Train material volume detection method, device and equipment based on millimeter wave radar

Abstract

The invention provides a train material volume detection method, device and equipment based on millimeter wave radar, and relates to the technical field of railway transportation intelligent detection and three-dimensional point cloud data processing. The method comprises the following steps: acquiring point cloud data obtained by scanning a train through a millimeter wave radar, and preprocessing the point cloud data to obtain a train main body point cloud; wherein the point cloud data comprises position coordinates of points; on the basis of density change characteristics of the train main body point cloud in the train advancing direction, carriage connection gaps are recognized; based on the compartment connection gap and the preset compartment length, compartment section point clouds are divided from the train main body point clouds; and based on the coordinates of the points in the compartment section point clouds in the train height direction, material surface point clouds are extracted, and material volume calculation is carried out based on the material surface point clouds. The system can stably run in a loading building environment with large dust, strong shielding and complex illumination, and provides a reliable technical means for real-time loading monitoring of bulk materials of a train.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection and three-dimensional point cloud data processing technology in railway transportation, and in particular to a method, device and equipment for detecting the volume of train materials based on millimeter-wave radar. Background Technology

[0002] Point cloud data processing and volume estimation technologies have been applied to some extent in fields such as material yard measurement, building surveying and industrial inspection, but these methods still have significant limitations in the complex environment of railway loading buildings.

[0003] In their patent "A Method for Detecting the Volume and Bulk Density of Cargo Cars Based on Point Cloud Data Processing" (CN109029254B), Zhang Na et al. proposed using a laser scanner to collect point clouds of the car cars and then measuring the volume through the removal of non-cargo points, the construction of triangulation networks, and the integration of height differences. Although this method is applicable to railway open wagons with regular structures, it relies on high-density point clouds from lidar, making it sensitive to dust, strong reflections, and ambient light. In environments with high dust levels and multiple metal components in loading towers, the model's stability is limited, making it difficult to guarantee accurate reconstruction of material surfaces.

[0004] Another patent, "An Automatic Measurement Method for Bulk Material Stack Volume Based on Point Cloud" (CN112345678A), estimates the stockpile volume through voxelization. The voxel method requires high point cloud density and is particularly prone to discretization errors in boundary areas. Furthermore, its algorithm is more suitable for open material yards and does not have good adaptability to the "enclosed space + multiple obstructions" structure inside the car.

[0005] In academic research, Ding Weili proposed estimating the volume of bulk materials using slice-based and projection methods in his paper "A fast volume measurement method for obtaining point cloud data from bulk stockpiles". This method relies on regular slice layering and assumes a relatively smooth material surface; therefore, the accuracy of volume estimation tends to decrease when facing stacking scenarios with highly abrupt changes, complex structures, or sparse data. Another related paper, "Point Cloud Reconstruction and Volume Measurement Method for Dynamic Bulk Material Flow", proposed a point cloud reconstruction framework for dynamic bulk material flows, but it focuses on continuous flow scenarios such as conveyor belts and is insufficiently adaptable to static stacking conditions like railway carriages with irregular surfaces and complex occlusions. Summary of the Invention

[0006] This invention provides a method, apparatus, and equipment for detecting the volume of train materials based on millimeter-wave radar, in order to solve the problem of poor stability and accuracy of train material volume detection based on vision and lidar.

[0007] In a first aspect, embodiments of the present invention provide a method for detecting the volume of materials on a train based on millimeter-wave radar, comprising: The point cloud data obtained by scanning the train using millimeter-wave radar is acquired and preprocessed to obtain the main body point cloud of the train; the point cloud data includes the position coordinates of the points. Based on the density variation characteristics of the train's main point cloud in the direction of train travel, the gaps between carriages are identified. Based on the gap between carriages and the preset carriage length, the point cloud of carriage sections is divided in the point cloud of the main body of the train; Based on the coordinates of the midpoint of the point cloud in the carriage section along the train's height, the point cloud of the material surface is extracted, and the material volume is calculated based on the point cloud of the material surface.

[0008] In one possible implementation, the point cloud data also includes the velocity of the points; preprocessing includes displacement compensation; displacement compensation of the point cloud data includes: The point cloud data is divided into multiple point cloud groups, each point cloud group is numbered according to time sequence, and the velocity components of all points in each point cloud group in the direction of train movement are counted; wherein, a point cloud group is defined as N consecutive frames of point cloud data, and N is a preset positive integer. Points with velocity components greater than 0 in each point cloud group are identified as moving points, and the proportion of moving points in each point cloud group is calculated. If the proportion of moving points in any point cloud group is greater than or equal to a preset threshold, then based on the number of the point cloud group and the preset compensation step size, all points in the point cloud group are translated in the opposite direction of the train's forward movement to obtain the displacement-compensated point cloud data.

[0009] In one possible implementation, preprocessing includes multi-stage filtering; multi-stage filtering of the point cloud data includes: Remove points whose position coordinates are outside the preset working range from the point cloud data to obtain point cloud data after physical boundary filtering; The neighborhood distance of each point in the point cloud data after physical boundary filtering is statistically analyzed, and points whose difference between their neighborhood distance and the average neighborhood distance exceeds a preset distance threshold are removed to obtain the statistically filtered point cloud data. Based on the DBSCAN filtering algorithm, the statistically filtered point cloud data is clustered into multiple clusters, and clusters with fewer than a preset number threshold and / or a vertical length less than a preset length threshold are removed to obtain multi-level filtered point cloud data.

[0010] In one possible implementation, preprocessing includes width center calibration; performing width center calibration on the point cloud data includes: Select a section of a first preset length in the direction of train travel; For all points within the interval, extract the coordinate values ​​in the train width direction to form a width sample set; Remove extreme values ​​from the width sample set and calculate the median or weighted average of the remaining data as the center of the train width; Subtract the center value of the train width from the coordinates of all points in the point cloud data along the train width direction to obtain the point cloud data after width center calibration.

[0011] In one possible implementation, the gaps between carriages are identified based on the density variation characteristics of the train's main point cloud in the direction of train travel, including: Along the direction of train travel, the point cloud of the main body of the train is divided into multiple intervals of a second preset length, and the number of point clouds in each interval is counted. The number of point clouds in each interval is converted into point cloud density, and then arranged according to the spatial order of the intervals to obtain a density sequence; For each minimum value in the density sequence, if the slope of the minimum value is less than a preset slope threshold, then the interval where the minimum value is located is determined to be a gap between carriages.

[0012] In one possible implementation, the material surface point cloud is extracted based on the coordinates of the midpoint of the point cloud in the carriage section along the train's height, including: In the point cloud of the carriage section, select the height statistics window of the third preset length; For all points within the height statistics window, extract the coordinate values ​​in the height direction of the train to form a height sample set; Remove outliers from the high-value sample set and count the maximum and minimum values ​​of the remaining data; Add the preset allowance to the maximum value and subtract the preset allowance from the minimum value to form the effective height range; By retaining the points in the point cloud of the carriage section whose coordinate values ​​in the train's height direction are within the effective height range, the point cloud of the material surface is obtained.

[0013] In one possible implementation, material volume calculation is performed based on the point cloud of the material surface, including: The projection of the point cloud on the material surface onto the YZ plane is used as a set of nodes, and a mesh structure covering the entire interior area of ​​the carriage is generated using the two-dimensional Delaunay triangulation method, resulting in multiple triangular elements; where the YZ plane is the plane formed by the train's forward direction and the train's width direction. For each triangular unit, if the vertex span of the triangular unit is greater than a preset span threshold, the triangular unit is marked as an abnormal unit and removed. For each remaining triangle, if the height difference of the vertices of the normal unit is negative, then it is cut off with the plane of the carriage floor as the boundary, and the remaining part after the cut is divided into triangular units. Calculate the product of the projected area and average height of each triangular element on the Y–Z plane to obtain the material volume corresponding to that triangular element; Calculate the sum of the material volumes corresponding to all triangular units to obtain the material volume corresponding to the point cloud of the carriage section.

[0014] Secondly, embodiments of the present invention provide a train material volume detection device based on millimeter-wave radar, comprising: The acquisition module is used to acquire point cloud data obtained by scanning the train with millimeter-wave radar and to preprocess it to obtain the main body point cloud of the train; wherein, the point cloud data includes the position coordinates of the points; The gap recognition module is used to identify the gaps between carriages based on the density variation characteristics of the train's main point cloud in the direction of train travel. The carriage segmentation module is used to divide the carriage segment point cloud into the main train point cloud based on the carriage connection gap and the preset carriage length. The volume calculation module is used to extract the material surface point cloud based on the coordinates of the midpoint of the point cloud of the carriage section in the direction of train height, and to calculate the material volume based on the material surface point cloud.

[0015] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0017] The present invention provides a method, apparatus, and equipment for detecting train material volume based on millimeter-wave radar. By employing millimeter-wave radar as the core acquisition method, it effectively utilizes its advantages of stable operation under harsh conditions such as rain, snow, fog, dust, and strong sunlight, as well as its wide coverage and strong environmental adaptability. Through point cloud decoding, filtering, and dynamic alignment, a stable point cloud of the train carriages and materials is obtained, solving problems such as sparse point clouds, multipath reflection, and high noise point ratios in millimeter-wave radar. Combining the train's morphological characteristics, the density variation features of the point cloud are used to identify the carriage connections and achieve automatic segmentation, providing a good data foundation for material volume calculation and ensuring the accuracy of the volume calculation. This invention can operate stably in loading building environments with high dust levels, strong obstruction, and complex lighting, providing a reliable technical means for real-time monitoring of the loading of bulk materials on trains. Attached Figure Description

[0018] Figure 1 This is an application scenario diagram of the train material volume detection method based on millimeter-wave radar provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the implementation of the train material volume detection method based on millimeter-wave radar provided in this embodiment of the invention. Figure 3 This is a schematic diagram of the 4D millimeter-wave radar deployment and a schematic diagram of the radar coordinate system being converted into the world coordinate system, provided in an embodiment of the present invention. Figure 4 This is the original three-dimensional point cloud data after coordinate system transformation provided in the embodiments of the present invention; Figure 5 This is a schematic diagram of the overall frame of the train obtained after the algorithm provided in this embodiment of the invention identifies the gaps between carriages; Figure 6 This is a schematic diagram of the three-dimensional point cloud data of the carriage directly below and in front of the radar, provided by an embodiment of the present invention. Figure 7 This is a schematic diagram of the reference plane for constructing a Delaunay triangulation and the bottom of the carriage using three-dimensional point cloud data provided in this embodiment of the invention; Figure 8 This is a schematic diagram of the actual environment of the train after loading, provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of the train material volume detection device based on millimeter-wave radar provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] Based on the above analysis, the application of existing technologies in railway loading terminal environments still faces the following common problems: (1) Both visual and lidar are susceptible to dust, reflection and occlusion, resulting in unstable point cloud quality; (2) Most methods rely on regular grids, voxels or slice models, which are insufficient for the detailed reproduction of the real load surface; (3) It is difficult to achieve accurate dynamic alignment of point clouds across frames when the train is moving slowly, which easily produces ghosting and ghosting. (4) Lack of a complete processing chain from parsing the original sensor logs to car identification, region division, surface reconstruction and volume calculation.

[0020] Compared with the above solutions, this invention uses 4D millimeter-wave radar as the core acquisition method, which has irreplaceable advantages over other sensors: millimeter waves have good penetration ability and environmental robustness, and can work stably under harsh conditions such as rain, snow, fog, dust, and strong light; its echo signal does not depend on the imaging light and will not fail due to lens contamination or smoke and dust obstruction. Although millimeter-wave point clouds are relatively sparse, they have a wide coverage area and strong environmental adaptability, making them very suitable for use in scenarios such as railway loading buildings with dense dust, strong reflection, and complex spatial structures.

[0021] To address the challenges of sparse point clouds, multipath reflection, and high noise levels in millimeter-wave radar, this invention proposes a complete 3D reconstruction and volume calculation method based on raw radar logs. This method obtains stable point clouds of carriages and materials through point cloud decoding, filtering, and dynamic alignment; it achieves automatic segmentation by identifying carriage connections; and it constructs a TIN surface model using Delaunay triangulation, eliminating abnormal triangles through height difference and side length constraints, resulting in a more continuous and realistic reconstructed surface. Finally, this invention calculates material volume using triangular prism integration based on the projected area and average height of the triangular mesh. This method can operate stably in loading environments with high dust levels, strong obstruction, and complex lighting, providing a reliable technical means for real-time monitoring of bulk material loading on trains. The embodiments of this invention will be described in detail below with reference to the accompanying drawings.

[0022] Figure 1 This diagram illustrates an application scenario of the train material volume detection method based on millimeter-wave radar provided in an embodiment of the present invention. (Example:) Figure 1 As shown, the first step is to install a 4D millimeter-wave radar at an appropriate location in the railway loading tower. The radar should be mounted on a steel structure above or to the side of the loading tower, ensuring its main antenna beam covers the area of ​​multiple train carriages above the railway tracks. The radar should be installed at a safe distance above the top of the carriages, while avoiding obstruction by unloading chutes, hoppers, steel beams, or other structural components. The radar's elevation and azimuth angles should be pre-adjusted according to the track position, ensuring the radar's field of view is aligned with the track centerline and the center of the loading area.

[0023] See Figure 2The document illustrates a flowchart of the implementation of a train material volume detection method based on millimeter-wave radar provided in an embodiment of the present invention, detailed below: Step 201: Obtain point cloud data obtained by scanning the train with millimeter-wave radar, and perform preprocessing to obtain the main body point cloud of the train; wherein, the point cloud data includes the position coordinates of the points.

[0024] In this embodiment, after the radar is installed and calibrated, it operates according to a preset scanning cycle, continuously sending raw log data containing multi-target information to the host computer via a communication protocol within a fixed scanning cycle. The encoding format of the log data is defined by the radar manufacturer, and it is stored frame by frame in hexadecimal or octal form. Each frame corresponds to one radar scan result and contains parameters such as the three-dimensional coordinates of multiple target points, velocities in each of the three dimensions, and reflection intensity. The host computer receives and stores this log data in chronological order, providing the raw data source for subsequent point cloud analysis and processing.

[0025] The raw logs output by the radar are parsed and converted into three-dimensional point cloud data in the world coordinate system of the loading terminal. Specifically, the host computer program identifies the frame header marker, the number of targets, and the encoded fields of each target from each log frame according to the radar communication protocol. Based on the radar's built-in range settings and scaling factors, it calculates the three-dimensional coordinate data, velocity in each direction, and reflection intensity value of each target.

[0026] like Figure 3 As shown, after obtaining the three-dimensional coordinate information of each target, based on the radar's installation position and attitude within the loading tower structure, this invention maps the radar coordinate system to the loading tower's world coordinate system through a three-dimensional rotation matrix transformation, thus obtaining... Figure 4 The original point cloud data shown.

[0027]

[0028] Where (X1, Y1, Z1) are the point cloud coordinates in the original radar coordinate system. Let (X2, Y2, Z2) be the angle between the millimeter-wave radar and the vertical direction, and (X2, Y2, Z2) be the point cloud coordinates in the world coordinate system.

[0029] In the world coordinate system, the X-axis points vertically downwards, representing the altitude to the radar; the Y-axis runs along the railway track, representing the direction of train movement; and the Z-axis extends laterally along the sleepers, representing the width of the train body.

[0030] To facilitate subsequent processing, this invention also adds auxiliary information to each point, including the point's reflection intensity, velocity component, and the index of the time frame to which it belongs, and stores it in a structured form in memory or a local file, so that each point cloud data record contains at least its three-dimensional coordinates, velocity, and reflection intensity features in the world coordinate system, thereby forming a set of original point clouds with physical meaning.

[0031] Step 202: Identify the gaps between carriages based on the density variation characteristics of the train's main point cloud in the direction of train travel.

[0032] In this embodiment, the gap between two carriages refers to the physical connection point between them (such as the area where the coupler is located). Its spatial characteristics are that the point cloud density is significantly lower in the direction of train travel than in the carriage body area. The principle is that the carriage body structure (side walls, end walls) reflects radar waves strongly and continuously, resulting in a dense point cloud; while the connection point is mostly a narrow gap or a complex metal component, with a small reflective area and severe scattering, leading to a sharp reduction in the number of effective points detected by the radar at this location.

[0033] To achieve identification, this step, based on the train body point cloud obtained in step 201 and having calibrated to determine the train's forward direction (usually defined as the Y-axis of the coordinate system), identifies the density variation characteristics of the point cloud, forming a density distribution curve along the Y-axis. It then searches for local minima within this curve and further calculates the ratio of the point cloud density in the "valley" region near this minima to the point cloud density in the adjacent "shoulder" regions (representing the carriage body). When this ratio is below a preset threshold, the location is determined to be a carriage connection gap. This allows for the identification of one or more carriage gaps. The overall train frame obtained after identifying the carriage gaps is as follows: Figure 5 As shown, Compared to algorithms that rely on complex 3D shape matching, density analysis has low computational cost, high real-time performance, and lower requirements for the overall integrity of the point cloud. Even if some points are missing on the surface of the carriage, as long as the low-density features at the connection point exist, they can be reliably detected, demonstrating strong robustness and practicality.

[0034] Step 203: Based on the gap between carriages and the preset carriage length, divide the carriage section point cloud into the main body point cloud.

[0035] In this embodiment, after identifying the connection point, the Y-direction is segmented based on the preset standard car length, dividing the train into several car sections. When a connection gap is identified in the middle, the Y-range of the front and rear cars can be drawn on both sides according to the standard car length. If there are still more cars, the line continues to extend forward or backward at equal intervals to determine the section boundaries of each car in turn.

[0036] After completing the division of the carriage sections, this invention groups and stores the three-dimensional point cloud according to each carriage, laying the foundation for the next step of analyzing the surface of the materials inside each carriage and calculating the volume.

[0037] Step 204: Based on the coordinates of the midpoint of the point cloud of the carriage section in the direction of train height, extract the point cloud of the material surface, and calculate the material volume based on the point cloud of the material surface.

[0038] In this embodiment, a single carriage can be extracted and its internal point cloud can be processed. In the case of multiple carriages, the point cloud can be extracted sequentially and step 204 can be executed to calculate the material volume in all carriages.

[0039] Material surface point cloud specifically refers to the reflection points on the surface of the material (such as coal and ore) loaded inside the car. Since the surface of the material accumulation is uneven and lower than the side walls of the car, it is difficult to separate it directly from the point cloud of the car section. This embodiment utilizes the continuous distribution characteristics formed by the material surface in the height direction (Z-axis) to extract interference points belonging to the car structure (side walls at higher levels and floor plates at lower levels) through statistical analysis.

[0040] After extracting the clean point cloud of the material surface, volume calculation is performed. The core implementation method is 3D reconstruction and spatial integration. First, the surface point cloud is triangulated to construct a continuous surface model consisting of triangular meshes covering the entire material surface. Then, the planar equation of the carriage floor is determined (its height can be obtained from the carriage outline recognition results or design drawings). Finally, the volume of the 3D space enclosed by the triangular mesh surface and the floor plane is calculated. For each prism under the triangular mesh, its volume can be approximated by multiplying the base area by the average height difference; summing the volumes of all prisms yields the total loading volume.

[0041] This method does not rely on regular assumptions about the shape of material piles and can truly reflect their complex three-dimensional morphology. Through rigorous geometric and mathematical methods, discrete point clouds are transformed into continuous volume quantities, and the calculation results are accurate and reliable, providing an objective measurement basis for realizing intelligent loading and trade settlement.

[0042] This invention employs millimeter-wave radar as the core data acquisition method, effectively leveraging its advantages of stable operation under harsh conditions such as rain, snow, fog, dust, and strong sunlight, as well as its wide coverage and strong environmental adaptability. By decoding, filtering, and dynamically aligning point clouds, stable point clouds of carriages and materials are obtained, solving problems such as sparse point clouds, multipath reflection, and high noise levels associated with millimeter-wave radar. Combining the train's morphological characteristics, the density variation features of the point cloud are used to identify carriage connections and achieve automatic segmentation, providing a solid data foundation for material volume calculation and ensuring its accuracy. This invention can operate stably in loading building environments with high dust levels, strong obstructions, and complex lighting, providing a reliable technical means for real-time monitoring of bulk material loading on trains.

[0043] In one possible implementation, the point cloud data also includes the velocity of the points; preprocessing includes displacement compensation; displacement compensation of the point cloud data includes: The point cloud data is divided into multiple point cloud groups, each point cloud group is numbered according to time sequence, and the velocity components of all points in each point cloud group in the direction of train movement are counted; wherein, a point cloud group is defined as N consecutive frames of point cloud data, and N is a preset positive integer. Points with velocity components greater than 0 in each point cloud group are identified as moving points, and the proportion of moving points in each point cloud group is calculated. If the proportion of moving points in any point cloud group is greater than or equal to a preset threshold, then based on the number of the point cloud group and the preset compensation step size, all points in the point cloud group are translated in the opposite direction of the train's forward movement to obtain the displacement-compensated point cloud data.

[0044] In this embodiment, since the train is usually not completely stationary when passing under the loading tower, but moves continuously or intermittently at a low speed, the point clouds collected by the radar at different time frames correspond to the cross-sections of the carriage at different positions. If the point clouds of all time frames are directly superimposed without processing, the carriage outline will be elongated in the Y direction or "ghosting" will occur, which is not conducive to subsequent carriage recognition and geometric modeling.

[0045] To address this issue, this invention groups and dynamically aligns point clouds from different time periods. Specifically, the original logs are structured according to data frames, with N frames forming a point cloud group, each group representing the radar's scan results within a certain time period. For each point cloud group, the velocity components of all points in the train's forward direction are statistically analyzed, and the proportion of points with significantly non-zero velocities is calculated. When this proportion reaches or exceeds a preset threshold, it is determined that the train has significant motion relative to the radar during that time period; when the proportion is low, the train is considered approximately stationary.

[0046] For point cloud groups determined to be in motion, this invention assigns them group numbers according to time sequence and estimates the displacement H of the carriage along the Y direction per unit time based on the average train speed. Then, the Y coordinates of all points within the group are translated in the reverse direction according to the rule of "group number × compensation step size," so that the later-collected point cloud is "pulled back" to the reference position of the previous point cloud. After such displacement compensation, the relative misalignment of multiple frames of point clouds on the Y-axis is significantly reduced, and the positions of fixed structures such as the bottom and side walls of the carriage tend to stabilize in three-dimensional space, thus forming an approximately static carriage point cloud model. The specific value of the compensation step size H is determined based on the actual calibration on site and the speed at which the ox pulls the train.

[0047] In one possible implementation, preprocessing includes multi-stage filtering; multi-stage filtering of the point cloud data includes: Remove points whose position coordinates are outside the preset working range from the point cloud data to obtain point cloud data after physical boundary filtering; The neighborhood distance of each point in the point cloud data after physical boundary filtering is statistically analyzed, and points whose difference between their neighborhood distance and the average neighborhood distance exceeds a preset distance threshold are removed to obtain the statistically filtered point cloud data. Based on the DBSCAN filtering algorithm, the statistically filtered point cloud data is clustered into multiple clusters, and clusters with fewer than a preset number threshold and / or a vertical length less than a preset length threshold are removed to obtain multi-level filtered point cloud data.

[0048] In this embodiment, although the dynamically aligned point cloud is closer to the actual shape of the carriage as a whole, it still contains a large number of interference points caused by ground reflection, upper steel structure, multipath reflection and other environmental objects. To address this, multi-level filtering can be applied to the point cloud based on displacement compensation to significantly improve the signal-to-noise ratio and structural clarity of the point cloud.

[0049] First, a pass-through filter based on physical boundaries is performed. Based on the loading tower design drawings and on-site measurements, a three-dimensional spatial "working zone" covering all the wagons to be inspected is determined. This zone extends vertically from a safe distance above the sleepers to a certain range below the lower structure of the loading tower; horizontally, it extends from several meters in front of the loading tower to several meters behind it; and horizontally, it covers a certain distance on both sides of the track centerline. All points outside this working zone are discarded one by one, thus quickly eliminating point clouds reflecting distant buildings, the ground, bridges, and other irrelevant information.

[0050] Secondly, statistical filtering is performed on the point cloud within the working range. For each point, its nearest neighbors in three-dimensional space are found, and the average distance from the point to its neighbors is calculated. If the average distance between a point and its neighbors is significantly greater than the average neighborhood distance of the entire point cloud, the point can be considered to have a weak connection with the surrounding structure and is an isolated noise point. This invention sets an appropriate distance threshold to delete these point clouds identified as outliers, thereby reducing local noise caused by multipath reflections, accidental clutter, etc.

[0051] Furthermore, after statistical filtering, this invention introduces the DBSCAN filtering algorithm, a density-based clustering algorithm, to perform clustering analysis on the point cloud. In three-dimensional space, for each point, the number of its neighboring points is searched with a certain radius. When the number of neighboring points exceeds a set threshold, the point is considered a "density core point." The mutually adjacent core points and their surrounding neighborhood points together form a density cluster. By traversing all points, multiple clusters composed of spatially connected and high-density points are obtained. This invention filters clusters based on the number of points in the cluster and the cluster's extension length in the Y direction. Clusters with too few points or too short a longitudinal length are often environmental noise or isolated reflections and can be completely removed; clusters with a large number of points and a longitudinal extension close to the length of the carriage are retained as valid point clouds related to the carriage and materials. Through the above triple filtering, a set of high-quality point cloud data, mainly concentrated inside and around the carriage, is finally obtained.

[0052] In one possible implementation, preprocessing includes width center calibration; performing width center calibration on the point cloud data includes: Select a section of a first preset length in the direction of train travel; For all points within the interval, extract the coordinate values ​​in the train width direction to form a width sample set; Remove extreme values ​​from the width sample set and calculate the median or weighted average of the remaining data as the center of the train width; Subtract the center value of the train width from the coordinates of all points in the point cloud data along the train width direction to obtain the point cloud data after width center calibration.

[0053] In this embodiment, based on multi-level filtering, the point cloud distribution along the width of the vehicle body can be calibrated to obtain the center position of the carriage in the lateral direction, providing a basis for subsequent carriage recognition. Specifically, a long interval covering multiple carriages is selected in the Y direction, and the Z-coordinate values ​​are extracted from all point clouds within this interval to form a sample set in the width direction. To reduce the impact of a few outliers, these Z-coordinates can be sorted, and a small number of extreme values ​​at the top and bottom can be removed. Then, the median or weighted average of the remaining data is calculated as the width center position of the entire point cloud.

[0054] Using this width center position as the Z-reference plane, the Z-coordinates of all points are subtracted from this center value, thereby translating the entire carriage in the coordinate system to a position with Z as the axis of symmetry. This operation can offset the effects of factors such as radar lateral offset, slight track curvature, and slight deviation of the carriage from the track center, resulting in a better symmetry in the distribution of the carriage point cloud in the width direction.

[0055] At this stage, the invention also incorporates the projection distribution of point clouds in the Y and Z directions to preliminarily identify the approximate outline of the carriages. Multiple regions with approximately rectangular outlines can be observed in the point cloud distribution on the YZ plane; these regions correspond to the main structure of the carriages. Between these rectangular regions, there are narrow, elongated regions with smaller widths and lower density, corresponding to the connection gaps between carriages. This observation provides an intuitive basis for the precise positioning of the carriage connections in the next step.

[0056] In one possible implementation, the gaps between carriages are identified based on the density variation characteristics of the train's main point cloud in the direction of train travel, including: Along the direction of train travel, the point cloud of the main body of the train is divided into multiple intervals of a second preset length, and the number of point clouds in each interval is counted. The number of point clouds in each interval is converted into point cloud density, and then arranged according to the spatial order of the intervals to obtain a density sequence; For each minimum value in the density sequence, if the slope of the minimum value is less than a preset slope threshold, then the interval where the minimum value is located is determined to be a gap between carriages.

[0057] In this embodiment, the location of the connection between carriages is automatically identified by utilizing the density variation characteristics of the point cloud in the forward direction, and the sections of each carriage are divided accordingly. Specifically, the point cloud, after alignment and filtering, is divided into multiple equal-length intervals along the Y direction, with each interval's length L being 0.2 meters. The interval length L is selected based on the distribution density of the radar point cloud in the forward direction and the typical dimensions of the carriage connection structure to ensure the stability of the point cloud statistics and enhance the identifiability of the point cloud density variation characteristics at the carriage connections.

[0058] After dividing the train into intervals, the number of point clouds within each interval is counted, and the number of point clouds is normalized by combining the interval length. The ratio of the number of point clouds in each interval to the interval length is taken as the point cloud density within that interval, thus obtaining a one-dimensional point cloud density sequence that varies along the train's direction of travel. This sequence is then used for subsequent identification of carriage connection positions. Subsequently, a moving average method is used to smooth the density sequence, reducing high-frequency jitter caused by the randomness of the point clouds, and generating a density curve corresponding to the density sequence to calculate the slope of the minimum point.

[0059] On the smoothed density curve, the point cloud density is higher inside the carriage and near the side walls, while the point cloud density is significantly lower at the carriage connection points due to the gap between the two carriages, resulting in less fixed structure and material accumulation. This invention identifies local minima as candidate points for the connection points. To avoid false minima caused by random noise, this invention calculates the average density of the interval containing each candidate point and several neighboring intervals to its left and right, and constructs an evaluation index: the ratio of the valley interval density (i.e., the point cloud density of the minimum point) to the density of the surrounding annular intervals. Only when this ratio is lower than a set threshold is the candidate minimum value considered to correspond to a real carriage connection gap.

[0060] For a train with a long carriage length, the density curve may contain multiple valley points (minimum points) that meet the requirements. This invention selects one or more of these candidate points as the effective connection location based on factors such as the estimated number of carriages, valley location, valley depth, and spacing. Specifically, all candidate valley points are sorted according to their position in the train's direction of travel, and a corresponding number of valley points are selected based on the preset number of carriages. When the number of candidate valley points exceeds the required number, valley points with greater depth and adjacent valley spacing that meet the minimum interval requirement are preferentially selected as the final carriage connection gap.

[0061] After completing the identification of the gap between the carriages, there will still be some errors between the actual situation and the identified gap. Therefore, position compensation is added based on the actual situation on site. Position compensation is achieved by correcting the position of the identified gap along the direction of train movement. The amount of compensation is determined based on the forward displacement of the train during the point cloud sampling time. After compensation, it is more consistent with the actual situation.

[0062] In one possible implementation, the material surface point cloud is extracted based on the coordinates of the midpoint of the point cloud in the carriage section along the train's height, including: In the point cloud of the carriage section, select the height statistics window of the third preset length; For all points within the height statistics window, extract the coordinate values ​​in the height direction of the train to form a height sample set; Remove outliers from the high-value sample set and count the maximum and minimum values ​​of the remaining data; Add the preset allowance to the maximum value and subtract the preset allowance from the minimum value to form the effective height range; By retaining the points in the point cloud of the carriage section whose coordinate values ​​in the train's height direction are within the effective height range, the point cloud of the material surface is obtained.

[0063] In this embodiment, a single train carriage is extracted and its internal point cloud is processed. Considering that the carriage directly below the radar is not fully complete, and that the carriage directly below it in front experiences less interference from the train structure and millimeter-wave reflection interference, the carriage directly below the radar in front is selected for subsequent processing. Figure 6 As shown, the extracted 3D point cloud information for this carriage is complete. As time goes by, more interfering point clouds will appear, so timely processing and calculation will be more accurate.

[0064] For this carriage, the forward section in the Y-direction of the train's forward direction is selected as the height statistics window. Within this Y-section, a window of length M (e.g., 3-5 meters) is selected towards the front. The height X-coordinates of all point clouds within this window are extracted to form a height sample set. The window length can be specifically set to 40% of the length of a single carriage in the forward direction. After removing a very small number of abnormally high or low points, the minimum and maximum heights of the statistical samples are determined, and then extended upwards and downwards by a margin to form the effective height range of the material surface. For example, an additional margin is added above the maximum height to accommodate local peaks, and a slight extension is made below the minimum height to compensate for some depressions or areas with missing measurements.

[0065] Subsequently, all points within section Y of the car are compared with the aforementioned height range. Points with heights below the lower limit or above the upper limit are considered non-material points, such as upper steel beams, suspended equipment, and reflection points scattered on the lower part of the car, and are deleted. Only points with heights within the effective range are retained as material surface point clouds. After this adaptive height filtering step, each car will obtain a relatively clean set of material surface point clouds, in which the spatial positions of the points basically cover the entire material stacking surface inside the car.

[0066] In one possible implementation, material volume calculation is performed based on the point cloud of the material surface, including: The projection of the point cloud on the material surface onto the YZ plane is used as a set of nodes, and a mesh structure covering the entire interior area of ​​the carriage is generated using the two-dimensional Delaunay triangulation method, resulting in multiple triangular elements; where the YZ plane is the plane formed by the train's forward direction and the train's width direction. For each triangular unit, if the vertex span of the triangular unit is greater than a preset span threshold, the triangular unit is marked as an abnormal unit and removed. For each remaining triangle, if the height difference of the vertices of the normal unit is negative, then it is cut off with the plane of the carriage floor as the boundary, and the remaining part after the cut is divided into triangular units. Calculate the product of the projected area and average height of each triangular element on the Y–Z plane to obtain the material volume corresponding to that triangular element; Calculate the sum of the material volumes corresponding to all triangular units to obtain the material volume corresponding to the point cloud of the carriage section.

[0067] In this embodiment, a three-dimensional curved surface model of the material surface is constructed based on the point cloud of the material surface through triangulation. The volume between this curved surface and the plane of the carriage floor is calculated by integration to obtain the loading volume of the material inside the carriage. The plane of the carriage floor is a preset parameter, specifically determined based on the installation height of the millimeter-wave radar and the distance between the bottom of the carriage and the track sleepers. By converting the radar installation height with the aforementioned distance relationship, the height position of the carriage floor in the world coordinate system is obtained.

[0068] First, using the extracted point cloud of this carriage as a reference, an irregular triangulation mesh is constructed based on its spatial distribution in the Y-Z plane to establish a continuous geometric model of the material surface inside the carriage. Specifically, the projection of the material point cloud onto the Y-Z plane is used as a node set, and a 2D Delaunay triangulation method is employed to generate a mesh structure covering the entire interior area of ​​the carriage. The resulting mesh is shown in the image. Figure 7 As shown, this method can improve the quality and stability of surface modeling by avoiding excessively elongated or twisted triangular elements while ensuring node connectivity.

[0069] Then, for each triangular element, the maximum and minimum heights of its three vertices are calculated. When the height span of a triangular element exceeds a certain threshold, it is considered that the triangular element may be affected by local anomalies, and is temporarily marked as an anomalous element that does not participate in the volume calculation, in order to reduce the impact of local anomalies on the overall result. For the remaining normal triangular elements, their projected area on the Y-Z plane and the difference between the heights of their three vertices and the height of the floor plane are calculated respectively. If all three height differences are positive, the triangular element is completely located on the floor plane, and its volume contribution can be directly used as the product of its average height and projected area.

[0070] If the height difference between one or two vertices is negative, it indicates that the triangular unit crosses the floor of the carriage, requiring it to be cut according to the plane of the floor. Using the plane of the carriage floor as the boundary, find the intersection points of the triangle's sides with the plane. Cut the original triangle into one or two new triangles or quadrilaterals above the plane. Then, divide the quadrilateral into two triangles and calculate their volumes separately. This ensures that only the space actually occupied by the material above the carriage floor is counted.

[0071] By summing the values ​​of all triangular units within a single car, the estimated material volume for that car is obtained. Multiplying this value by the average bulk density of the material allows for the calculation of the material's mass within the car. Combined with measurements from a track scale, the bulk density can also be calibrated and analyzed. For multiple cars, the above calculations can be performed separately to obtain the loading volume of each car. This data is then used by the loading control system to determine whether the target loading volume has been reached, whether the loading speed needs adjustment, or whether loading should be stopped early.

[0072] Based on the above technical solutions, the present invention can be further extended to applications such as multi-radar collaborative measurement, multi-carriage continuous monitoring, and joint calibration of volume and track scale mass, thereby improving the robustness and accuracy of volume estimation.

[0073] In one specific embodiment, the method of the present invention specifically includes the following steps: Step 1: Analyze the raw log data output by the 4D millimeter-wave radar, extracting information such as distance, velocity, azimuth, and reflection intensity for each point. Considering the radar's installation location and attitude, transform the point cloud from the radar's own coordinate system to the loading tower's world coordinate system, which is referenced to the track. This ensures that the height direction, train direction of travel, and vehicle width direction have clear meanings in the world coordinate system. In this embodiment... .

[0074] Step 2: Group the continuous point cloud data according to empty lines, so that... The points are grouped into sets, each representing a radar scan within 1 second. By statistically analyzing the velocity distribution of points within each set, it is determined whether the train is in motion. For point cloud sets in motion, displacement compensation is applied in the forward direction based on their order in the time series. This allows for the dynamic alignment of multiple point clouds in three-dimensional space, resulting in an approximately static carriage point cloud model.

[0075] Step 3: Perform multi-level filtering on the aligned point cloud. First, set the physical range based on the height of the loading tower structure, track position, and effective detection area. Points whose height, distance, and width exceed the reasonable range are filtered out using a pass-through filter. Second, calculate the nearest neighbor distance for each point using a statistical filtering method and delete outliers that are far from local dense areas. Finally, use the DBSCAN clustering algorithm to divide the point cloud into clusters in three-dimensional space. Clusters with a sufficient number of points and sufficient vertical extension are retained as point clouds related to the carriage and materials, while short, scattered, or isolated clusters are identified as noise clusters and removed.

[0076] Step 4: Perform width-direction center calibration on the filtered point cloud. By statistically analyzing the distribution of the Z coordinates of the point cloud within a certain Y range, removing some extreme values, and calculating the median as the overall width centerline of the carriage, the Z coordinates of all points are translated to align the carriage point cloud with a unified reference center in the width direction, so as to facilitate subsequent analysis of carriage width and material distribution.

[0077] Step 5: In the carriage recognition stage, the distribution of point cloud along the train's direction of travel is statistically analyzed, and the direction of travel is divided into several equally spaced intervals. The system calculates the number of points in each segment to obtain a point cloud density profile curve along the Y direction. By smoothing the density curve and performing a minimum value search, it identifies valleys where the point cloud density decreases significantly within the candidate region. The location of the connection gap between carriages is automatically identified by determining the density ratio between the valley interval and its two adjacent intervals.

[0078] Step Six: Based on the identified connection locations and the pre-set standard carriage length and connection gap length, the forward direction is segmented to form the Y-direction range of multiple carriages. Width distribution analysis is performed on the point cloud inside each carriage to estimate the effective width range and width centerline of the carriage, thereby constructing a 3D bounding box that fits the actual carriage dimensions, providing spatial constraints for subsequent volume calculations.

[0079] Step 7: In the material surface extraction stage, this invention selects the tail end of each carriage. As a height statistics window, the minimum and maximum heights of the point cloud within this window are calculated, and upper and lower floating ranges are added based on practical experience. This height range is then used as the effective height range for the material surface. Subsequently, this height range is applied to the point cloud of the entire carriage for pass-through filtering, removing points that are higher or lower than this range, and retaining only the point cloud that is determined to be the material surface or close to the material surface.

[0080] Step 8: In the volume calculation stage, using the interior space of the carriage as the calculation area, an irregular triangular mesh is constructed based on the projection planes of the material surface point cloud in the width and forward directions. For each triangular element, the volume between the material surface and the carriage floor plane is calculated using the height information of its vertices. When all triangular elements are located above the floor, their volume contribution is calculated by multiplying the average height by the area of ​​the triangular pieces. When a triangular element is partially located below the floor, it is geometrically trimmed according to the intersection line with the floor plane, retaining only the portion above the floor for volume calculation. By summing the volumes of all triangular elements, the total volume of the bulk material inside the carriage is obtained.

[0081] The method for 3D modeling and volume calculation based on point cloud data processing proposed in this invention has been verified. It can be seen that the Delaunay triangulation and TIN modeling method constructed in this invention can stably obtain continuous and regular material surface structures in complex environments. Combined with the point cloud preprocessing, projection transformation and triangular prism volume summation model proposed in this invention, the true 3D shape of the cargo body can be accurately restored. Figure 7 The material three-dimensional model generated by the method of this invention. Figure 8 The images were collected from actual sites, and the two images showed a high degree of consistency in terms of surface morphology, geometric features, and overall contour, indicating that the modeling algorithm of this invention has high reconstruction accuracy and stability and is suitable for engineering site applications.

[0082] Based on this, this embodiment further verified the volume and weight calculations of the iron ore loading data. According to the 3D modeling results, the loading volume of the iron ore is approximately 28.710 m³; based on the average density of the crushed and piled iron ore of 2.4 t / m³, its weight is estimated to be approximately 68.904 t. The rated load capacity of the wagon is 70 t, and the error is within 5%. Therefore, the material weight measured in this embodiment is close to the rated range. If used for dynamic loading monitoring in actual transportation processes, it can effectively indicate potential overloading risks. The method of this invention can not only accurately assess the actual loading volume of bulk materials but also calculate the weight by combining density parameters, providing reliable technical support for the safe operation and intelligent management of transportation equipment.

[0083] In summary, the three-dimensional point cloud modeling and volume calculation method proposed in this invention demonstrates good modeling accuracy, computational stability, and environmental adaptability in actual testing. It can meet the engineering requirements for real-time loading monitoring in bulk material transportation scenarios and has significant practical application value and promotion prospects.

[0084] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0085] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0086] Figure 9 A schematic diagram of a train material volume detection device based on millimeter-wave radar provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 9 As shown, the train material volume detection device 9 based on millimeter-wave radar includes: The acquisition module 91 is used to acquire point cloud data obtained by scanning the train with millimeter-wave radar and to preprocess it to obtain the main body point cloud of the train; wherein, the point cloud data includes the position coordinates of the points; The gap recognition module 92 is used to identify the gaps between carriages based on the density change characteristics of the train body point cloud in the direction of train travel. The carriage segmentation module 93 is used to divide the carriage section point cloud in the main body point cloud based on the carriage connection gap and the preset carriage length. The volume calculation module 94 is used to extract the material surface point cloud based on the coordinates of the midpoint of the point cloud of the carriage section in the direction of train height, and to calculate the material volume based on the material surface point cloud.

[0087] In one possible implementation, the point cloud data also includes the velocity of the points; preprocessing includes displacement compensation; the acquisition module 91 is specifically used for: The point cloud data is divided into multiple point cloud groups, each point cloud group is numbered according to time sequence, and the velocity components of all points in each point cloud group in the direction of train movement are counted; wherein, a point cloud group is defined as N consecutive frames of point cloud data, and N is a preset positive integer. Points with velocity components greater than 0 in each point cloud group are identified as moving points, and the proportion of moving points in each point cloud group is calculated. If the proportion of moving points in any point cloud group is greater than or equal to a preset threshold, then based on the number of the point cloud group and the preset compensation step size, all points in the point cloud group are translated in the opposite direction of the train's forward movement to obtain the displacement-compensated point cloud data.

[0088] In one possible implementation, preprocessing includes multi-stage filtering; the acquisition module 91 is specifically used for: Remove points whose position coordinates are outside the preset working range from the point cloud data to obtain point cloud data after physical boundary filtering; The neighborhood distance of each point in the point cloud data after physical boundary filtering is statistically analyzed, and points whose difference between their neighborhood distance and the average neighborhood distance exceeds a preset distance threshold are removed to obtain the statistically filtered point cloud data. Based on the DBSCAN filtering algorithm, the statistically filtered point cloud data is clustered into multiple clusters, and clusters with fewer than a preset number threshold and / or a vertical length less than a preset length threshold are removed to obtain multi-level filtered point cloud data.

[0089] In one possible implementation, preprocessing includes width center calibration; the acquisition module 91 is specifically used for: Select a section of a first preset length in the direction of train travel; For all points within the interval, extract the coordinate values ​​in the train width direction to form a width sample set; Remove extreme values ​​from the width sample set and calculate the median or weighted average of the remaining data as the center of the train width; Subtract the center value of the train width from the coordinates of all points in the point cloud data along the train width direction to obtain the point cloud data after width center calibration.

[0090] In one possible implementation, the gap recognition module 92 is specifically used for: Along the direction of train travel, the point cloud of the main body of the train is divided into multiple intervals of a second preset length, and the number of point clouds in each interval is counted. The number of point clouds in each interval is converted into point cloud density, and then arranged according to the spatial order of the intervals to obtain a density sequence; For each minimum value in the density sequence, if the slope of the minimum value is less than a preset slope threshold, then the interval where the minimum value is located is determined to be a gap between carriages.

[0091] In one possible implementation, the volume calculation module 94 is specifically used for: In the point cloud of the carriage section, select the height statistics window of the third preset length; For all points within the height statistics window, extract the coordinate values ​​in the height direction of the train to form a height sample set; Remove outliers from the high-value sample set and count the maximum and minimum values ​​of the remaining data; Add the preset allowance to the maximum value and subtract the preset allowance from the minimum value to form the effective height range; By retaining the points in the point cloud of the carriage section whose coordinate values ​​in the train's height direction are within the effective height range, the point cloud of the material surface is obtained.

[0092] In one possible implementation, the volume calculation module 94 is specifically used for: The projection of the point cloud on the material surface onto the YZ plane is used as a set of nodes, and a mesh structure covering the entire interior area of ​​the carriage is generated using the two-dimensional Delaunay triangulation method, resulting in multiple triangular elements; where the YZ plane is the plane formed by the train's forward direction and the train's width direction. For each triangular unit, if the vertex span of the triangular unit is greater than a preset span threshold, the triangular unit is marked as an abnormal unit and removed. For each remaining triangle, if the height difference of the vertices of the normal unit is negative, then it is cut off with the plane of the carriage floor as the boundary, and the remaining part after the cut is divided into triangular units. Calculate the product of the projected area and average height of each triangular element on the Y–Z plane to obtain the material volume corresponding to that triangular element; Calculate the sum of the material volumes corresponding to all triangular units to obtain the material volume corresponding to the point cloud of the carriage section.

[0093] This invention employs millimeter-wave radar as the core data acquisition method, effectively leveraging its advantages of stable operation under harsh conditions such as rain, snow, fog, dust, and strong sunlight, as well as its wide coverage and strong environmental adaptability. By decoding, filtering, and dynamically aligning point clouds, stable point clouds of carriages and materials are obtained, solving problems such as sparse point clouds, multipath reflection, and high noise levels associated with millimeter-wave radar. Combining the train's morphological characteristics, the density variation features of the point cloud are used to identify carriage connections and achieve automatic segmentation, providing a solid data foundation for material volume calculation and ensuring its accuracy. This invention can operate stably in loading building environments with high dust levels, strong obstructions, and complex lighting, providing a reliable technical means for real-time monitoring of bulk material loading on trains.

[0094] Figure 10 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 10 As shown, the electronic device 10 of this embodiment includes a processor 100 and a memory 101. The memory 101 stores a computer program 102. When the processor 100 executes the computer program 102, it implements the steps in the various method embodiments described above. Alternatively, when the processor 100 executes the computer program 102, it implements the functions of each module / unit in the various device embodiments described above.

[0095] For example, computer program 102 may be divided into one or more modules / units, which are stored in memory 101 and executed by processor 100 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 102 in electronic device 10.

[0096] Electronic device 10 may include, but is not limited to, processor 100 and memory 101. Those skilled in the art will understand that... Figure 10 This is merely an example of electronic device 10 and does not constitute a limitation on electronic device 10. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 10 may also include input / output devices, network access devices, buses, etc.

[0097] The processor 100 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0098] The memory 101 can be an internal storage unit of the electronic device 10, such as a hard disk or RAM of the electronic device 10. The memory 101 can also be an external storage device of the electronic device 10, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 10. Furthermore, the memory 101 can include both internal and external storage units of the electronic device 10. The memory 101 is used to store the computer program 102 and other programs and data required by the electronic device 10. The memory 101 can also be used to temporarily store data that has been output or will be output.

[0099] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0100] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0101] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0102] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0103] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0104] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting the volume of train materials based on millimeter-wave radar, characterized in that, include: Point cloud data obtained by scanning the train using millimeter-wave radar is acquired and preprocessed to obtain the main body point cloud of the train; wherein, the point cloud data includes the position coordinates of the points; Based on the density variation characteristics of the train body point cloud in the direction of train travel, the gaps between carriages are identified. Based on the gap between the carriages and the preset carriage length, the carriage section point cloud is divided in the main body point cloud of the train. Based on the coordinates of the midpoint of the point cloud of the carriage section in the direction of train height, the point cloud of the material surface is extracted, and the material volume is calculated based on the point cloud of the material surface.

2. The method for detecting train material volume based on millimeter-wave radar according to claim 1, characterized in that, The point cloud data also includes the velocity of the points; the preprocessing includes displacement compensation; displacement compensation of the point cloud data includes: The point cloud data is divided into multiple point cloud groups, each point cloud group is numbered according to time sequence, and the velocity components of all points in each point cloud group in the direction of train movement are counted; wherein, a point cloud group is defined as N consecutive frames of point cloud data, and N is a preset positive integer. Points with velocity components greater than 0 in each point cloud group are identified as moving points, and the proportion of moving points in each point cloud group is calculated. If the proportion of moving points in any point cloud group is greater than or equal to a preset threshold, then based on the number of the point cloud group and the preset compensation step size, all points in the point cloud group are translated in the opposite direction of the train's forward movement to obtain the displacement-compensated point cloud data.

3. The method for detecting train material volume based on millimeter-wave radar according to claim 1, characterized in that, The preprocessing includes multi-level filtering; multi-level filtering of point cloud data includes: Remove points whose position coordinates are outside the preset working range from the point cloud data to obtain point cloud data after physical boundary filtering; The neighborhood distance of each point in the point cloud data after physical boundary filtering is statistically analyzed, and points whose difference between their neighborhood distance and the average neighborhood distance exceeds a preset distance threshold are removed to obtain the statistically filtered point cloud data. Based on the DBSCAN filtering algorithm, the statistically filtered point cloud data is clustered into multiple clusters, and clusters with fewer than a preset number threshold and / or a vertical length less than a preset length threshold are removed to obtain multi-level filtered point cloud data.

4. The method for detecting train material volume based on millimeter-wave radar according to claim 1, characterized in that, The preprocessing includes width center calibration; Width center calibration of point cloud data includes: Select a section of a first preset length in the direction of train travel; For all points within the interval, extract the coordinate values ​​in the train width direction to form a width sample set; Remove extreme values ​​from the width sample set and calculate the median or weighted average of the remaining data as the center of the train width; Subtract the train width center value from the coordinates of all points in the point cloud data in the train width direction to obtain the point cloud data after width center calibration.

5. The method for detecting train material volume based on millimeter-wave radar according to any one of claims 1 to 4, characterized in that, The method of identifying the gaps between carriages based on the density variation characteristics of the train's main point cloud in the direction of train travel includes: Along the direction of train travel, the point cloud of the main body of the train is divided into multiple intervals of a second preset length, and the number of point clouds in each interval is counted. The number of point clouds in each interval is converted into point cloud density, and then arranged according to the spatial order of the intervals to obtain a density sequence; For each minimum value in the density sequence, if the slope of the minimum value is less than a preset slope threshold, then the interval where the minimum value is located is determined to be a gap between carriages.

6. The method for detecting train material volume based on millimeter-wave radar according to any one of claims 1 to 4, characterized in that, Extracting material surface point clouds based on the coordinates of the midpoint of the point cloud of the carriage section in the train height direction includes: In the point cloud of the carriage section, select a height statistics window of a third preset length; For all points within the height statistics window, extract the coordinate values ​​in the height direction of the train to form a height sample set; Remove outliers from the height sample set and count the maximum and minimum values ​​of the remaining data; The effective height range is formed by adding the preset allowance to the maximum value and subtracting the preset allowance from the minimum value. By retaining the points in the point cloud of the carriage section whose coordinate values ​​in the train height direction are within the effective height range, the point cloud of the material surface is obtained.

7. The method for detecting train material volume based on millimeter-wave radar according to any one of claims 1 to 4, characterized in that, The calculation of material volume based on the point cloud of the material surface includes: The projection of the point cloud on the surface of the material onto the YZ plane is used as a set of nodes, and a mesh structure covering the entire interior area of ​​the carriage is generated using the two-dimensional Delaunay triangulation method to obtain multiple triangular elements; wherein, the YZ plane is the plane formed by the train's forward direction and the train's width direction; For each triangular unit, if the vertex span of the triangular unit is greater than a preset span threshold, the triangular unit is marked as an abnormal unit and removed. For each remaining triangle, if the height difference of the vertices of the normal unit is negative, then it is cut off with the plane of the carriage floor as the boundary, and the remaining part after the cut is divided into triangular units. Calculate the product of the projected area and average height of each triangular element on the Y–Z plane to obtain the material volume corresponding to that triangular element; Calculate the sum of the material volumes corresponding to all triangular units to obtain the material volume corresponding to the point cloud of the carriage section.

8. A train material volume detection device based on millimeter-wave radar, characterized in that, include: The acquisition module is used to acquire point cloud data obtained by scanning the train with millimeter-wave radar, and to preprocess the data to obtain the main body point cloud of the train; wherein, the point cloud data includes the position coordinates of the points; The gap recognition module is used to identify the gaps between carriages based on the density change characteristics of the point cloud of the main body of the train in the direction of train travel; The carriage segmentation module is used to divide the carriage segment point cloud into the main body point cloud based on the carriage connection gap and the preset carriage length. The volume calculation module is used to extract the material surface point cloud based on the coordinates of the midpoint of the point cloud of the carriage section in the train height direction, and to calculate the material volume based on the material surface point cloud.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images