An irregular cargo volume measurement system and method based on multi-source radar fusion

Abstract

The application provides an irregular cargo volume measurement system and method based on multi-source radar fusion. The method belongs to the field of cargo volume measurement, and comprises the following steps: collecting point cloud data of each radar and a yaw angle of a turntable in real time during rotation of the cargo on the turntable, and obtaining rotating point cloud data; constructing a three-dimensional voxel grid for a measurement space, and updating an occupancy probability of each voxel by using a logarithmic probability state updating model to distinguish an internal entity, a surface and an external idle area of an object; performing slicing processing on the three-dimensional voxel grid after the occupancy probability is updated along a height direction, detecting and identifying a gap blind area formed due to laser penetration in each slice layer, and reclassifying voxels that are incorrectly marked as occupied into an idle state; and counting a number of voxels with an occupancy probability greater than a preset threshold in all voxels, and calculating a real physical volume of the cargo according to the number of voxels and a volume of a single voxel. The method can realize high-precision real-time measurement of the cargo volume under low computing power consumption.

Description

Technical Field

[0001] This application relates to the field of cargo volume measurement, and more specifically, to a system and method for measuring the volume of irregular cargo based on multi-source radar fusion. Background Technology

[0002] In logistics warehousing, freight hubs, and industrial manufacturing, cargo volume measurement is fundamental for billing, load optimization, and warehouse space planning. Currently, most mainstream volume measurement solutions employ the "cubic-box" method, which measures the maximum length, width, and height of the cargo and multiplies them. However, in actual operations, many cargoes are irregularly shaped—such as soft sacks filled with agricultural products or chemical raw materials, multiple cardboard boxes of various shapes stacked haphazardly, and mechanical parts with protrusions or indentations. Traditional box measurement methods include external gaps in the cargo's dimensions, leading to overestimation of the volume and failing to reflect the cargo's true physical volume. Furthermore, existing solutions using 3D depth cameras or single LiDAR units have blind spots; a single device may fail to scan the back or recessed areas of the cargo due to occlusion effects, resulting in missing point clouds. Moreover, high-density 3D point clouds consume significant computational resources during multi-frame stitching and volume integration, making it difficult to meet the real-time output requirements of logistics lines. Summary of the Invention

[0003] The purpose of this application is to provide a system and method for measuring the volume of irregular cargo based on multi-source radar fusion, which can achieve high-precision real-time measurement of cargo volume with low computing power consumption.

[0004] This application is implemented as follows: In a first aspect, this application provides a method for measuring the volume of irregular cargo based on multi-source radar fusion, including: S1: Perform joint calibration on multiple single-line lidars, and transform the original point cloud data of each lidar from its own coordinate system to the world coordinate system with the center of the turntable as the origin. S2: The turntable carries the cargo and rotates at least one revolution. During the rotation, the point cloud data of each radar and the yaw angle of the turntable are collected in real time. The point cloud data in the world coordinate system is rotated and transformed according to the yaw angle to obtain the rotating point cloud data in the simulated state of rotating and scanning around the cargo. S3: Construct a three-dimensional voxel mesh in the measurement space. For each laser point in each frame of rotated point cloud data, emit a virtual ray from the optical center of the corresponding radar towards that laser point. Based on the voxels through which the virtual ray passes and the voxel where the endpoint is located, use a logarithmic probability state update model to update the occupancy probability of each voxel to distinguish between internal entities, surfaces, and external vacant areas of an object. In the logarithmic probability state update model, voxels are defined as follows: The probability of occupancy is Its logarithmic probability state quantity Defined as: ; For intermediate voxels traversed by the ray, the idle probability is updated according to the following formula: ; For the voxel where the ray endpoint is located, the occupancy probability is updated according to the following formula: ; in, voxels at the current moment The logarithmic probability state quantity, voxels of the previous moment The logarithmic probability state quantity, The preset hit probability constant, The preset probability constant for crossing; S4: Slice the 3D voxel mesh that has completed the occupancy probability update along the height direction, detect and identify the gap blind zone formed by the laser's inability to penetrate in each slice layer, and reclassify the voxels that have been incorrectly marked as occupied in the gap blind zone as idle state in order to remove false volumes. S5: Count the number of voxels with a probability greater than a preset threshold among all voxels, and calculate the actual physical volume of the cargo based on the number of voxels and the volume of a single voxel.

[0005] Based on the first aspect, in step S1, the coordinate system transformation is expressed as: ;in, This is the raw radar point cloud data. The transformed world coordinate system point cloud data, Let be a rotation matrix. This is the offset.

[0006] Based on the first aspect, in step S1, the joint calibration includes: Adjust the yaw angle of each radar so that the laser illumination surface of that radar is perpendicular to the plane of the turntable and passes through the center point of the turntable; Adjust the roll angle of each radar so that the horizontal and vertical directions of its laser point cloud are parallel to the Y-axis and Z-axis of the world coordinate system, respectively. The translation vectors of each radar in the world coordinate system are calibrated.

[0007] Based on the first aspect, in step S2, let the yaw angle be... The corresponding rotation matrix Represented as: ; Let the coordinates of the point cloud after transformation be... Combined with the rotation matrix obtained from calibration and offset Rotated point cloud data can be obtained. Compared with raw radar point cloud data The relationship is represented as: .

[0008] Based on the first aspect, the preset hit probability constant in step S3 The preset pass-through probability constant is 0.7. It is 0.4.

[0009] Based on the first aspect, step S3 also includes: processing the voxel at the current time. Log-probability state quantity Set upper and lower thresholds to prevent probability-based deadlock.

[0010] Based on the first aspect, in step S4, the specific steps for detecting and identifying gap blind areas formed by laser penetration in each slice layer, and reclassifying voxels that have been incorrectly marked as occupied in the gap blind areas as free, in order to remove false volumes, include: Extract the local point set of the identified gap edge R represents a real number, and N represents the total number of two-dimensional laser points contained within the local point set of the extracted and identified slit boundary line in the current slice layer; after calculating the centroid and decentering, we obtain: ; ; in, This represents the original coordinates of the i-th point in the set of gap boundary line points on the current slice 2D image. ; This represents the coordinates of the geometric centroid of the set of points representing the gap boundary line on the current slice of the 2D image. ; This represents the relative coordinate vector of the i-th boundary line point after centroid subtraction and decentering. Construct the matrix , represented as: ; SVD decomposition Where U represents the left singular matrix, Σ represents the singular value diagonal matrix, and V represents the right singular matrix; take the right singular matrix The first column is used as the principal direction vector of the line. Then, the slope of the straight line equation parameters on both sides of the gap can be obtained. and intercept , represented as: ; ; The equation of the corresponding line is For each gap, two straight lines are obtained. and ; For unknown points inside the gap blind zone Calculate its distance to the imaginary reference closure line outside the gap. vertical distance , represented as: ; in, , These represent the hypothetical reference closure lines. The equation of the straight line, its slope, and its intercept; Assigning probability weights to it based on a Gaussian distribution, indicating that it belongs to the outer free space. , represented as: ; Among them, variance This represents the preset Gaussian distribution variance, a manually set hyperparameter used to control the sensitivity and range of voxels within the gap being identified as free. A larger value results in more voxels being identified as free within the gap. Whether a voxel within the gap is considered free depends on the hyperparameter threshold. To determine whether a voxel is free, we can use the following method: voxels with values ​​greater than the threshold are considered free. Let the endpoints of the gap and their mathematical projections on the opposite side form a quadrilateral, with the clockwise vertex being... ; Represents the coordinates of the four vertices of a clockwise convex quadrilateral formed by the two ends of the gap and their mathematical projections on the opposite sides; Determines the projected coordinates of unknown blind zone points. Whether a space is completely inside a deep crevice depends on the following four conditions being met: The cross product of these four vectors must have the same sign. ; If satisfied, then the unknown blind spot point If the elimination weight is forcibly set to 1.0, then the unknown blind spot point is marked. It is an empty voxel.

[0011] Secondly, this application provides a system for measuring the volume of irregular cargo based on multi-source radar fusion, comprising: The multi-source data acquisition module includes at least two single-line lidars, which are mounted on the measuring area bracket at different roll angles. A turntable is used to carry goods and rotate them. The computing unit is connected to the multi-source data acquisition module and the turntable for communication purposes, and is used to execute the methods described above.

[0012] Thirdly, this application provides an electronic device, comprising: Memory, used to store one or more programs; processor; The above method is implemented when one or more programs are executed by the processor.

[0013] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0014] Compared with the prior art, this application has at least the following advantages or beneficial effects: This invention provides a method and system for measuring the volume of irregular goods based on multi-source radar fusion. Through probabilistic voxel meshing and spatial engraving technology, it accurately distinguishes the internal solids, surfaces, and external empty areas of an object, and removes false volumes incorrectly marked due to blind spots in fan-shaped scanning. The synergy of these two techniques ensures that the final measured volume closely matches the actual physical shape of the goods, with an error far less than that of the traditional cuboid bounding box method. A logarithmic probability state update model is employed, effectively suppressing noise interference such as missing point clouds caused by black light-absorbing packaging or diffraction flypoints caused by metallic reflections through multi-frame probability accumulation, ensuring the stability and reliability of the volume measurement results. Using a single-line lidar instead of a high-beam radar, combined with voxel meshing and an efficient ray projection algorithm, the data volume is far lower than that of high-density 3D point clouds. Real-time volume calculation and mesh updates can be achieved on an embedded intelligent terminal, meeting the real-time requirements of high-speed logistics operations. In summary, this invention, from multi-source data fusion, probabilistic modeling, false volume removal to voxel statistical output, has each step closely linked and mutually coordinated, jointly achieving rapid and accurate measurement of the true physical volume of irregular goods. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of an embodiment of a method for measuring the volume of irregular cargo based on multi-source radar fusion according to this application; Figure 2 This is a schematic diagram of the hardware structure in one embodiment of the irregular cargo volume measurement method based on multi-source radar fusion according to this application; Figure 3 This is a top view of a turntable and cargo in one embodiment of a method for measuring the volume of irregular cargo based on multi-source radar fusion according to this application. The notch on the left is a gap. Figure 4This is a slice of a single-line lidar point cloud in one embodiment of a method for measuring the volume of irregular cargo based on multi-source radar fusion in this application. The red area is the radar blind zone, which is assumed to be the voxel occupied inside the box. Figure 5 In one embodiment of the irregular cargo volume measurement method based on multi-source radar fusion of this application, straight lines L1 and L2 are obtained by fitting local points, and L3 is obtained by the gap edge points; Figure 6 This is an embodiment of a method for measuring the volume of irregular cargo based on multi-source radar fusion according to this application. Schematic diagram; Figure 7 This is a schematic diagram of the structure of an electronic device according to this application.

[0017] icon: 1. A hybrid solid-state single-line lidar; 2. A second hybrid solid-state single-line lidar; 3. A computing unit; 4. A turntable; 5. Cargo; 6. A processor; 7. A memory; 8. A communication interface. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the various embodiments and features described below can be combined with each other. Example

[0020] This application provides a system and method for measuring the volume of irregular cargo based on multi-source radar fusion, which can achieve high-precision real-time measurement of cargo volume with low computing power consumption.

[0021] Please refer to Figure 1 , Figure 1 The flowchart of the method in this application shows that the method for measuring the volume of irregular cargo based on multi-source radar fusion includes the following steps: S1: Perform joint calibration on multiple single-line lidars, and transform the original point cloud data of each lidar from its own coordinate system to the world coordinate system with the center of the turntable as the origin. Please refer to Figure 2 , Figure 2This is a schematic diagram of the hardware structure of this application. This embodiment uses two hybrid solid-state single-line lidars, namely a first hybrid solid-state single-line lidar 1 and a second hybrid solid-state single-line lidar 2. Both the first hybrid solid-state single-line lidar 1 and the second hybrid solid-state single-line lidar 2 are connected to the computing unit 3. A turntable 4 is used to carry the cargo 5 and rotate it. The shape of the cargo 5 is not specifically limited and can be various irregular shapes. The two hybrid solid-state single-line lidars are installed above and below the measurement station. The first hybrid solid-state single-line lidar is installed approximately 2.5 meters above the ground at an angle of approximately 45 degrees downwards; the second hybrid solid-state single-line lidar is installed approximately 0.3 meters above the ground horizontally.

[0022] Since each radar's point cloud data originates from its own sensor, the coordinate system of each radar needs to be transformed to a world coordinate system with the turntable center as the origin before joint measurement. The coordinate system transformation consists of two parts: a rotation matrix and a translation vector, expressed as follows: ;in, This is the raw radar point cloud data. The transformed world coordinate system point cloud data, Let be a rotation matrix. This is the offset.

[0023] The specific calibration process includes: The yaw angle of each radar is adjusted so that the laser illumination surface of the radar is perpendicular to the plane of the turntable and passes through the center point of the turntable. Specifically, after starting the first hybrid solid-state single-line lidar, the second hybrid solid-state single-line lidar and the computing unit, the light rays of the laser illuminating the turntable are displayed with the help of an infrared camera. The lidar is moved so that the light rays displayed on the turntable pass through the center point of the turntable, so that the laser illumination surface is perpendicular to the plane of the turntable and passes through the center point of the turntable.

[0024] Adjust the roll angle of each radar so that the horizontal and vertical directions of its laser point cloud are parallel to the Y-axis and Z-axis of the world coordinate system, respectively. Specifically, place a box on the turntable and use laser data visualization software to display the effect of the laser in the world coordinate system. Adjust the roll angle to make the horizontal and vertical directions of the laser point cloud parallel to the Y-axis and Z-axis of the world coordinate system, respectively.

[0025] The translation vectors of each radar in the world coordinate system are calibrated. Specifically, after the yaw angle calibration is completed, the offset of the laser in the X-axis direction is 0. By adjusting the offsets in the Y-axis and Z-axis directions, the position of the origin in the laser radar point cloud is made to coincide with the position of the origin in the world coordinate system.

[0026] Repeat the above calibration steps for the first hybrid solid-state single-line lidar and the second hybrid solid-state single-line lidar respectively to complete all joint calibrations.

[0027] By unifying the coordinate systems of multiple radars to the same world coordinate system through joint calibration, coordinate system conflicts between multi-source data are eliminated, providing a precise spatial reference for subsequent fusion processing of multi-radar point cloud data and ensuring the accuracy of subsequent volume measurement results.

[0028] S2: The turntable carries the cargo and rotates at least one revolution. During the rotation, the point cloud data of each radar and the yaw angle of the turntable are collected in real time. The point cloud data in the world coordinate system is rotated and transformed according to the yaw angle to obtain the rotating point cloud data in the simulated state of rotating and scanning around the cargo. Specifically, the forklift places the stacked sacks onto an industrial turntable, which then begins to rotate at a constant speed, with a rotation angle of at least 360 degrees. During the rotation, the computing unit collects point cloud data from two radars and the yaw angle fed back by the turntable encoder in real time.

[0029] Let the yaw angle be The corresponding rotation matrix Represented as: ; Let the coordinates of the point cloud after transformation be... Combined with the rotation matrix obtained from calibration and offset Rotated point cloud data can be obtained. Compared with raw radar point cloud data The relationship is represented as: .

[0030] As the turntable rotates continuously, the point cloud data acquired by each radar at different angles is converted and accumulated frame by frame, forming a complete point cloud coverage of the cargo surface. This step utilizes the 360-degree rotation of the turntable, enabling the single-line lidar to scan the cargo at different angles. Combined with the real-time rotational transformation of the yaw angle, it effectively achieves omnidirectional scanning of the cargo by the radar, effectively compensating for the limited field of view of a single radar and providing a complete source of point cloud data for subsequent 3D voxel reconstruction.

[0031] Considering that a single lidar scan can only acquire point clouds of the object's surface, this step introduces a 3D voxel grid and fast raycasting method to distinguish between "internal entities," "surface," and "external air." The measurement space is divided into a 3D voxel grid of specified side lengths. Instead of using a simple "0 or 1" binary state for each voxel, a log-odds spatial state update model is employed to eliminate diffraction flypoints and penetration noise caused by surface materials (such as reflectivity and black light absorption). Please refer to step S3 below for details: S3: Construct a three-dimensional voxel mesh in the measurement space. For each laser point in each frame of rotated point cloud data, emit a virtual ray from the optical center of the corresponding radar towards that laser point. Based on the voxels through which the virtual ray passes and the voxel where the endpoint is located, use a logarithmic probability state update model to update the occupancy probability of each voxel to distinguish between internal entities, surfaces, and external vacant areas of an object. In the logarithmic probability state update model, voxels are defined as follows: The probability of occupancy is Its logarithmic probability state quantity Defined as: ; For intermediate voxels traversed by the ray, the idle probability is updated according to the following formula: ; For the voxel where the ray endpoint is located, the occupancy probability is updated according to the following formula: ; in, voxels at the current moment The logarithmic probability state quantity, voxels of the previous moment The logarithmic probability state quantity, The preset hit probability constant, The preset probability constant for crossing; Furthermore, the preset hit probability constant The preset pass-through probability constant is 0.7. The value is 0.4. For the current voxel... Log-probability state quantity Set upper and lower thresholds to prevent probability-based deadlock.

[0032] After a voxel has been updated multiple times, its logarithmic probability state value will stop changing once it reaches its upper or lower limit. After the turntable rotates more than 360 degrees, the cumulative update stops. At this point, the logarithmic probability state value of each voxel has stabilized and can effectively distinguish between "internal entities of an object", "surface" and "external air".

[0033] This step employs a log-probability state update model instead of a simple binary state, allowing the occupancy probability of each voxel to accumulate and optimize continuously across multiple observation frames. This effectively suppresses noise interference caused by surface material issues such as missing point clouds due to black light-absorbing packaging or diffraction flypoints caused by metallic reflections. The probability update method ensures the robustness and smoothness of the state estimation, providing a reliable data foundation for the subsequent accurate calculation of the actual physical volume.

[0034] Further research revealed that during rotating scanning, the illumination surface of the single-line lidar passes through the center of the turntable, exhibiting a fan-shaped distribution in space. When scanning multiple irregularly stacked cartons or goods with deep grooves, because the light is blocked by the outer edges, there will inevitably be blind spots that the laser cannot reach for all physical gaps. Please refer to [reference needed]. Figure 3 and Figure 4 , Figure 3 This is a top view of a turntable and cargo in one embodiment of a method for measuring the volume of irregular cargo; the notch on the left is a gap. Figure 4 This is a slice of a single-line lidar point cloud. The red area represents the lidar blind zone, which is assumed to be the occupied voxels inside the container. In the spatial sculpting algorithm of the lidar ray, the illuminated surface is marked as "Hit," the space traversed by the ray is marked as "Miss," and the blind zone that the light cannot reach remains in the initial "Unknown" state. Since the actual solid interior of the cargo also lacks light illumination and is in an "Unknown" state, gap blind zones are misclassified as solid regions similar to the interior of the object. If left untreated, these gaps, which should be external air, will be accumulated and integrated, resulting in an abnormally large calculated volume.

[0035] To address the volume expansion issue caused by the physical characteristics of sector scanning, this invention designs a gap filtering algorithm. This algorithm slices the 3D point cloud along the Z-axis using a multi-segment RDP algorithm, principal component analysis (PCA), and Gaussian weight distribution to accurately identify and forcibly remove these spurious volumes. Please refer to step S4 below for details: S4: Slice the 3D voxel mesh that has completed the occupancy probability update along the height direction, detect and identify the gap blind zone formed by the laser's inability to penetrate in each slice layer, and reclassify the voxels that have been incorrectly marked as occupied in the gap blind zone as idle state in order to remove false volumes. The specific steps involved in detecting and identifying gaps and blind spots formed in each slice layer due to laser penetration, and reclassifying voxels that have been incorrectly marked as occupied in these gap and blind spots as idle, in order to remove false volumes, include: Extract the local point set of the identified gap edge R represents a real number, and N represents the total number of two-dimensional laser points contained within the local point set of the extracted and identified slit boundary line in the current slice layer; after calculating the centroid and decentering, we obtain: ; ; in, This represents the original coordinates of the i-th point in the set of gap boundary line points on the current slice 2D image. ; This represents the coordinates of the geometric centroid of the set of points representing the gap boundary line on the current slice of the 2D image. ; This represents the relative coordinate vector of the i-th boundary line point after centroid subtraction; (The rest of the text appears to be a fragment and requires further context for accurate translation.) Construct the matrix , represented as: ; SVD decomposition Where U represents the left singular matrix, Σ represents the singular value diagonal matrix, and V represents the right singular matrix; take the right singular matrix The first column is used as the principal direction vector of the line. Then, the slope of the straight line equation parameters on both sides of the gap can be obtained. and intercept , represented as: ; ; The equation of the corresponding line is For each gap, two straight lines are obtained. and ,like Figure 5 As shown, Figure 5 Lines L1 and L2 are obtained by fitting local points, and L3 is obtained from the edge points of the gap. For unknown points inside the gap blind zone Calculate its distance to the imaginary reference closure line outside the gap. vertical distance , represented as: ; in, , These represent the hypothetical reference closure lines. The equation of the straight line, its slope, and its intercept; Assigning probability weights to it based on a Gaussian distribution, indicating that it belongs to the outer free space. , represented as: ; Among them, variance This represents the preset Gaussian distribution variance, a manually set hyperparameter used to control the sensitivity and range of voxels within the gap being identified as free. A larger value results in more voxels being identified as free within the gap. Whether a voxel within the gap is considered free depends on the hyperparameter threshold. To determine whether a voxel is free, we can use the following method: voxels with values ​​greater than the threshold are considered free. Let the endpoints of the gap and their mathematical projections on the opposite side form a quadrilateral, with the clockwise vertex being... Please refer to Figure 6 , Figure 6 for Schematic diagram; showing the coordinates of the four vertices of a clockwise convex quadrilateral formed by the two ends of the gap and their mathematical projections on the opposite sides; determining the projected coordinates of unknown blind zone points. Whether a space is completely inside a deep crevice depends on the following four conditions being met: The cross product of these four vectors must have the same sign. ; If satisfied, then the unknown blind spot point If the elimination weight is forcibly set to 1.0, then the unknown blind spot point is marked. It is an empty voxel.

[0036] This step addresses the gap / blind zone problem caused by the physical characteristics of single-line LiDAR fan-shaped scanning. Through dimensionality reduction slicing and edge contour analysis, it accurately identifies false volume regions incorrectly marked as "occupied." Using a multi-segment RDP algorithm and principal component analysis to extract gap boundaries, and combining Gaussian distribution weights and a deep blind zone detection mechanism, false voxels are removed from the occupied state. This effectively solves the problem of abnormal volume expansion caused by stacking depressions, ensuring that the final volume measurement results reflect the true physical shape of the goods.

[0037] S5: Count the number of voxels with a probability greater than a preset threshold among all voxels, and calculate the actual physical volume of the cargo based on the number of voxels and the volume of a single voxel.

[0038] The calculated volume results are uploaded to the host computer for display and recording, and used for subsequent billing, load optimization, or warehouse space planning. By calculating volume through voxel accumulation, compared to the traditional cuboid bounding box measurement method, this method can accurately determine the actual space volume occupied by the goods, eliminating invalid volumes from gaps and recessed areas outside the goods. The measurement results closely match the actual physical shape of the goods, providing accurate data for logistics billing and warehouse management.

[0039] In some embodiments of the present invention, an irregular cargo volume measurement system based on multi-source radar fusion is also provided, comprising: The multi-source data acquisition module includes at least two single-line lidars, which are mounted on the measurement area bracket at different roll angles; the two single-line lidars are a first hybrid solid-state single-line lidar 1 and a second hybrid solid-state single-line lidar 2. Turntable 4 is used to carry goods 5 and drive goods 5 to rotate; The computing unit 3 is communicatively connected to the multi-source data acquisition module and the turntable 4, respectively, and is used to execute all or part of the above methods.

[0040] Please refer to Figure 7 In some embodiments of the present invention, an electronic device is also provided, comprising: Memory 7 is used to store one or more programs; Processor 6; Processor 6 is connected to memory 7 via communication interface 8; When one or more programs are executed by processor 6, all or some of the above methods are implemented.

[0041] In some embodiments of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor 6, implements all or part of the methods described above.

[0042] The memory 7 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0043] Processor 6 can be an integrated circuit chip with signal processing capabilities. This processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0044] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for measuring the volume of irregular cargo based on multi-source radar fusion, characterized in that, include: S1: Perform joint calibration on multiple single-line lidars, and transform the original point cloud data of each lidar from its own coordinate system to the world coordinate system with the center of the turntable as the origin. S2: The turntable carries the cargo and rotates at least one revolution. During the rotation, the point cloud data of each radar and the yaw angle of the turntable are collected in real time. The point cloud data in the world coordinate system is rotated and transformed according to the yaw angle to obtain the rotating point cloud data in the simulated rotating scanning state around the cargo. S3: Construct a three-dimensional voxel mesh in the measurement space. For each laser point in each frame of the rotated point cloud data, emit a virtual ray from the optical center of the corresponding radar towards that laser point. Based on the voxels through which the virtual ray passes and the voxel where the endpoint is located, update the occupancy probability of each voxel using a logarithmic probability state update model to distinguish between internal entities, surfaces, and external vacant areas of an object. In the logarithmic probability state update model, voxels are defined as follows: The probability of occupancy is Its logarithmic probability state quantity Defined as: ； For intermediate voxels traversed by the ray, the idle probability is updated according to the following formula: ； For the voxel where the ray endpoint is located, the occupancy probability is updated according to the following formula: ； in, voxels at the current moment Log-probability state quantity, voxels of the previous moment Log-probability state quantity, The preset hit probability constant, The preset probability constant for crossing; S4: The three-dimensional voxel mesh that has completed the occupancy probability update is sliced ​​along the height direction. In each slice layer, gap blind areas formed due to the inability of the laser to penetrate are detected and identified. Voxels that have been incorrectly marked as occupied in the gap blind areas are reclassified as idle to remove false volumes. S5: Count the number of voxels with a probability greater than a preset threshold among all voxels, and calculate the actual physical volume of the cargo based on the number of voxels and the volume of a single voxel.

2. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 1, characterized in that, In step S1, the coordinate system transformation is expressed as: ;in, This is the raw radar point cloud data. The transformed world coordinate system point cloud data, For rotation matrix, This is the offset.

3. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 1, characterized in that, In step S1, the joint calibration includes: Adjust the yaw angle of each radar so that the laser illumination surface of that radar is perpendicular to the plane of the turntable and passes through the center point of the turntable; Adjust the roll angle of each radar so that the horizontal and vertical directions of its laser point cloud are parallel to the Y-axis and Z-axis of the world coordinate system, respectively. The translation vectors of each radar in the world coordinate system are calibrated.

4. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 2, characterized in that, In step S2, let the yaw angle be... The corresponding rotation matrix Represented as: ； Let the coordinates of the point cloud after transformation be... Combined with the rotation matrix obtained from calibration and offset Rotated point cloud data can be obtained. Compared with the original radar point cloud data The relationship is represented as: 。 5. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 1, characterized in that, The preset hit probability constant mentioned in step S3 The preset probability constant for crossing is 0.

7. It is 0.

4.

6. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 1, characterized in that, Step S3 also includes: processing the voxel at the current time. Log-probability state quantity Set upper and lower thresholds to prevent probability-based deadlock.

7. The method for measuring the volume of irregular cargo based on multi-source radar fusion according to claim 1, characterized in that, In step S4, the specific steps for detecting and identifying gap blind zones formed in each slice layer due to the inability of the laser to penetrate, and reclassifying voxels that have been incorrectly marked as occupied in the gap blind zones as free, to remove false volumes include: Extract the local point set of the identified gap edge R represents a real number, and N represents the total number of two-dimensional laser points contained within the local point set of the extracted and identified slit boundary line in the current slice layer; after calculating the centroid and decentering, we obtain: ； ； in, This represents the original coordinates of the i-th point in the set of gap boundary line points on the current slice 2D image. ; This represents the coordinates of the geometric centroid of the set of points representing the gap boundary line on the current slice of the 2D image. ; This represents the relative coordinate vector of the i-th boundary line point after centroid subtraction and decentering. Construct the matrix , represented as: ； SVD decomposition Where U represents the left singular matrix, Σ represents the singular value diagonal matrix, and V represents the right singular matrix; take the right singular matrix The first column is used as the principal direction vector of the line. Then, the slope of the straight line equation parameters on both sides of the gap can be obtained. and intercept , represented as: ； ； The equation of the corresponding line is For each gap, two straight lines are obtained. and ; For unknown points inside the gap blind zone Calculate its distance to the imaginary reference closure line outside the gap. vertical distance , represented as: ； in, , These represent the hypothetical reference closure lines. The equation of the straight line, its slope, and its intercept; Assigning probability weights to it based on a Gaussian distribution, indicating that it belongs to the outer free space. , represented as: ； Among them, variance This represents the preset Gaussian distribution variance, a manually set hyperparameter used to control the sensitivity and range of voxels within the gap being identified as free. A larger value results in more voxels being identified as free within the gap. Whether a voxel within the gap is considered free depends on the hyperparameter threshold. To determine whether a voxel is free, we can use the following method: voxels with values ​​greater than the threshold are considered free. Let the endpoints of the gap and their mathematical projections on the opposite side form a quadrilateral, with the clockwise vertex being... ; Represents the coordinates of the four vertices of a clockwise convex quadrilateral formed by the two ends of the gap and their mathematical projections on the opposite sides; Determines the projected coordinates of unknown blind zone points. Whether a space is completely inside a deep crevice depends on the following four conditions being met: The cross product of these four vectors must have the same sign. ； If satisfied, then the unknown blind spot point If the elimination weight is forcibly set to 1.0, then the unknown blind spot point is marked. It is an empty voxel.

8. A system for measuring the volume of irregular cargo based on multi-source radar fusion, characterized in that, include: The multi-source data acquisition module includes at least two single-line lidars, which are mounted on the bracket in the measurement area at different roll angles. A turntable is used to carry goods and rotate them. The computing unit is communicatively connected to the multi-source data acquisition module and the turntable, respectively, and is used to execute the method as described in any one of claims 1-7.

9. An electronic device, characterized in that, include: Memory, used to store one or more programs; processor; When the one or more programs are executed by the processor, the method as described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images