node whose redundancy rate is higher than the preset redundancy threshold and which is not marked as an unadjustable node.

[0077] The preset redundancy threshold is defined as the maximum upper limit of the redundancy rate allowed for a partial redundant node to remain in the sensor subset. Its specific value is determined according to the coverage reliability safety margin of the method design. For example, it is set to 0.2, which means that a node is allowed to have at most 20% of its coverage area overlapping with other nodes, as long as its remaining independent coverage contribution rate can still play a substantial role in ensuring reliable coverage of its associated area. If the redundancy rate exceeds this threshold, it is considered that the contribution of the node to improving coverage reliability is not significant, and its resource occupation becomes unreasonable, thereby triggering the adjustment or elimination process.

[0078] The initial sensor subset after final optimization adjustment is generated as the selected sensor subset.

[0079] The embodiment of the application constructs an intelligent sensor coordination strategy based on geometric constraints and coverage optimization. When scheduling slave scanning nodes to form a sensor subset, not only the geometric relationship between a single node and the target area is considered, but also the complementary nature of the viewing angles among multiple sensors is emphasized. Through redundant node identification and elimination based on coverage range intersection, and fine-tuning of the postures of partial redundant nodes without affecting full coverage, the final selected sensor subset can achieve efficient and non-missing coverage of all high-heat-change areas with the least number of nodes and the optimal combination of observation viewing angles.

[0080] S4, activating the sensor subset and improving its scanning angle resolution, and performing a coordinated enhanced scan on the high-heat-change area.

[0081] The coordinated enhanced scan aims to use the scheduled sensor subset to obtain refined point cloud data of the high-heat-change area that far exceeds the basic resolution of the main scanning node. This process is achieved through precise synchronization control and parameter improvement, specifically including: assigning the same time reference to all slave scanning nodes in the sensor subset. This time reference can be achieved through a clock synchronization protocol within the scanning network, or a time stamp instruction broadcast by the master system, and the purpose is to establish a common time coordinate system for the scanning actions of all nodes.

[0082] Under the time reference, each slave scanning node in the sensor subset is controlled to start scanning synchronously to avoid the scanning laser beams interfering with each other in the high-heat-variation region, while the scanning angle resolution of each slave scanning node in the sensor subset is improved to a level higher than that of the master scanning node.

[0083] S5, merging the high-density local point cloud obtained by the enhanced scanning with the global point cloud of the master scanning node to generate a three-dimensional surface model of the bulk material pile.

[0084] The three-dimensional surface model generation process of the bulk material pile includes: converting the local high-density point cloud data obtained for the same high-heat-variation region and the global point cloud data obtained by the master scanning node in the region to the same coordinate system to perform registration.

[0085] For the registered point cloud data from different sensors, the confidence weight of each data point is calculated based on the distance of the source sensor and the laser incidence angle, which is inversely proportional to the square of the distance between the data point and its source sensor, and is proportional to the cosine value of the laser beam incidence angle at the point. The basis is that the closer the distance, the closer the incidence angle, the smaller the measurement error theory, and the more reliable the data.

[0086] As an example of confidence weight calculation, the square of the distance between all data points and their source sensors is calculated, and the arithmetic mean is calculated as the numerator. The square of the distance between a certain data point and its source sensor is taken as the denominator, the ratio is calculated, and the ratio operation result is multiplied by the cosine value of the laser beam incidence angle at the point to obtain the initial confidence. In order to facilitate subsequent weighted average calculation, the initial confidence of all points from the same sensor or the same fusion region can be normalized so that their sum is 1, so as to obtain the confidence weight of each data point.

[0087] The weighted average method is used for fusion to generate the final point cloud set of the high-heat-variation region. The direct technical effect of using the weighted average method for fusion is that in the overlapping area, the points with high confidence contribute more to the fusion result, thereby suppressing noise and improving the geometric accuracy of the fused point cloud at the pixel level, which helps to generate the final high-precision point cloud set of the region.

[0088] Using the final point cloud set, a three-dimensional surface model of the high-heat-variation region is reconstructed by a triangular meshing algorithm.

[0089] The model part of the corresponding region in the original global three-dimensional surface model is replaced and updated with the reconstructed three-dimensional surface model to generate a global three-dimensional surface model of the bulk material pile.

[0090] After the bulk material pile three-dimensional surface model generation step is completed, a reset and preparation phase is entered, specifically including: controlling all slave scanning nodes participating in collaborative enhanced scanning to resume low-power mode until the next time they are selected as sensor subset members, and taking the generated bulk material pile three-dimensional surface model as the typical spatial profile model of the newly input bulk material pile.

[0091] The embodiment of the present application realizes dynamic on-demand scheduling and energy efficiency optimization of scanning resources, discards the inefficient mode of all laser radar sensors working at full power all the time in the prior art, sets up a master scanning node for global monitoring, and places slave scanning nodes in low-power mode. Only when a high-heat change area caused by pile and reclaimer operation is identified, the optimal slave scanning node subset is dynamically activated and scheduled for focused scanning. This mechanism fundamentally avoids generating redundant data and unnecessary energy consumption during material static or slow change periods, significantly reduces scanning data processing load, and prolongs the overall service life of the sensor network.

[0092] The above is merely an example and description of the present application, and those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, as long as they do not deviate from the concept of the present application or exceed the scope defined by the present application, and should belong to the protection scope of the present application.

Claims

1. A method for scanning bulk materials based on a lidar sensor, characterized in that, include: S1. Among the multiple lidar sensors deployed around the bulk material pile, a master scanning node is designated to continuously monitor the bulk material pile, and the remaining sensors are set as slave scanning nodes and placed in low power mode. Obtain the fixed installation positions and spatial attitude parameters of all LiDAR sensors in the scanning network. Combined with a pre-input typical spatial contour model of the bulk material pile, perform line-of-sight simulation analysis to calculate the visible area ratio of each LiDAR sensor on the surface of the typical spatial contour model. Based on the visible area ratio, determine whether there is a LiDAR sensor with full surface coverage. If so, directly designate it as the main scanning node. If not, perform iterative construction: a. Select the LiDAR sensor with the highest visible area ratio as the first main scanning node. b. Calibrate the blind zone on the surface of the typical spatial contour model corresponding to the first main scanning node, and select the sensors with observation angles for this blind zone from the remaining sensors to add as new main scanning nodes; c. Repeat steps a and b until the blind zone on the surface of the typical spatial contour model is eliminated; during the specified process, if multiple sensors that meet the current specified conditions appear in parallel, the sensor with the higher theoretical average point cloud density shall be selected first. S2. Register the point cloud data collected by the main scanning node in adjacent scanning cycles, identify the dynamic changes on the surface of the bulk material pile through voxelization differential calculation, generate a heat map of the morphological change intensity of different spatial regions per unit time through spatial clustering, and retrieve the high heat change regions in the map. S3. For the high-temperature change area, based on the geometric relationship between the sensor and the area and the complementarity of the multi-sensor perspective, select a subset of sensors from all subordinate scanning nodes; S4. Activate the sensor subset and improve its scanning angular resolution to perform a collaborative enhancement scan on the high-thermal-change region; S5. The high-density local point cloud obtained by the enhanced scan is fused with the global point cloud of the main scan node to generate a three-dimensional surface model of the bulk material pile.

2. The bulk material scanning method based on a lidar sensor according to claim 1, characterized in that, The registration of point cloud data collected by the main scanning node in adjacent scanning cycles includes: The point cloud data from two adjacent scanning cycles are defined as the source point cloud and the target point cloud, respectively, and a spatial index of the target point cloud is constructed. For each point in the source point cloud, the spatial index is used to find its spatial nearest neighbor in the target point cloud to form a temporary point pair correspondence. Based on the angle between the normal vectors of the two points in the temporary corresponding point pair, remove point pairs that do not conform to geometric consistency to obtain the optimized corresponding point pair set; Based on the optimized set of corresponding point pairs, the rigid body transformation matrix that minimizes the cumulative spacing between point pairs is solved by least squares optimization. The spatial position of the source point cloud is updated by applying the rigid body transformation matrix, and the average spacing of the optimized corresponding point pair set after transformation is calculated simultaneously as the registration residual for this iteration. Using the source point cloud transformed in the previous iteration as new input, the process of finding corresponding points and calculating rigid body transformation matrices is iteratively executed until the termination condition is met, and the two-phase point cloud data that have been registered and are in the same coordinate system are output.

3. The bulk material scanning method based on a lidar sensor according to claim 2, characterized in that, The method of identifying dynamic changes on the surface of bulk material piles through voxelized differential calculation includes: The coordinate system of the two phases of point cloud data after registration is divided into a three-dimensional voxel mesh; The number of valid points and the centroid position of the point cloud for each voxel in two scanning cycles are counted, and the difference in the number of valid points and the displacement of the centroid of the point cloud are calculated for the same voxel between the two scanning cycles. The registration residual from the last iteration is retrieved to set the permissible threshold for the centroid displacement of the point cloud. The statistical distribution of the number of valid points for each voxel in the historical scan cycle is retrieved to update the permissible threshold for the difference in the number of valid points. If the difference in the number of valid points or the displacement of the centroid of the point cloud exceeds its corresponding permissible threshold between two scan cycles, the voxel is marked as a variable voxel. All variable voxels are retrieved to identify dynamic changes on the surface of the bulk material pile.

4. The bulk material scanning method based on a lidar sensor according to claim 3, characterized in that, The process of generating the heatmap includes: Three-dimensional spatial connectivity clustering is performed on all labeled variant voxels to form several variant clusters; Based on the total volume percentage of the voxels contained in the variation cluster and the average rate of change of its point cloud features within the scanning interval, the morphological change intensity of each variation cluster is calculated, and the variation heatmap is generated based on the spatial distribution of each variation cluster and its corresponding morphological change intensity.

5. The bulk material scanning method based on a lidar sensor according to claim 4, characterized in that, The process of searching for high-temperature change areas includes: Obtain heat maps showing changes in the current and previous preset number of continuous monitoring periods to form a time series; For each cluster of changes in the heat map, the trend of its morphological change intensity over the time series is analyzed to divide the spatial location into a continuous change zone, an intermittent change zone, and a stable zone. Calculate the median and standard deviation of the intensity of morphological changes in the current heat map, and use this to define the high heat change threshold for different regional types. Clusters of changes whose morphological change intensity exceeds the high-temperature change area retrieval threshold defined by their respective region type are identified as high-temperature change areas.

6. The bulk material scanning method based on a lidar sensor according to claim 1, characterized in that, The step of selecting a subset of sensors from all subordinate scanning nodes includes: For each region of high heat change, the subordinate scanning node with the smallest angle between the incident direction of the laser beam and the normal direction of its surface is selected as the associated node, and all associated nodes are merged to form an initial sensor subset. Based on the theoretical coverage of each node in the initial sensor subset, the intersection of coverage between nodes is identified, and fully redundant nodes and partially redundant nodes are defined according to the intersection of coverage. Fully redundant nodes are removed from the initial sensor subset. For some redundant nodes, sort them according to their redundancy rate, and select the node with the highest redundancy rate as the node to be adjusted. Within the allowable attitude adjustment range of the node to be adjusted, optimize its laser beam incident direction to reduce its coverage of redundant areas in the intersection of the coverage range; After the verification direction is optimized, it is determined whether the node to be adjusted still maintains effective coverage of its original associated high-heat change area; If the verification passes, update the attitude information of the node and the theoretical coverage and intersection of the coverage of all nodes in the initial sensor subset; If the verification fails, the node is restored to its original pose and marked as an unadjustable node. Repeat the attitude adjustment process for redundant nodes until the convergence condition is met, and then use the final optimized initial sensor subset to generate the selected sensor subset.

7. The bulk material scanning method based on a lidar sensor according to claim 1, characterized in that, The collaborative enhanced scanning execution process includes: Assign the same time base to all subordinate scanning nodes in the sensor subset; Under the stated time reference, each subordinate scanning node in the sensor subset is controlled to start scanning synchronously, and the scanning angular resolution of the subordinate scanning node is required to be higher than that of the master scanning node.

8. The bulk material scanning method based on a lidar sensor according to claim 1, characterized in that, The process of generating the three-dimensional surface model of the bulk material pile includes: The local high-density point cloud data acquired for the same high-temperature change area, and the global point cloud data acquired by the main scanning node in the same area, are uniformly converted to the same coordinate system for registration. For the registered point cloud data from different sensors, the confidence weight of each data point is calculated based on the distance from the source sensor and the laser incident angle. The data is then fused using a weighted average method to generate the final point cloud set of the high heat change area. Using the final point cloud, a three-dimensional surface model of the high-temperature change region is reconstructed through a triangular meshing algorithm; The model portion of the corresponding region in the original global 3D surface model is replaced and updated with the reconstructed 3D surface model to generate a global 3D surface model of the bulk material pile.

9. The bulk material scanning method based on a lidar sensor according to claim 1, characterized in that, After completing the step of generating the 3D surface model of the bulk material pile, the following steps are also included: Control all subordinate scanning nodes participating in the collaborative enhancement scan to return to low-power mode until they are selected as a member of the sensor subset again, and use the generated 3D surface model of the bulk material pile as the typical spatial contour model of the bulk material pile as the new input.

Images