Solid waste residue volume monitoring method and device based on unmanned aerial vehicle oblique photography

Abstract

The application provides a solid waste residue volume monitoring method and device based on unmanned aerial vehicle oblique photography, the method comprises the following steps: acquiring multi-period three-dimensional point cloud data of a residue field, calculating a point cloud local information entropy gradient based on a local neighborhood, dissecting a space analysis unit and determining a local structure complexity parameter; determining a regional semantic label in combination with an average normal vector inclination and an elevation distribution feature; constructing a multi-dimensional high-elevation semantic vector in each unit, generating a unit dominant semantic vector and a diagnostic modulus, and removing high-elevation gross error points according to an elevation semantic deviation distance, a positive elevation difference and a planarity improvement condition; and finally, constructing a DEM based on clean point cloud and differentiating to obtain a volume monitoring result.

Description

Technical Field

[0001] This application belongs to the field of monitoring, and in particular relates to a method and device for monitoring the volume of solid waste residue based on oblique photography by unmanned aerial vehicles. Background Technology

[0002] Mining, construction, and industrial production processes generate large amounts of solid waste, which is typically stockpiled in slag heaps, tailings ponds, or spoil heaps. The safety, stability, and environmental compliance of solid waste heaps are crucial aspects of enterprise safety management. Monitoring the volume of slag heaps is essential for assessing their actual carrying capacity, calculating production progress, and providing early warnings of geological hazards such as landslides. Current technologies typically utilize 3D laser scanning and UAV aerial photogrammetry to acquire 3D point cloud data of the monitored area at different times. By constructing a digital elevation model and performing multi-period spatial difference calculations, the volume changes of solid waste are calculated.

[0003] In actual monitoring environments, solid waste dumps are typically located in open-air conditions, making it easy for high-level gross points to be mixed into the acquired 3D point cloud data. These high-level gross points include temporary facilities such as excavators, transport vehicles, and conveyor belts, flying targets, and scattered vegetation canopies. Furthermore, solid waste dumps have complex spatial topographic structures, typically including gentle working platforms, steep dump slopes, and settlement zones, with significant differences in point cloud spatial distribution characteristics and surface undulation at different locations. If the point cloud data is not processed in a targeted manner, taking into account the local topographic complexity and spatial semantic features of different areas, before constructing the elevation model, insufficient removal of high-level gross points or misclassification of actual steep slopes and dump edges as gross points can easily occur. This affects the accuracy of the surface representation in the digital elevation model and the reliability of volume change results during multi-stage difference calculations. Therefore, how to obtain more reliable surface point cloud data by balancing the preservation of real terrain features with the identification and elimination of high-level coarse points under complex slag dump terrain conditions is a technical problem that needs to be solved in the field of solid waste volume monitoring. Summary of the Invention

[0004] To address the problems in existing technologies where complex slag yard point clouds are easily affected by high-level gross errors and real slope features are easily deleted, resulting in insufficient accuracy of digital elevation models and reliability of volume monitoring, this invention provides a method and device for monitoring the volume of solid waste slag based on UAV oblique photography.

[0005] In a first aspect, the present invention provides a method for monitoring the volume of solid waste based on oblique photography by unmanned aerial vehicles (UAVs), comprising the following steps: Acquire three-dimensional point cloud data of the solid waste slag site to be monitored, including at least the current period and the reference period, and perform data processing steps independently on the three-dimensional point cloud data of each period; Based on the local neighborhood of the data points in the three-dimensional point cloud data, the local information entropy gradient of the point cloud is calculated. According to the local information entropy gradient of the point cloud, the three-dimensional point cloud data is divided into multiple spatial analysis units, and the local structural complexity parameter of each spatial analysis unit is calculated. The region semantic segmentation of the solid waste slag site to be monitored is performed, and the region semantic label is determined for each spatial analysis unit. Within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, a multidimensional high-level elevation semantic vector is calculated for each data point. The unit-dominant semantic vector of each spatial analysis unit is determined, and a corresponding diagnostic modulus is generated according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit-dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic modulus, then the data point is determined to be a high-level gross error point and removed. Digital elevation models were constructed and differential calculations were performed on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the monitoring results of the solid waste volume.

[0006] Optionally, the step of dividing the 3D point cloud data into multiple spatial analysis units based on the local information entropy gradient of the point cloud, and calculating the local structural complexity parameter of each spatial analysis unit, includes: Construct an initial 3D bounding box containing the 3D point cloud data as the root node; Calculate the average value of the information entropy gradient of all data points within the current node as the local structural complexity parameter of the current node; Determine whether the local structural complexity parameter of the current node exceeds the preset partitioning threshold. If it does, and the size of the current node is larger than the preset minimum scale, then partition the current node into eight child nodes evenly. Repeat the above calculation and judgment process until the local structural complexity parameters of all child nodes are not greater than the partitioning threshold or the node size reaches the minimum scale, and then use the leaf nodes as the multiple spatial analysis units.

[0007] Optionally, the step of performing regional semantic segmentation on the solid waste residue site to be monitored and determining regional semantic labels for each spatial analysis unit includes: The average angle of inclination of the average normal vector is obtained by calculating the average angle between the normal vector of all data points in the spatial analysis unit and the vertical direction, and the elevation range of the data points in the spatial analysis unit is calculated. If the average normal vector tilt angle is less than the first tilt angle threshold and the elevation range is less than the first elevation threshold, then the regional semantic label of the spatial analysis unit is determined as a flat area at the top of the slope. If the average normal vector tilt angle is greater than the second tilt angle threshold, then the regional semantic label of the spatial analysis unit is determined as a steep slope area; If the spatial analysis unit is not identified as a flat area at the top of the slope or a steep area on the slope, then the regional semantic label of the spatial analysis unit is identified as a transition area at the foot of the slope.

[0008] Optionally, the step of calculating a multidimensional elevation semantic vector for each data point within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, includes: Sort the elevation values ​​of all data points within the spatial analysis unit, take the median elevation, and obtain the reference elevation. Calculate the difference between the elevation of the data point itself and the reference elevation; Calculate the average of the elevation differences obtained by subtracting the elevations of each of the nearest neighbor points from the elevation of the data point itself; The reference elevation, the difference, and the average of the elevation difference are combined sequentially to form a multidimensional elevation semantic vector for the data point.

[0009] Optionally, the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies a second threshold condition based on the diagnostic modulus, including: A local point set is obtained with the data point as the center and a preset radius, and a covariance matrix is ​​constructed. The first eigenvalue, the second eigenvalue, and the third eigenvalue of the covariance matrix are calculated, and the first eigenvalue is not less than the second eigenvalue, and the second eigenvalue is not less than the third eigenvalue. The quotient obtained by dividing the difference between the second eigenvalue and the third eigenvalue by the first eigenvalue is used as the initial planarity measure factor before removing the data point. Remove the data points from the local point set, reconstruct the covariance matrix, and calculate the planarity measure factor after removal. The difference between the removed planarity measure factor and the initial planarity measure factor is calculated to obtain the planarity improvement amount; If the initial flatness measure factor is less than the first flatness threshold and the flatness improvement amount is greater than the flatness improvement threshold set based on the diagnostic module, then the second threshold condition is determined to be satisfied, wherein the first flatness threshold and the flatness improvement threshold are determined based on the diagnostic module.

[0010] Optionally, the step of constructing digital elevation models and performing differential calculations on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the solid waste volume monitoring results includes: The clean point cloud data from each phase are projected onto the same reference horizontal plane and divided into regular grids of the same resolution. Interpolate the point cloud elevation within each regular grid to generate digital elevation models for the current period and the reference period; Calculate the elevation difference between the digital elevation models of the current period and the reference period under the same planar grid coordinates; The product of the elevation difference of all regular grids and the area of ​​a single grid is summed to obtain the monitoring result of the volume of solid waste.

[0011] Secondly, the present invention also provides a solid waste volume monitoring device based on UAV oblique photography, comprising the following modules: The acquisition module is used to acquire three-dimensional point cloud data of each phase of the solid waste slag site to be monitored, including at least the current phase and the reference phase, and to independently perform data processing steps on the three-dimensional point cloud data of each phase. The calculation module is used to calculate the local information entropy gradient of the point cloud based on the local neighborhood of the data points in the three-dimensional point cloud data, divide the three-dimensional point cloud data into multiple spatial analysis units according to the local information entropy gradient of the point cloud, calculate the local structural complexity parameters of each spatial analysis unit, perform regional semantic segmentation on the solid waste slag site to be monitored, and determine regional semantic labels for each spatial analysis unit. The determination module is used to calculate a multidimensional high-level elevation semantic vector for each data point within each spatial analysis unit based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, determine the unit dominant semantic vector of each spatial analysis unit, and generate a corresponding diagnostic module according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic module, then the data point is determined to be a high-level gross error point and removed. The monitoring module is used to construct digital elevation models and perform differential calculations on the clean point cloud data of each period after removing high-level coarse-difference points to obtain the monitoring results of the solid waste volume.

[0012] Further, the step of dividing the 3D point cloud data into multiple spatial analysis units based on the local information entropy gradient of the point cloud, and calculating the local structural complexity parameter of each spatial analysis unit, includes: Construct an initial 3D bounding box containing the 3D point cloud data as the root node; Calculate the average value of the information entropy gradient of all data points within the current node as the local structural complexity parameter of the current node; Determine whether the local structural complexity parameter of the current node exceeds the preset partitioning threshold. If it does, and the size of the current node is larger than the preset minimum scale, then partition the current node into eight child nodes evenly. Repeat the above calculation and judgment process until the local structural complexity parameters of all child nodes are not greater than the partitioning threshold or the node size reaches the minimum scale, and then use the leaf nodes as the multiple spatial analysis units.

[0013] Furthermore, the step of performing regional semantic segmentation on the solid waste slag site to be monitored, and determining regional semantic labels for each spatial analysis unit, includes: The average angle of inclination of the average normal vector is obtained by calculating the average angle between the normal vector of all data points in the spatial analysis unit and the vertical direction, and the elevation range of the data points in the spatial analysis unit is calculated. If the average normal vector tilt angle is less than the first tilt angle threshold and the elevation range is less than the first elevation threshold, then the regional semantic label of the spatial analysis unit is determined as a flat area at the top of the slope. If the average normal vector tilt angle is greater than the second tilt angle threshold, then the regional semantic label of the spatial analysis unit is determined as a steep slope area; If the spatial analysis unit is not identified as a flat area at the top of the slope or a steep area on the slope, then the regional semantic label of the spatial analysis unit is identified as a transition area at the foot of the slope.

[0014] Furthermore, within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, the calculation of a multidimensional elevation semantic vector for each data point includes: Sort the elevation values ​​of all data points within the spatial analysis unit, take the median elevation, and obtain the reference elevation. Calculate the difference between the elevation of the data point itself and the reference elevation; Calculate the average of the elevation differences obtained by subtracting the elevations of each of the nearest neighbor points from the elevation of the data point itself; The reference elevation, the difference, and the average of the elevation difference are combined sequentially to form a multidimensional elevation semantic vector for the data point.

[0015] Furthermore, the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies a second threshold condition based on the diagnostic modulus, including: A local point set is obtained with the data point as the center and a preset radius, and a covariance matrix is ​​constructed. The first eigenvalue, the second eigenvalue, and the third eigenvalue of the covariance matrix are calculated, and the first eigenvalue is not less than the second eigenvalue, and the second eigenvalue is not less than the third eigenvalue. The quotient obtained by dividing the difference between the second eigenvalue and the third eigenvalue by the first eigenvalue is used as the initial planarity measure factor before removing the data point. Remove the data points from the local point set, reconstruct the covariance matrix, and calculate the planarity measure factor after removal. The difference between the removed planarity measure factor and the initial planarity measure factor is calculated to obtain the planarity improvement amount; If the initial flatness measure factor is less than the first flatness threshold and the flatness improvement amount is greater than the flatness improvement threshold set based on the diagnostic module, then the second threshold condition is determined to be satisfied, wherein the first flatness threshold and the flatness improvement threshold are determined based on the diagnostic module.

[0016] Furthermore, the step of constructing digital elevation models and performing differential calculations on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the solid waste volume monitoring results includes: The clean point cloud data from each phase are projected onto the same reference horizontal plane and divided into regular grids of the same resolution. Interpolate the point cloud elevation within each regular grid to generate digital elevation models for the current period and the reference period; Calculate the elevation difference between the digital elevation models of the current period and the reference period under the same planar grid coordinates; The product of the elevation difference of all regular grids and the area of ​​a single grid is summed to obtain the monitoring result of the volume of solid waste.

[0017] This invention adaptively spatially partitions 3D point cloud data by calculating the local information entropy gradient of the point cloud, and determines regional semantic labels by combining local structural complexity, average normal vector tilt angle, and elevation distribution characteristics, thereby characterizing the differences in terrain structure in different areas of the solid waste slag disposal site. By constructing a multidimensional high-level elevation semantic vector and generating a diagnostic modulus by combining structural complexity and semantic labels, and identifying and removing high-level gross errors by combining elevation semantic deviation distance, positive elevation difference, and planarity improvement conditions, the invention further utilizes the processed clean point cloud to construct a digital elevation model and perform differential operations, which helps to reduce gross error interference and improve the stability and reliability of solid waste volume monitoring results. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for monitoring the volume of solid waste based on UAV oblique photography; Figure 2This is a schematic diagram of the scatter distribution of node parameters in the partition space. Detailed Implementation

[0019] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] Example 1

[0021] In Embodiment 1 of the present invention, a method for monitoring the volume of solid waste based on oblique photography by unmanned aerial vehicles (UAVs) is provided, such as... Figure 1 As shown, it includes the following steps: S1, acquire three-dimensional point cloud data of each phase of the solid waste slag site to be monitored, including at least the current phase and the reference phase, and independently perform data processing steps on the three-dimensional point cloud data of each phase.

[0022] Multi-view images of the solid waste disposal site under monitoring at different time points were collected using an UAV oblique photography system. These images were then processed through aerial triangulation, dense matching, and point cloud reconstruction to generate original 3D point clouds. The current period represents the most recently acquired point cloud, while the reference period represents the historical baseline point cloud. The reading interface loads the original 3D point cloud files from each period into computer memory and converts them into point coordinate arrays. Voxelization downsampling is performed on the 3D point cloud data for each period, and spatial points are merged and thinned according to a preset voxel size to reduce point cloud redundancy and unify the overall data density. Subsequently, a statistical outlier removal method is used to search for a specified number of neighboring points for each point cloud data point and calculate the average neighborhood distance. Data points with an average neighborhood distance exceeding the global distance threshold are identified as outliers and removed, thereby eliminating low-altitude random noise and discrete flying points.

[0023] S2, calculate the local information entropy gradient of the point cloud based on the local neighborhood of the data points in the three-dimensional point cloud data, divide the three-dimensional point cloud data into multiple spatial analysis units according to the local information entropy gradient of the point cloud, calculate the local structural complexity parameter of each spatial analysis unit, perform regional semantic segmentation on the solid waste slag site to be monitored, and determine regional semantic labels for each spatial analysis unit.

[0024] The Fast Nearest Neighbor Search algorithm is invoked to construct a spherical local neighborhood containing a preset number of nearest neighbors for each data point in the 3D point cloud data. The 3D coordinate point set within the local neighborhood is extracted, a 3×3 spatial covariance matrix is ​​calculated, and a principal component analysis algorithm is executed to obtain three eigenvalues. The three normalized eigenvalues ​​are substituted into the Shannon information entropy formula to calculate the local spatial information entropy of a single point. The spatial difference rate between the information entropy of the current point and the information entropy of the neighboring points is calculated to obtain the local information entropy gradient of the point cloud. The magnitude of the local information entropy gradient of the point cloud is then normalized.

[0025] An octree structure is constructed, using the average value of the normalized local information entropy gradient magnitude of the point cloud within each node as the local structural complexity parameter. Whether a node continues to split is determined based on whether this local structural complexity parameter exceeds a preset splitting threshold. The 3D space is recursively partitioned until the splitting termination condition is met or the maximum tree depth is reached. Each generated leaf node is treated as an independent spatial analysis unit. The normal estimation module is used to calculate the mean normal vector of all data points within the spatial analysis unit. The angle between the dot product of this mean normal vector and the vertical Z-axis is calculated as the mean normal vector inclination angle. The elevation range of all points within the spatial analysis unit is extracted as the elevation distribution feature. A density-based spatial clustering algorithm is used to perform unsupervised clustering analysis on the 2D feature matrix containing the mean normal vector inclination angle and elevation distribution features. The entire solid waste slag site to be monitored is divided into regions such as flat top areas, steep slope areas, and transitional slope areas. The corresponding clustering category identifier is assigned to each spatial analysis unit as a region semantic label.

[0026] In one embodiment, the step of dividing the 3D point cloud data into multiple spatial analysis units based on the local information entropy gradient of the point cloud, and calculating the local structural complexity parameter of each spatial analysis unit, includes: Construct an initial 3D bounding box containing the 3D point cloud data as the root node; Calculate the average value of the information entropy gradient of all data points within the current node as the local structural complexity parameter of the current node; Determine whether the local structural complexity parameter of the current node exceeds the preset partitioning threshold. If it does, and the size of the current node is larger than the preset minimum scale, then partition the current node into eight child nodes evenly. Repeat the above calculation and judgment process until the local structural complexity parameters of all child nodes are not greater than the partitioning threshold or the node size reaches the minimum scale, and then use the leaf nodes as the multiple spatial analysis units.

[0027] After acquiring 3D point cloud data, based on coordinate extrema... and Construct an initial 3D bounding box, i.e., the root node of an octree. Assume the point cloud data size is 1 million points, spatially distributed over a solid waste disposal site of 500m × 500m × 100m. For each data point within the root node, search for K nearest neighbors, where K is preferably between 15 and 30, for example, K=20. Based on the 3D coordinates of the data point and its nearest neighbors, construct a covariance matrix, extract eigenvalues, and calculate the local information entropy. Calculate the spatial change rate of the information entropy of adjacent points. After normalizing the magnitude of the spatial change rate, calculate the arithmetic mean of the normalized magnitudes of all points within the node, denoted as the local structural complexity parameter. Set a preset partitioning threshold, preferably between 0.4 and 0.8, here 0.6 is used as an example; set a preset minimum scale between 1.0m and 5.0m, for example, 2.0m, to reduce the risk of excessive memory consumption due to over-segmentation, while preserving local micro-topographical features. The scatter distribution of the partitioning spatial node parameters is as follows: Figure 2 As shown.

[0028] In the recursive subdivision of the octree, when the complexity parameter of the current node reaches 0.75, which is greater than 0.6, and the side length of the current bounding box node is 125m, which is greater than the minimum scale of 2.0m, the node is symmetrically cut along the three orthogonal coordinate axes X, Y, and Z at the three-dimensional geometric center point, generating 8 non-overlapping sub-cube nodes of equal volume. Each generated sub-node is traversed, and the calculation and judgment of the covariance and average gradient of information entropy are repeated. When the local structural complexity parameter of a sub-node drops to 0.55, which is less than 0.6, or when the side length of the sub-node reaches 1.95m, which is less than 2.0m after multiple subdivisions, the subdivision of that branch is stopped, and the sub-node is designated as a leaf node. All leaf nodes formed through recursive termination constitute the set of spatial analysis units, achieving a sparse representation of point cloud data. Steep slopes with dramatic terrain changes are subdivided more densely, while flat areas are subdivided more sparsely.

[0029] In one embodiment, the step of performing regional semantic segmentation on the solid waste slag site to be monitored, and determining regional semantic labels for each spatial analysis unit, includes: The average angle of inclination of the average normal vector is obtained by calculating the average angle between the normal vector of all data points in the spatial analysis unit and the vertical direction, and the elevation range of the data points in the spatial analysis unit is calculated. If the average normal vector tilt angle is less than the first tilt angle threshold and the elevation range is less than the first elevation threshold, then the regional semantic label of the spatial analysis unit is determined as a flat area at the top of the slope. If the average normal vector tilt angle is greater than the second tilt angle threshold, then the regional semantic label of the spatial analysis unit is determined as a steep slope area; If the spatial analysis unit is not identified as a flat area at the top of the slope or a steep area on the slope, then the regional semantic label of the spatial analysis unit is identified as a transition area at the foot of the slope.

[0030] For the leaf nodes of the spatial analysis unit, the surface normal vector of each data point within the node is extracted based on principal component analysis. The inverse cosine of the dot product of the normal vector and the vertical upward direction vector (0,0,1) is calculated to obtain the normal vector tilt angle of a single point. The arithmetic mean of the tilt angle values ​​of all data points within the unit is calculated to obtain the average normal vector tilt angle. The maximum and minimum Z-axis coordinates of all data points within the spatial analysis unit are extracted, and the difference is used to obtain the elevation range. The preferred range for the first tilt angle threshold is set to 10° to 20°, for example, 15°; the preferred range for the second tilt angle threshold is set to 35° to 45°, for example, 40°; and the preferred range for the first elevation threshold is set to 0.5m to 2.0m, for example, 1.0m.

[0031] In determining the semantic label of a region, if the average normal vector inclination angle calculated for the current analysis unit is 12° (less than 15°) and the elevation range is 0.6m (less than 1.0m), then the semantic label of the analysis unit is determined as "flat top region"; if the calculated average normal vector inclination angle is 42° (greater than 40°), then the semantic label of the analysis unit is determined as "steep slope region"; when unit parameters such as the average normal vector inclination angle are 25°, or the average normal vector inclination angle is 10° and the elevation range is 1.5m, and these conditions for flat top region and steep slope region are not met, the semantic label of the analysis unit is determined as "slope toe transition region". This segmentation process converts the point cloud into blocks with terrain category identifiers.

[0032] S3. Within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, a multidimensional high-level elevation semantic vector is calculated for each data point. The unit-dominant semantic vector of each spatial analysis unit is determined, and a corresponding diagnostic modulus is generated according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit-dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic modulus, then the data point is determined to be a high-level gross error point and removed.

[0033] By traversing all data points within a single spatial analysis unit, a three-dimensional elevation semantic vector is constructed by combining the average of the elevation difference between the reference elevation of the spatial analysis unit, the elevation of the central data point minus the reference elevation, and the elevation difference between the central data point and each nearest neighbor point.

[0034] The dominant semantic vector of each spatial analysis unit is determined, and a corresponding diagnostic modulus is generated based on the local structural complexity parameter and the regional semantic label. Specifically, for each spatial analysis unit, the multidimensional elevation semantic vector of all data points within the unit is extracted, and the arithmetic mean of each dimension component is calculated. The resulting average vector is used as the dominant semantic vector of the spatial analysis unit. A mapping relationship between regional semantic labels and semantic weight coefficients is pre-established, where the flat area at the top of the slope corresponds to the first semantic weight coefficient, the transition area at the bottom of the slope corresponds to the second semantic weight coefficient, and the steep area on the slope corresponds to the third semantic weight coefficient. The third semantic weight coefficient is greater than the second semantic weight coefficient, and the second semantic weight coefficient is greater than the first semantic weight coefficient. The semantic weight coefficient corresponding to the regional semantic label of the current spatial analysis unit is multiplied by the normalized local structural complexity parameter to obtain the diagnostic modulus corresponding to the spatial analysis unit.

[0035] Multiply the global baseline elevation difference threshold parameter by the adjustment factor corresponding to the diagnostic modulus to solve the Euclidean distance between the single-point 3D elevation semantic vector and the unit dominant semantic vector. When the distance scalar is greater than the first threshold and the self-elevation difference is positive, local geometric verification is performed. Principal component analysis is used to calculate the initial planarity measure factor of the initial local point set containing the central data point. The planarity measure factor of the remaining local point set after removing the central data point is calculated again in the same way. The planarity improvement amount is obtained by subtracting the initial planarity measure factor from the planarity measure factor after removal. A planarity improvement threshold is set according to the diagnostic modulus, and the planarity improvement threshold is negatively correlated with the diagnostic modulus. When the initial planarity measure factor is less than the first planarity threshold and the planarity improvement amount is greater than the planarity improvement threshold, it indicates that the local surface geometric consistency has been improved after removing the central data point. The central data point is identified as an elevation gross point caused by a conveyor belt support or heavy machinery cantilever, etc., and the array slice deletion operation is called to remove the index row data of the point from memory.

[0036] In one embodiment, the step of calculating a multidimensional elevation semantic vector for each data point within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, includes: Sort the elevation values ​​of all data points within the spatial analysis unit, take the median elevation, and obtain the reference elevation. Calculate the difference between the elevation of the data point itself and the reference elevation; Calculate the average of the elevation differences obtained by subtracting the elevations of each of the nearest neighbor points from the elevation of the data point itself; The reference elevation, the difference, and the average of the elevation difference are combined sequentially to form a multidimensional elevation semantic vector for the data point.

[0037] Within each spatial analysis unit already labeled with regional semantic tags, an elevation assessment benchmark is established. Using a quicksort algorithm, the absolute Z-coordinate values ​​of all data points within the unit (e.g., 1000 points with Z-coordinates ranging from 85.0m to 88.0m) are sorted in ascending order. The Z-value at the 50th percentile is extracted as the benchmark elevation. For example, if the benchmark elevation of the current unit is measured to be 86.5m, and a data point within the unit has an elevation of 87.2m, the difference between the data point's elevation and the benchmark elevation is calculated to obtain a self-elevation difference of +0.7m.

[0038] Based on this, a KD-tree nearest neighbor search is used to obtain a predetermined number of nearest neighbors for the data point in the Euclidean distance space. The preferred number of nearest neighbors is between 8 and 16; 10 is used as an example. The elevation difference between the data point's own elevation and the elevations of the 10 nearest neighbors is calculated one by one. The arithmetic mean of these 10 elevation differences is then calculated to obtain the average elevation difference, for example, +0.4m. The baseline elevation of 86.5m, the difference of 0.7m, and the average elevation difference of 0.4m are then combined sequentially to form a three-dimensional vector. This serves as the multidimensional high-level elevation semantic vector for that data point.

[0039] In one embodiment, the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies a second threshold condition based on the diagnostic modulus, including: A local point set is obtained with the data point as the center and a preset radius, and a covariance matrix is ​​constructed. The first eigenvalue, the second eigenvalue, and the third eigenvalue of the covariance matrix are calculated, and the first eigenvalue is not less than the second eigenvalue, and the second eigenvalue is not less than the third eigenvalue. The quotient obtained by dividing the difference between the second eigenvalue and the third eigenvalue by the first eigenvalue is used as the initial planarity measure factor before removing the data point. Remove the data points from the local point set, reconstruct the covariance matrix, and calculate the planarity measure factor after removal. The difference between the removed planarity measure factor and the initial planarity measure factor is calculated to obtain the planarity improvement amount; If the initial flatness measure factor is less than the first flatness threshold and the flatness improvement amount is greater than the flatness improvement threshold set based on the diagnostic module, then the second threshold condition is determined to be satisfied, wherein the first flatness threshold and the flatness improvement threshold are determined based on the diagnostic module.

[0040] Using the data point as the center, a radius-nearest neighbor search is used to obtain all neighboring points within a preset spatial radius to form a local point set. The preferred range of the preset radius is 0.5m to 2.5m; in this example, it is set to 1.5m. The three-dimensional coordinates of all points in the local point set are extracted to calculate a 3×3 spatial covariance matrix. Singular value decomposition is then used to obtain three non-negative eigenvalues, which are arranged in descending order as the first eigenvalue, the second eigenvalue, and the third eigenvalue, for example, 0.85, 0.65, and 0.30, respectively. The initial planarity measure factor is calculated using the formula (second eigenvalue - third eigenvalue) / first eigenvalue. In the aforementioned example, the initial planarity measure factor is approximately 0.41.

[0041] The data point is removed from the local point set. The covariance matrix is ​​reconstructed and eigenvalues ​​are calculated using the remaining points. For example, the new eigenvalues ​​become 0.88, 0.72, and 0.05 respectively. The planarity measure factor after removal is calculated using the same formula to be approximately 0.76. A first planarity threshold is set based on the diagnostic modulus, and a planarity improvement threshold that is negatively correlated with the diagnostic modulus is also set. For example, if the diagnostic modulus is between 0.1 and 1.0, and the current diagnostic modulus is 0.8, the first planarity threshold is the diagnostic modulus multiplied by 0.6, i.e., 0.48. The planarity improvement threshold can be determined by dividing the basic improvement threshold by (1 + diagnostic modulus). If the basic improvement threshold is 0.3, then the planarity improvement threshold is 0.3 / (1 + 0.8), approximately 0.17. In the example, the initial flatness measure factor of 0.41 is less than the first flatness threshold of 0.48. After removal, the difference between the flatness measure factor of 0.76 and the initial flatness measure factor of 0.41 is 0.35, and 0.35 is greater than the flatness improvement threshold of 0.17. Therefore, the second threshold condition is met, and the data point is determined to be a high-order gross error point and removed.

[0042] S4. Digital elevation models are constructed and differential calculations are performed on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the monitoring results of the solid waste volume.

[0043] On the XY two-dimensional horizontal projection plane, an orthogonal regular grid with fixed grid spacing is generated based on the maximum boundary coordinates of the solid waste slag site to be monitored. The clean point cloud data of the reference period and the current period are traversed. The Z-axis elevation value of each data point is projected and mapped onto the corresponding two-dimensional orthogonal grid. For grids containing multiple data points, an inverse distance weighted average algorithm is used to determine the initial elevation of the grid. For grids without data points, the `griddata` function is called to fill in missing values ​​using linear or cubic interpolation, generating a two-dimensional raster matrix format digital elevation model for each period after missing value filling. Element-wise subtraction of corresponding matrices is performed, and the reference period's digital elevation model matrix is ​​subtracted from the current period's digital elevation model matrix to obtain the elevation difference digital model representing the elevation change. The volume of newly added waste in landfill is obtained by summing up the elevation change pixel values ​​greater than zero in the elevation difference digital model and multiplying them by the area of ​​a single grid. The volume of waste sedimentation loss is obtained by summing up the absolute values ​​of elevation change pixels less than zero and multiplying them by the area of ​​a single grid. The combined monitoring results of the increase and loss are output as the basis for evaluation.

[0044] In one embodiment, the step of constructing digital elevation models and performing differential calculations on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the solid waste volume monitoring results includes: The clean point cloud data from each phase are projected onto the same reference horizontal plane and divided into regular grids of the same resolution. Interpolate the point cloud elevation within each regular grid to generate digital elevation models for the current period and the reference period; Calculate the elevation difference between the digital elevation models of the current period and the reference period under the same planar grid coordinates; The product of the elevation difference of all regular grids and the area of ​​a single grid is summed to obtain the monitoring result of the volume of solid waste.

[0045] The 3D point cloud data of each cleanroom phase is projected onto a predetermined reference horizontal plane, dividing the projection plane into a regular grid of equal length and width. The grid resolution, i.e., the side length of a single grid cell, is preferably set between 0.1m and 0.5m; in this example, 0.2m is used, corresponding to a single grid area of ​​0.04m². 2 For each point cloud falling within a grid, the inverse distance weighting method or kriging interpolation method is used to calculate the interpolation result of the elevation of the center point of that grid, thereby constructing digital elevation models for the current period and the reference period respectively.

[0046] In differential analysis, for each grid coordinate, the current grid elevation is subtracted from the reference grid elevation to obtain the elevation difference between the grids. For example, if the grid elevation changes from 80.5m to 82.0m, the elevation difference is +1.5m, indicating fill; a negative value indicates cut. The calculated elevation difference for each grid is then multiplied by the area of ​​a single grid, 0.04m². 2 Calculate the volume represented by a single grid cell, such as the elevation difference of 1.5m multiplied by the grid area of ​​0.04m². 2 A single volume of +0.06m was obtained. 3 By traversing all regular grids and summing the volume values, the monitoring results of the volume change of the solid waste slag site to be monitored are obtained.

[0047] Example 2

[0048] Embodiment 2 of the present invention proposes a solid waste volume monitoring device based on UAV oblique photography, comprising the following modules: The acquisition module is used to acquire three-dimensional point cloud data of each phase of the solid waste slag site to be monitored, including at least the current phase and the reference phase, and to independently perform data processing steps on the three-dimensional point cloud data of each phase. The calculation module is used to calculate the local information entropy gradient of the point cloud based on the local neighborhood of the data points in the three-dimensional point cloud data, divide the three-dimensional point cloud data into multiple spatial analysis units according to the local information entropy gradient of the point cloud, calculate the local structural complexity parameters of each spatial analysis unit, perform regional semantic segmentation on the solid waste slag site to be monitored, and determine regional semantic labels for each spatial analysis unit. The determination module is used to calculate a multidimensional high-level elevation semantic vector for each data point within each spatial analysis unit based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, determine the unit dominant semantic vector of each spatial analysis unit, and generate a corresponding diagnostic module according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic module, then the data point is determined to be a high-level gross error point and removed. The monitoring module is used to construct digital elevation models and perform differential calculations on the clean point cloud data of each period after removing high-level coarse-difference points to obtain the monitoring results of the solid waste volume.

[0049] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0050] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for monitoring the volume of solid waste based on UAV oblique photography, characterized in that, Includes the following steps: Acquire three-dimensional point cloud data of the solid waste slag site to be monitored, including at least the current period and the reference period, and perform data processing steps independently on the three-dimensional point cloud data of each period; Based on the local neighborhood of the data points in the three-dimensional point cloud data, the local information entropy gradient of the point cloud is calculated. According to the local information entropy gradient of the point cloud, the three-dimensional point cloud data is divided into multiple spatial analysis units, and the local structural complexity parameter of each spatial analysis unit is calculated. The region semantic segmentation of the solid waste slag site to be monitored is performed, and the region semantic label is determined for each spatial analysis unit. Within each spatial analysis unit, based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, a multidimensional high-level elevation semantic vector is calculated for each data point. The unit-dominant semantic vector of each spatial analysis unit is determined, and a corresponding diagnostic modulus is generated according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit-dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic modulus, then the data point is determined to be a high-level gross error point and removed. Digital elevation models were constructed and differential calculations were performed on the clean point cloud data of each period after removing high-level coarse-point errors to obtain the monitoring results of the solid waste volume.

2. The method according to claim 1, characterized in that, The step of dividing the 3D point cloud data into multiple spatial analysis units based on the local information entropy gradient of the point cloud, and calculating the local structural complexity parameters of each spatial analysis unit, includes: Construct an initial 3D bounding box containing the 3D point cloud data as the root node; Calculate the average value of the information entropy gradient of all data points within the current node as the local structural complexity parameter of the current node; Determine whether the local structural complexity parameter of the current node exceeds the preset partitioning threshold. If it does, and the size of the current node is larger than the preset minimum scale, then partition the current node into eight child nodes evenly. Repeat the above calculation and judgment process until the local structural complexity parameters of all child nodes are not greater than the partitioning threshold or the node size reaches the minimum scale, and then use the leaf nodes as the multiple spatial analysis units.

3. The method according to claim 2, characterized in that, The step of performing regional semantic segmentation on the solid waste slag site to be monitored, and determining regional semantic labels for each spatial analysis unit, includes: The average angle of inclination of the average normal vector is obtained by calculating the average angle between the normal vector of all data points in the spatial analysis unit and the vertical direction, and the elevation range of the data points in the spatial analysis unit is calculated. If the average normal vector tilt angle is less than the first tilt angle threshold and the elevation range is less than the first elevation threshold, then the regional semantic label of the spatial analysis unit is determined as a flat area at the top of the slope. If the average normal vector tilt angle is greater than the second tilt angle threshold, then the regional semantic label of the spatial analysis unit is determined as a steep slope area; If the spatial analysis unit is not identified as a flat area at the top of the slope or a steep area on the slope, then the regional semantic label of the spatial analysis unit is identified as a transition area at the foot of the slope.

4. The method according to claim 1, characterized in that, Within each spatial analysis unit, based on the elevation of the data points and the elevation relationship between the data points and neighboring points, a multidimensional elevation semantic vector is calculated for each data point, including: Sort the elevation values ​​of all data points within the spatial analysis unit, take the median elevation, and obtain the reference elevation. Calculate the difference between the elevation of the data point itself and the reference elevation; Calculate the average of the elevation differences obtained by subtracting the elevations of each of the nearest neighbor points from the elevation of the data point itself; The reference elevation, the difference, and the average of the elevation difference are combined sequentially to form a multidimensional elevation semantic vector for the data point.

5. The method according to claim 4, characterized in that, The planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic modulus, including: A local point set is obtained with the data point as the center and a preset radius, and a covariance matrix is ​​constructed. The first eigenvalue, the second eigenvalue, and the third eigenvalue of the covariance matrix are calculated, and the first eigenvalue is not less than the second eigenvalue, and the second eigenvalue is not less than the third eigenvalue. The quotient obtained by dividing the difference between the second eigenvalue and the third eigenvalue by the first eigenvalue is used as the initial planarity measure factor before removing the data point. Remove the data points from the local point set, reconstruct the covariance matrix, and calculate the planarity measure factor after removal. The difference between the removed planarity measure factor and the initial planarity measure factor is calculated to obtain the planarity improvement amount; If the initial flatness measure factor is less than the first flatness threshold and the flatness improvement amount is greater than the flatness improvement threshold set based on the diagnostic module, then the second threshold condition is determined to be satisfied, wherein the first flatness threshold and the flatness improvement threshold are determined based on the diagnostic module.

6. The method according to claim 1, characterized in that, The process of constructing digital elevation models and performing differential calculations on the clean point cloud data of each phase after removing high-level gross errors to obtain the solid waste volume monitoring results includes: The clean point cloud data from each phase are projected onto the same reference horizontal plane and divided into regular grids of the same resolution. Interpolate the point cloud elevation within each regular grid to generate digital elevation models for the current period and the reference period; Calculate the elevation difference between the digital elevation models of the current period and the reference period under the same planar grid coordinates; The product of the elevation difference of all regular grids and the area of ​​a single grid is summed to obtain the monitoring result of the volume of solid waste.

7. A solid waste volume monitoring device based on UAV oblique photography, characterized in that, Includes the following modules: The acquisition module is used to acquire three-dimensional point cloud data of each phase of the solid waste slag site to be monitored, including at least the current phase and the reference phase, and to independently perform data processing steps on the three-dimensional point cloud data of each phase. The calculation module is used to calculate the local information entropy gradient of the point cloud based on the local neighborhood of the data points in the three-dimensional point cloud data, divide the three-dimensional point cloud data into multiple spatial analysis units according to the local information entropy gradient of the point cloud, calculate the local structural complexity parameters of each spatial analysis unit, perform regional semantic segmentation on the solid waste slag site to be monitored, and determine regional semantic labels for each spatial analysis unit. The determination module is used to calculate a multidimensional high-level elevation semantic vector for each data point within each spatial analysis unit based on the elevation of the data point and the elevation relationship between the data point and its neighboring points, determine the unit dominant semantic vector of each spatial analysis unit, and generate a corresponding diagnostic module according to the local structural complexity parameter and the regional semantic label. If the deviation distance of the multidimensional high-level elevation semantic vector of the data point from the unit dominant semantic vector is greater than a first threshold and has a positive elevation difference, and the planarity measure factor of the local point set centered on the data point before and after removing the data point satisfies the second threshold condition set based on the diagnostic module, then the data point is determined to be a high-level gross error point and removed. The monitoring module is used to construct digital elevation models and perform differential calculations on the clean point cloud data of each period after removing high-level coarse-difference points to obtain the monitoring results of the solid waste volume.

8. The apparatus according to claim 7, characterized in that, The step of dividing the 3D point cloud data into multiple spatial analysis units based on the local information entropy gradient of the point cloud, and calculating the local structural complexity parameters of each spatial analysis unit, includes: Construct an initial 3D bounding box containing the 3D point cloud data as the root node; Calculate the average value of the information entropy gradient of all data points within the current node as the local structural complexity parameter of the current node; Determine whether the local structural complexity parameter of the current node exceeds the preset partitioning threshold. If it does, and the size of the current node is larger than the preset minimum scale, then partition the current node into eight child nodes evenly. Repeat the above calculation and judgment process until the local structural complexity parameters of all child nodes are not greater than the partitioning threshold or the node size reaches the minimum scale, and then use the leaf nodes as the multiple spatial analysis units.

9. The apparatus according to claim 7, characterized in that, The step of performing regional semantic segmentation on the solid waste slag site to be monitored, and determining regional semantic labels for each spatial analysis unit, includes: The average angle of inclination of the average normal vector is obtained by calculating the average angle between the normal vector of all data points in the spatial analysis unit and the vertical direction, and the elevation range of the data points in the spatial analysis unit is calculated. If the average normal vector tilt angle is less than the first tilt angle threshold and the elevation range is less than the first elevation threshold, then the regional semantic label of the spatial analysis unit is determined as a flat area at the top of the slope. If the average normal vector tilt angle is greater than the second tilt angle threshold, then the regional semantic label of the spatial analysis unit is determined as a steep slope area; If the spatial analysis unit is not identified as a flat area at the top of the slope or a steep area on the slope, then the regional semantic label of the spatial analysis unit is identified as a transition area at the foot of the slope.

10. The apparatus according to claim 7, characterized in that, Within each spatial analysis unit, based on the elevation of the data points and the elevation relationship between the data points and neighboring points, a multidimensional elevation semantic vector is calculated for each data point, including: Sort the elevation values ​​of all data points within the spatial analysis unit, take the median elevation, and obtain the reference elevation. Calculate the difference between the elevation of the data point itself and the reference elevation; Calculate the average of the elevation differences obtained by subtracting the elevations of each of the nearest neighbor points from the elevation of the data point itself; The reference elevation, the difference, and the average of the elevation difference are combined sequentially to form a multidimensional elevation semantic vector for the data point.

Images