A large stockyard-oriented multi-machine point cloud splicing and measurement accuracy checking method

Abstract

A kind of multi-machine point cloud splicing and measurement precision checking method for large stockyard, it relates to three-dimensional perception and bulk storage measurement technical field.The present application is to solve the problem of existing large stockyard in single three-dimensional laser radar, multiple radar point cloud is difficult to unify splicing, splicing seam volume is inconsistent, point cloud quality is uneven and the problem of lack of measurement precision checking.The present application arranges multiple radars along the large stockyard, and calibrates the point cloud to the global coordinate system by means of target or feature ground calibration;ICP is used for precise registration in overlapping area;dynamic weight is constructed based on point cloud density, incident angle and signal-to-noise ratio for quality weighted fusion splicing;average height difference is used to monitor volume consistency in overlapping area, and re-registration is triggered when exceeding threshold;volume calculation is only counted once in overlapping area;standard body calibration traceability, weighing true value back calculation and multi-radar cross-checking are combined for checking.It is used for accurate measurement of coal, ore, grain and other bulk storage.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional perception and bulk material storage metering technology, specifically to a method for multi-three-dimensional lidar point cloud stitching and metering accuracy verification for large material yards. It is applicable to the accurate metering of material volume and inventory in large strip material yards, circular material yards and multi-position storage facilities. Multiple three-dimensional lidars work together to acquire complete material surface point clouds and perform multi-method verification and traceability of volume and inventory weight metering accuracy. Background Technology

[0002] In bulk material storage management in industries such as power, metallurgy, and building materials, accurate measurement of material yard inventory is directly related to production scheduling, trade settlement, and safety supervision. Three-dimensional lidar can acquire planar, high-coverage point clouds of material surfaces, making it an important means of measuring the volume of material surfaces in material yards. However, large strip-shaped material yards can reach hundreds of meters in length. A single three-dimensional lidar unit, limited by its range, field of view, and self-occlusion of the material pile, cannot achieve complete, blind-spot-free coverage of the entire material surface, often resulting in significant errors in the measured volume due to incomplete coverage. Furthermore, existing solutions based on single-point level gauges, RTK-GPS manual measurement, or photogrammetry suffer from low measurement efficiency, insufficient coverage, and susceptibility to environmental interference, making it difficult to meet the all-weather, high-precision measurement requirements of large material yards.

[0003] To improve coverage, multiple 3D lidar units are typically deployed along the material yard for zonal scanning. However, the point clouds collected by multiple radars exist in their respective local coordinate systems, presenting the following technical challenges: First, the installation pose (extrinsic parameters) of each radar is difficult to predict precisely, and the local point clouds cannot be directly stitched together into a unified global point cloud. Second, inconsistencies exist in the point clouds in overlapping areas of adjacent radar scanning ranges; simple superposition will result in duplicate volume calculations or missing data at the seams. Third, the dusty environment and rain / fog attenuation in the material yard will degrade the quality of some radar point clouds; indiscriminate fusion will introduce errors. Fourth, there is a lack of effective verification and traceability methods for the accuracy of the stitched volume and inventory weight measurements, making it difficult to guarantee the reliability of the measurement results.

[0004] Existing methods primarily focus on volume measurement or single-point material level measurement using a single radar. For example, volume integration methods based on a single fixed radar are only suitable for small to medium-sized material yards and cannot solve the coverage blind spot problem caused by material pile obstruction. While scanning schemes based on mobile platforms can improve coverage, they are difficult to implement online real-time measurement. Current technologies lack a systematic solution for the collaborative calibration, point cloud stitching, and measurement accuracy verification of multiple 3D LiDARs in large material yards. Specifically, key technical issues for improving the reliability of volume and inventory weight measurement in large material yards include: accurately calibrating and stitching the local point clouds of multiple radars into a complete global point cloud; constraining the volume consistency of the stitching seams to avoid duplicate calculations or omissions; adaptively reducing the weight of point clouds with degraded quality in harsh environments such as dust, rain, and fog; and using multiple methods to verify and trace the measurement accuracy.

[0005] In summary, existing single-unit 3D lidar in large material yards suffers from problems such as incomplete coverage, difficulty in uniformly stitching together point clouds from multiple lidar units, inconsistent stitching seam volumes, inconsistent point cloud quality, and lack of verification of measurement accuracy. Summary of the Invention

[0006] The purpose of this invention is to address the problems of incomplete coverage by a single 3D LiDAR in existing large material yards, difficulty in uniformly stitching point clouds from multiple LiDARs, inconsistent seam volumes, inconsistent point cloud quality, and lack of verification of measurement accuracy. Therefore, this invention provides a method for multi-machine point cloud stitching and measurement accuracy verification in large material yards, achieving high-precision stitching of complete point clouds across the entire yard and reliable verification of volume and inventory weight measurement accuracy.

[0007] The technical solution of this invention is: a method for multi-machine point cloud stitching and measurement accuracy verification for large material yards, comprising the following steps:

[0008] Step 1: Deployment and coordinated calibration of multiple radar stations;

[0009] Step 11: Deploy multiple 3D LiDARs along the large material yard so that the scanning ranges of adjacent LiDARs overlap on the material surface.

[0010] Steps 1 and 2: Deploy several calibration targets with known coordinates or select stable feature features within the material yard. Estimate the installation pose of each radar relative to the global material yard coordinate system, i.e., the rotation matrix R, by using the observations of each radar on the targets or feature features. m With translation vector t m The local point clouds collected by each radar are uniformly transformed to the same global coordinate system;

[0011] Step 2: Fine-tuning the overlapping area;

[0012] After transforming each radar point cloud to the global coordinate system, the overlapping areas of adjacent radar scanning ranges are precisely registered to eliminate residual errors in external parameter estimation.

[0013] Step 3: Quality-weighted fusion and splicing;

[0014] For the registered overlapping area, dynamic weights are constructed based on one or more of the quality indicators of point cloud density, incident angle, and signal-to-noise ratio of each radar at each location. A weighted fusion method is used to generate a unified material surface elevation. The higher the point cloud quality, the greater the weight. When the point cloud quality of a radar at the corresponding location is lower than the threshold, its weight is set to zero and it is automatically removed. In this way, the multi-source point clouds are fused and stitched into a complete material surface point cloud that covers the entire field and is consistent.

[0015] Step 4: Calculation of splicing volume and consistency constraints;

[0016] The spliced ​​point cloud of the entire material surface is meshed. Combining the material yard geometric model and the reference plane, the space between the material surface and the reference plane is meshed and summed to obtain the volume of the entire field. Volume consistency constraints are introduced in the overlapping area. The splicing consistency is measured by the average elevation difference between the two radars in the overlapping area. When the average elevation difference exceeds the threshold, re-registration or external parameter optimization is triggered. When calculating the volume, the elevation of the overlapping area is taken as the fused value and only counted once.

[0017] Step 5: Measurement accuracy verification and closed-loop feedback;

[0018] The measurement accuracy is verified by a multi-method accuracy verification method that combines standard body calibration traceability, true value back calculation based on weighing device and multi-radar cross-verification. When the accuracy does not meet the standard, recalibration or re-splicing is triggered to form an accuracy closed loop.

[0019] Furthermore, in step two, the fine registration of adjacent overlapping radar regions employs an iterative nearest-point method, optimizing the relative pose in the overlapping region with the goal of minimizing the sum of squared point-to-point distances.

[0020]

[0021] In the formula, p j With q j R and t represent the corresponding point pairs from the two radars in the overlapping area, respectively, where R and t are the relative rotation and translation to be optimized, and n is the number of point pairs.

[0022] Let theta be a preset residual threshold. When the registration residual is greater than theta, the external parameters are adjusted and the registration is re-registered until the residual converges.

[0023] Furthermore, in the weighted fusion step three, the weight w of each radar at grid g is... m (g) Based on the aforementioned quality indicators, the elevation z(g) of the fused material surface is determined by the following formula:

[0024]

[0025] In the formula, z m (g) represents the surface elevation of the m-th radar at grid g; when the point cloud quality of a radar at grid g is lower than a preset threshold, its weight w m (g) is set to zero.

[0026] Furthermore, the average elevation difference δ in step four is determined by the following formula:

[0027]

[0028] In the formula, z a (j), z b (j) represents the surface elevation of the two radars at the j-th grid in the overlapping area, and n represents the number of grids in the overlapping area.

[0029] Furthermore, the total field volume V in step four is determined by the following formula:

[0030]

[0031] In the formula, G is the total number of grid cells, and z s (g) is the elevation of the material surface at grid g, z0(g) is the elevation of the reference surface at grid g, and Δs is the horizontal area of ​​the grid.

[0032] Furthermore, the standard body calibration and traceability in step five includes: placing a standard body with a known volume in the material yard, measuring the volume of the standard body by splicing point cloud and comparing it with the known true value, and performing traceability verification of the volume measurement.

[0033] Furthermore, step five, the back-calculation based on the true value of the weighing device, includes: back-calculating the reference volume V using the actual weight M calibrated by the truck scale or belt scale and the bulk density ρ. ref =M / ρ, compared with the volume V measured from the stitched point cloud, the volume relative error is defined as:

[0034] .

[0035] Furthermore, the multi-radar cross-verification in step five includes: cross-comparing the volumes of the overlapping area calculated by different radars, identifying and eliminating abnormal stations, and selecting the radar with the best quality as the main measurement source for the area.

[0036] Furthermore, the metrological accuracy verification in step five uses one or more of the following indicators to quantitatively evaluate metrological accuracy: mean absolute error, mean absolute percentage error, and maximum error ratio.

[0037] The mean absolute percentage error is defined as follows:

[0038]

[0039] In the formula, n is the sample size, and y i and These are the reference value and the measured value for the i-th sample, respectively.

[0040] Preferably, the three-dimensional lidar is a 3D lidar with explosion-proof certification.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. This invention proposes a multi-3D lidar collaborative calibration and point cloud stitching method for large material yards. It estimates the extrinsic parameters of each radar by using targets or stable feature features, and eliminates residual pose errors by ICP iterative fine registration in the overlapping area. The multi-source local point clouds are stitched into a complete material surface point cloud covering the entire field without blind spots, fundamentally solving the problem of incomplete coverage by a single radar.

[0043] 2. This invention proposes a weighted fusion mechanism based on multi-dimensional quality indicators such as point cloud density, incident angle and signal-to-noise ratio, which automatically reduces the weight and removes radar point clouds whose quality is degraded due to harsh environments such as dust, rain and fog, effectively improving the robustness and reliability of spliced ​​point clouds under complex working conditions.

[0044] 3. This invention proposes a method for constraining the consistency of splice seam volume based on the average elevation difference of the overlapping area. When the elevation difference exceeds the threshold, re-registration or external parameter optimization is triggered, and the volume of the overlapping area is only counted once. This mechanism avoids repeated calculation or missing volume and ensures the continuity and uniqueness of the volume of the entire field.

[0045] 4. This invention proposes a multi-method measurement accuracy verification method that combines standard body calibration traceability, true value back calculation based on weighing devices, and multi-radar cross-verification. It also constructs an accuracy closed loop of "verification-feedback-recalibration". When the accuracy does not meet the standard, it automatically triggers recalibration or reassembly, which significantly improves the reliability and traceability of the measurement of volume and inventory weight in large material yards.

[0046] 5. This invention, through the collaborative calibration of multiple three-dimensional lidars, point cloud stitching and fusion, and measurement accuracy verification, can achieve high-precision measurement of the complete material surface in large material yards and reliably guarantee measurement accuracy. It has significant advantages such as high coverage, good accuracy, strong robustness, and traceability.

[0047] (1) Multi-machine collaborative splicing breaks through the coverage range limitation of a single radar, realizing complete and blind-spot-free measurement of the material surface of a large strip material yard. The material surface coverage can be increased from about 60% of a single unit to about 99%, which is significantly better than the existing single unit solution.

[0048] (2) Based on the weighted fusion of multi-dimensional quality indicators and automatic removal of abnormal stations, it effectively copes with the interference of harsh environments such as dust and rain and fog, and ensures the measurement robustness of the point cloud quality degradation area;

[0049] (3) The consistency constraint of splice seam volume avoids the repeated calculation and missing of the volume of the overlapping area from the mechanism, ensuring the accuracy and continuity of the volume of the whole field, and the relative error of volume can be reduced to less than 1%.

[0050] (4) Multi-method accuracy verification and traceability mechanism, so that the volume and inventory weight measurement results can be verified and traced, and the inventory weight measurement MAPE can be controlled within 1.0%, which meets the accuracy requirements of trade measurement level;

[0051] (5) Prefer three-dimensional lidar with explosion-proof certification to meet the explosion-proof safety requirements of the material yard. The method can be connected with the company's existing metering and management system, with strong compatibility and easy engineering deployment.

[0052] Evaluation is conducted using indicators such as material coverage rate, relative volume error, registration residual in overlapping areas (RMSE), MAPE of inventory weight measurement, and maximum error ratio. In a specific embodiment of the present invention, compared to a single radar, three-unit splicing can increase the material coverage rate from approximately 60% to approximately 99%, reduce the relative volume error from approximately 3.8% to approximately 0.7%, converge the registration residual in overlapping areas to the sub-centimeter level (within approximately 0.05 to 0.1 m), and control the MAPE of inventory weight measurement to within 1.0% (the specific value depends on the equipment and site conditions). Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the multi-radar deployment and point cloud coverage in a large material yard according to the present invention. Figure 2 This is a schematic diagram of the multi-machine calibration and point cloud stitching process of the present invention. Figure 3 This is a schematic diagram of the measurement accuracy verification process of the present invention. Figure 4 This is a diagram showing the splicing accuracy result of a specific embodiment of the present invention. Detailed Implementation

[0054] Specific Implementation Method 1: This implementation method provides a multi-machine point cloud stitching and metering accuracy verification method for large material yards, including the following steps:

[0055] Step 1: Deployment and coordinated calibration of multiple radar stations;

[0056] Step 11: Deploy multiple 3D LiDARs along the large material yard so that the scanning ranges of adjacent LiDARs overlap on the material surface.

[0057] Steps 1 and 2: Deploy several calibration targets with known coordinates or select stable feature features within the material yard. Estimate the installation pose of each radar relative to the global material yard coordinate system, i.e., the rotation matrix R, by using the observations of each radar on the targets or feature features. m With translation vector t m The local point clouds collected by each radar are uniformly transformed to the same global coordinate system;

[0058] Step 2: Fine-tuning the overlapping area;

[0059] After transforming each radar point cloud to the global coordinate system, the overlapping areas of adjacent radar scanning ranges are precisely registered to eliminate residual errors in external parameter estimation.

[0060] Step 3: Quality-weighted fusion and splicing;

[0061] For the registered overlapping area, dynamic weights are constructed based on one or more of the quality indicators of point cloud density, incident angle, and signal-to-noise ratio of each radar at each location. A weighted fusion method is used to generate a unified material surface elevation. The higher the point cloud quality, the greater the weight. When the point cloud quality of a radar at the corresponding location is lower than the threshold, its weight is set to zero and it is automatically removed. In this way, the multi-source point clouds are fused and stitched into a complete material surface point cloud that covers the entire field and is consistent.

[0062] Step 4: Calculation of splicing volume and consistency constraints;

[0063] The spliced ​​point cloud of the entire material surface is meshed. Combining the material yard geometric model and the reference plane, the space between the material surface and the reference plane is meshed and summed to obtain the volume of the entire field. Volume consistency constraints are introduced in the overlapping area. The splicing consistency is measured by the average elevation difference between the two radars in the overlapping area. When the average elevation difference exceeds the threshold, re-registration or external parameter optimization is triggered. When calculating the volume, the elevation of the overlapping area is taken as the fused value and only counted once.

[0064] Step 5: Measurement accuracy verification and closed-loop feedback;

[0065] The measurement accuracy is verified using a multi-method accuracy verification approach that combines standard body calibration and traceability, true value back calculation based on weighing devices, and multi-radar cross-verification. When the accuracy fails to meet the standard, recalibration or reassembly is triggered to form an accuracy closed loop.

[0066] Specific Implementation Method Two: In step two of this implementation method, the fine registration of adjacent radar overlapping areas adopts an iterative nearest point method, optimizing the relative pose in the overlapping area with the goal of minimizing the sum of squared distances between point pairs.

[0067]

[0068] In the formula, p j With q jR and t represent the corresponding point pairs from the two radars in the overlapping area, respectively, where R and t are the relative rotation and translation to be optimized, and n is the number of point pairs.

[0069] Let theta be a preset residual threshold. When the registration residual is greater than theta, the external parameters are adjusted and the registration is re-registered until the residual converges.

[0070] Step two of this implementation method uses ICP iterative fine registration to find corresponding point pairs in the overlapping area. The relative pose is continuously fine-tuned with the goal of minimizing the sum of squared distances between point pairs, thereby eliminating the pose error remaining in step one and making adjacent radar point clouds accurately aligned in the overlapping area.

[0071] Specific Implementation Method 3: In the weighted fusion step 3 of this implementation method, the weight w of each radar at grid g is... m (g) Based on the aforementioned quality indicators, the elevation z(g) of the fused material surface is determined by the following formula:

[0072]

[0073] In the formula, z m (g) represents the surface elevation of the m-th radar at grid g; when the point cloud quality of a radar at grid g is lower than a preset threshold, its weight w m (g) is set to zero.

[0074] This implementation addresses the issue of inconsistent point cloud quality among different radars under harsh environments such as dust, rain, and fog, ensuring the accuracy and reliability of the fused material surface elevation. By dynamically assigning weights to each radar based on quality indicators such as point cloud density, incident angle, and signal-to-noise ratio, higher-quality radars contribute more, and lower-quality radars contribute less. When the quality of a radar at a certain grid point falls below a preset threshold, its weight is directly set to zero, effectively removing the radar's measurement value at that point. This processing method avoids including low-quality data in the fusion process, preventing inferior data from contaminating the final result. Therefore, even under conditions of localized dust or rain / fog, the fused material surface elevation accurately reflects the material surface shape, providing a reliable data foundation for subsequent volume calculations.

[0075] Specific Implementation Method Four: The average elevation difference δ in step four of this implementation method is determined by the following formula:

[0076]

[0077] In the formula, z a (j), z b (j) represents the surface elevation of the two radars at the j-th grid in the overlapping area, and n represents the number of grids in the overlapping area.

[0078] Step four of this implementation introduces the average elevation difference δ, whose core function is to monitor and ensure volume consistency in the splicing seam area. Although step two aligns the point cloud through ICP fine registration, slight elevation deviations may still exist in the overlapping area after registration due to factors such as changes in the field environment and radar drift. δ provides an objective value by quantitatively calculating the elevation difference between the two radars at the same location in the overlapping area—the smaller δ is, the better the splicing consistency; the larger δ is, the more significant the deviation at the splicing seam. After setting a threshold, once δ exceeds the limit, re-registration or extrinsic parameter optimization is immediately triggered, which is equivalent to setting a "quality threshold" before volume calculation to prevent errors caused by continuing to calculate the volume when the splicing quality is substandard. At the same time, the elevation of the overlapping area is taken as a fused value and only counted once in the volume calculation stage, fundamentally avoiding duplicate or missing volume calculations.

[0079] Specific Implementation Method Five: The total field volume V in step four of this implementation method is determined by the following formula:

[0080]

[0081] In the formula, G is the total number of grid cells, and z s (g) is the elevation of the material surface at grid g, z0(g) is the elevation of the reference surface at grid g, and Δs is the horizontal area of ​​the grid.

[0082] Specific Implementation Method Six: The standard body calibration and traceability in step five of this implementation method includes: placing a standard body with a known volume in the material yard, measuring the volume of the standard body by splicing point cloud and comparing it with the known true value, and performing traceability verification of the volume measurement.

[0083] This implementation method compares and verifies the volume measurement results of the stitched point cloud by using a standard body with a known volume, traces the measurement data to a unified measurement benchmark, ensures the accuracy and reliability of the measurement results, makes them traceable and credible, and meets the requirements of measurement reliability in scenarios such as trade settlement.

[0084] Specific Implementation Method Seven: Step five of this implementation method involves back-calculating the true value based on the weighing device, including: back-calculating the reference volume V using the true weight M calibrated by the truck scale or belt scale and the bulk density ρ. ref =M / ρ, compared with the volume V measured from the stitched point cloud, the volume relative error is defined as:

[0085] .

[0086] This implementation utilizes real weight data provided by external weighing devices such as truck scales or belt scales, and calculates a reference volume by combining it with the material's bulk density. This reference volume is then used as the true value, independent of the lidar measurement system, and compared with the volume measured by the stitched point cloud. This method provides supplementation and verification for standard body calibration and traceability. When it is inconvenient to place a standard body on-site, the true value calculated by the weighing device can provide a reliable external reference benchmark, effectively verifying the accuracy of the stitched point cloud volume measurement and enhancing the credibility of the measurement results.

[0087] Specific implementation method eight: The multi-radar cross-verification in step five of this implementation method includes: cross-comparing the volumes of the overlapping area calculated by different radars, identifying and eliminating abnormal stations, and selecting the radar with the best quality as the main measurement source for the area.

[0088] This implementation method enables the independent calculation and cross-comparison of volumes using different radars within overlapping areas. Consistency judgment identifies and eliminates stations with abnormal deviations, selecting the radar with the highest quality as the primary measurement source for that area. This mechanism does not rely on external benchmarks, utilizing redundant observation information from multiple radars within the system for self-verification. It effectively avoids measurement errors caused by single radar malfunctions, environmental interference, or calibration deviations, thus improving the overall fault tolerance and measurement reliability of the system.

[0089] Specific Implementation Method Nine: In step five of this implementation method, the metrological accuracy verification uses one or more of the following indicators to quantitatively evaluate the metrological accuracy: mean absolute error, mean absolute percentage error, and maximum error ratio.

[0090] The mean absolute percentage error is defined as follows:

[0091]

[0092] In the formula, n is the sample size, and y i and These are the reference value and the measured value for the i-th sample, respectively.

[0093] This facilitates the transformation of accuracy verification from a qualitative "accuracy" to a quantitative "accuracy level," providing quantifiable and comparable error evaluation standards for measurement results. These indicators objectively determine whether the current measurement accuracy meets application requirements. When the accuracy fails to meet the standards, recalibration or reassembly is triggered, forming a closed-loop mechanism of "verification-feedback-optimization," ensuring that the accuracy level of each measurement can be quantitatively evaluated and continuously improved.

[0094] Specific Implementation Method 10: The three-dimensional lidar in this implementation method is a 3D lidar with explosion-proof certification.

[0095] This embodiment specifies that the 3D lidar is an explosion-proof certified 3D lidar, enabling it to meet the mandatory explosion-proof safety requirements of equipment in large material yards (such as bulk material storage sites for coal, ore, etc.). Material yard environments typically contain combustible dust or flammable gases, posing safety hazards to the use of ordinary electronic equipment in such locations. Explosion-proof certified lidar ensures safe operation of equipment in the complex and hazardous environment of material yards, allowing the method of this invention to be truly implemented and deployed in industrial production sites, meeting the entry requirements for metering equipment in explosion-proof safety regulations for material yards in industries such as power, metallurgy, and building materials.

[0096] Furthermore, the technical solution of this invention includes four stages: multi-radar deployment and collaborative calibration, multi-source point cloud stitching and fusion, stitching volume calculation, and metrological accuracy verification. It should be specifically noted that the core of this invention lies in the overall scheme of "collaborative calibration of multiple three-dimensional lidars, point cloud stitching and fusion, and metrological accuracy verification." The specific algorithms for each stage are not limited to those listed in the embodiments below; anything falling within this overall scheme is within the protection scope of this invention.

[0097] (I) Multi-radar deployment and coordinated calibration

[0098] Multiple 3D lidar units are deployed along the large material yard, so that the scanning ranges of adjacent lidar units overlap to a certain extent on the material surface (e.g., Figure 1 (As shown). Several calibration targets with known coordinates are deployed within the material yard, or stable feature features are selected. The installation pose (i.e., rotation matrix R) of each radar relative to the global material yard coordinate system is estimated using observations of the targets / feature features by each radar. m With translation vector t m For any point p acquired by the m-th radar... (m) Its coordinates in the global material field coordinate system are:

[0099]

[0100] In the formula, R m t m Let p be the extrinsic parameter of the m-th radar. (G) This represents the coordinates of the point in the global coordinate system. This allows all radar point clouds to be uniformly transformed to the same global coordinate system.

[0101] (II) Multi-source point cloud stitching and fusion (core)

[0102] After transforming each radar point cloud to the global coordinate system, fine registration is performed on the overlapping areas of adjacent radar scan ranges to eliminate residual errors in extrinsic parameter estimation. This invention employs registration methods such as Iterative Closest Point (ICP) to optimize relative pose in the overlapping area with the goal of minimizing the sum of squared distances between point pairs.

[0103]

[0104] In the formula, p j With q j Let R and t be the corresponding point pairs from the two radars in the overlapping area, respectively, and let n be the relative rotation and translation to be optimized. When the registration residual is still large, the extrinsic parameters are adjusted based on feedback, and re-registration is performed until the residual converges (e.g., ...). Figure 3 (As shown).

[0105] For the overlapping areas after registration, a weighted fusion method is used to generate a unified material surface elevation to suppress the point cloud quality degradation caused by dust, rain, and fog attenuation from a single radar:

[0106]

[0107] In the formula, z m (g) represents the elevation of the material surface at grid g for the m-th radar, w m (g) represents the corresponding weight, which can be determined based on the radar's point cloud density, incident angle, signal-to-noise ratio, and other quality indicators at that location. The higher the quality, the greater the weight. When the point cloud quality of a radar at that location is too low, its weight approaches zero, and it is automatically downweighted and removed. In this way, multi-source point clouds are fused and stitched together into a complete, consistent point cloud covering the entire field without blind spots.

[0108] (III) Calculation of spliced ​​volume and consistency constraints

[0109] The spliced ​​point cloud of the entire material surface is meshed. Combining the material yard geometric model with the reference plane, the space between the material surface and the reference plane is summed by meshing to obtain the total volume of the entire area.

[0110]

[0111] In the formula, G is the total number of grid cells, and z s z(g) and z0(g) represent the material surface elevation and the reference surface elevation at grid g, respectively, and Δs is the horizontal area of ​​the grid. To avoid duplicate or missing volume calculations at the splicing seam, a volume consistency constraint is introduced in the overlapping area: the splicing consistency is measured by the average elevation difference between the two radars in the overlapping area.

[0112]

[0113] In the formula, z a (j), z b (j) represents the surface elevation of the two radars at the j-th grid in the overlapping area, δ is the average elevation difference, and n is the number of grids in the overlapping area. When δ exceeds the threshold, re-registration or extrinsic parameter optimization is triggered; during volume calculation, the fused elevation value of the overlapping area is taken and only counted once, thereby ensuring the continuity and uniqueness of the volume at the splice seam.

[0114] (iv) Measurement accuracy verification

[0115] To ensure the reliability of measurement results, this invention provides a multi-method precision verification and traceability mechanism (such as...). Figure 3 (as shown)

[0116] Firstly, standard body calibration traceability: Place a standard body with a known volume (such as a calibration pile or standard hopper) in the material yard, measure its volume by splicing point cloud and compare it with the known true value, and perform traceability verification of the volume measurement.

[0117] Secondly, reverse calculation of the true measurement value: the reference volume V is calculated by using the true weight M calibrated by weighing devices such as truck scales and belt scales and the bulk density ρ. ref = M / ρ, compared with the volume measured from the stitched point cloud. The volume relative error is defined as:

[0118]

[0119] Third, multi-radar cross-verification: cross-compare the volumes of overlapping areas calculated by different radars, identify and eliminate abnormal stations, and select the radar with the best quality as the main measurement source for the area.

[0120] Based on the above verification, the measurement accuracy can be quantitatively evaluated using indicators such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Maximum Error Percentage. MAPE is defined as:

[0121]

[0122] In the formula, n is the sample size, and y i With ŷ i These are the reference value and the measured value for the i-th sample, respectively. When the accuracy is not up to standard, recalibration or re-splicing is triggered to form an accuracy closed loop.

[0123] Combination Figures 1 to 4 Explanation of the working principle of this invention:

[0124] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the overall solution of the present invention without creative effort are all within the protection scope of the present invention.

[0125] As attached Figure 2 As shown, this invention provides a method for multi-3D lidar point cloud stitching and measurement accuracy verification for large material yards, specifically including the following steps:

[0126] P1: Multiple 3D LiDARs (preferably 3D LiDARs developed by Harbin Zhiwu Technology Co., Ltd.) are deployed along the large material yard to create an overlapping area on the material surface by the scanning range of adjacent LiDARs; each LiDAR independently collects point clouds of the material surface.

[0127] P2: Deploy several calibration targets with known coordinates or select stable feature features within the material yard, and estimate the external parameters R of each radar by observing them. m t m .

[0128] P3: Based on the external parameters, transform each radar point cloud to the global material field coordinate system.

[0129] P4: In the overlapping area of ​​adjacent radars, ICP and other methods are used to perform fine registration and consistency optimization of point clouds; when the registration residual is large, the external parameters are optimized and re-registered.

[0130] P5: The registered point cloud is subjected to quality-based weighted fusion to generate a complete material surface point cloud that covers the entire field without blind spots.

[0131] P6: Apply volume consistency constraints to the splice seams, remove redundancy in overlapping areas, and ensure continuous and unique volume.

[0132] P7: Perform gridded volume integration on the entire point cloud to obtain the entire volume V, and calculate the inventory weight by combining it with the bulk density model.

[0133] Example 1: Metering of a large strip material yard using three machines (as shown in the attached diagram) Figure 1 , Figure 4 (As shown)

[0134] This embodiment uses a large strip-shaped material yard approximately 200 meters long as the object. Three 3D lidars (R1, R2, R3) developed by Harbin Zhiwu Technology Co., Ltd. are arranged along the length of the yard. The scanning ranges of adjacent radars form an overlap area of ​​approximately tens of meters on the material surface. Four calibration targets with known coordinates are set up along the line within the yard to estimate the extrinsic parameters of each radar.

[0135] After each radar scans independently, the point cloud is unified to the global coordinate system based on extrinsic parameters, and ICP fine registration is performed in the overlapping area. As shown in Figure 4(b), the registration residual RMSE in the overlapping area converges rapidly with ICP iterations, converging to the sub-centimeter level after about ten iterations. After registration, a weighted fusion based on point cloud density and signal-to-noise ratio is used to generate a complete point cloud of the material surface across the entire field.

[0136] As shown in Figure 4(a), compared to using only a single radar R2 for measurement (material surface coverage of approximately 62%, volume relative error of approximately 3.8%), dual-machine splicing can increase the coverage to approximately 88% and reduce the volume relative error to approximately 1.6%. Triple-machine splicing further increases the coverage to approximately 99% and reduces the volume relative error to approximately 0.7%. Combined with the conversion of the bulk density model, the mean absolute percentage error (MAPE) of inventory weight measurement can be controlled within 1.0% (the specific value depends on the equipment and site conditions).

[0137] Example 2: Quality weighting and abnormal station removal under harsh environments

[0138] As a key function of this invention, when localized dust or rain / fog occurs in the material yard, the point cloud density of a radar in the corresponding area decreases and the noise increases. This embodiment evaluates the point cloud quality (point cloud density, incident angle, signal-to-noise ratio, etc.) of each radar at each grid in real time during the fusion process, and dynamically allocates fusion weights accordingly. When the quality of a radar at a certain location falls below a threshold, its weight approaches zero and it is automatically removed. The adjacent radar with better quality then dominates the material surface elevation in that area, thereby ensuring the robustness of the stitched point cloud and volume measurement under harsh conditions.

[0139] Example 3: Target-free feature identification

[0140] As another implementation method, when it is inconvenient to deploy calibration targets on-site, stable feature features within the material yard (such as fixed pillars, corridor supports, and perimeter walls) can be selected as calibration references. Constraints are established by utilizing observations of the same feature feature from various radars to estimate extrinsic parameters, achieving target-free collaborative calibration. When radar deployment, material yard structure, or environmental conditions change significantly, the calibration and stitching process is re-executed. Therefore, the overall solution of this invention is not limited to a specific calibration, registration, or fusion algorithm.

[0141] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. The method described in this invention is not only applicable to the metering of bulk materials such as coal and ore in material yards, but can also be extended to the storage and metering scenarios of various bulk materials such as grain, chemical raw materials, and building aggregates.

[0142] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make other changes within the spirit of the invention and apply it to fields not mentioned in the invention. Of course, all such changes made in accordance with the spirit of the invention should be included within the scope of protection claimed by the invention.

Claims

1. A method for multi-machine point cloud stitching and measurement accuracy verification for large material yards, characterized in that: Includes the following steps: Step 1: Deployment and coordinated calibration of multiple radar stations; Step 11: Deploy multiple 3D LiDARs along the large material yard so that the scanning ranges of adjacent LiDARs overlap on the material surface. Steps 1 and 2: Deploy several calibration targets with known coordinates or select stable feature features within the material yard. Estimate the installation pose of each radar relative to the global material yard coordinate system, i.e., the rotation matrix R, by using the observations of each radar on the targets or feature features. m With translation vector t m The local point clouds collected by each radar are uniformly transformed to the same global coordinate system; Step 2: Fine-tuning the overlapping area; After transforming each radar point cloud to the global coordinate system, the overlapping areas of adjacent radar scanning ranges are precisely registered to eliminate residual errors in external parameter estimation. Step 3: Quality-weighted fusion and splicing; For the registered overlapping area, dynamic weights are constructed based on one or more of the quality indicators of point cloud density, incident angle, and signal-to-noise ratio of each radar at each location. A weighted fusion method is used to generate a unified material surface elevation. The higher the point cloud quality, the greater the weight. When the point cloud quality of a radar at the corresponding location is lower than the threshold, its weight is set to zero and it is automatically removed. In this way, the multi-source point clouds are fused and stitched into a complete material surface point cloud that covers the entire field and is consistent. Step 4: Calculation of splicing volume and consistency constraints; The spliced ​​point cloud of the entire material surface is meshed. Combining the material yard geometric model and the reference plane, the space between the material surface and the reference plane is meshed and summed to obtain the volume of the entire field. Volume consistency constraints are introduced in the overlapping area. The splicing consistency is measured by the average elevation difference between the two radars in the overlapping area. When the average elevation difference exceeds the threshold, re-registration or external parameter optimization is triggered. When calculating the volume, the elevation of the overlapping area is taken as the fused value and only counted once. Step 5: Measurement accuracy verification and closed-loop feedback; The measurement accuracy is verified by a multi-method accuracy verification method that combines standard body calibration traceability, true value back calculation based on weighing device and multi-radar cross-verification. When the accuracy does not meet the standard, recalibration or re-splicing is triggered to form an accuracy closed loop.

2. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: In step two, the fine registration of adjacent overlapping radar regions uses an iterative nearest-point method to optimize the relative pose in the overlapping region with the goal of minimizing the sum of squared distances between point pairs. In the formula, p j With q j R and t represent the corresponding point pairs from the two radars in the overlapping area, respectively, where R and t are the relative rotation and translation to be optimized, and n is the number of point pairs. Let theta be a preset residual threshold. When the registration residual is greater than theta, the external parameters are adjusted and the registration is re-registered until the residual converges.

3. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: In the weighted fusion step three, the weight w of each radar at grid g is... m (g) Based on the aforementioned quality indicators, the elevation z(g) of the fused material surface is determined by the following formula: In the formula, z m (g) represents the surface elevation of the m-th radar at grid g; when the point cloud quality of a radar at grid g is lower than a preset threshold, its weight w m (g) is set to zero.

4. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: The average elevation difference δ in step four is determined by the following formula: In the formula, z a (j), z b (j) represents the surface elevation of the two radars at the j-th grid in the overlapping area, and n represents the number of grids in the overlapping area.

5. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: The total volume V in step four is determined by the following formula: In the formula, G is the total number of grid cells, and z s (g) is the elevation of the material surface at grid g, z0(g) is the elevation of the reference surface at grid g, and Δs is the horizontal area of ​​the grid.

6. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: Step 5, the standard body calibration and traceability, includes: placing a standard body with a known volume in the material yard, measuring the volume of the standard body by splicing point cloud and comparing it with the known true value, and performing traceability verification of the volume measurement.

7. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: Step five involves back-calculating the true value based on the weighing device, which includes calculating the reference volume V from the actual weight M calibrated by the truck scale or belt scale and the bulk density ρ. ref =M / ρ, compared with the volume V measured from the stitched point cloud, the volume relative error is defined as: 。 8. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: Step 5, multi-radar cross-verification, includes: cross-comparing the volumes of the overlapping area calculated by different radars, identifying and eliminating abnormal stations, and selecting the radar with the best quality as the main measurement source for the area.

9. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: Step 5, the metrological accuracy verification, uses one or more of the following indicators to quantitatively evaluate metrological accuracy: mean absolute error, mean absolute percentage error, and maximum error percentage. The mean absolute percentage error is defined as follows: In the formula, n is the sample size, and y i and These are the reference value and the measured value for the i-th sample, respectively.

10. The method for multi-machine point cloud stitching and measurement accuracy verification for large material yards according to claim 1, characterized in that: The three-dimensional lidar is a 3D lidar with explosion-proof certification.

Images