Strip mine area muck pile volume calculation method based on three-dimensional visual point cloud

Abstract

The invention discloses a method for calculating the volume of a muck pile in a strip mine area based on three-dimensional visual point cloud. The method comprises the following steps: step 1, extracting frames from a video of a whole muck pile shot by an unmanned aerial vehicle to obtain a muck pile picture, and processing the collected picture and video by using colmap to obtain internal reference and pose information of a camera; then, taking the picture and camera internal reference and pose information as input to be transmitted into a neural radiation field model, completing three-dimensional reconstruction of the muck pile, and obtaining point cloud data of the muck pile; step 2, processing the point cloud data of the muck pile to obtain morphological parameters of a longitudinal profile, a volume and a throwing distance of the muck pile; according to the method, three-dimensional reconstruction is performed on the muck pile through the neural radiation field, the muck pile point cloud is generated, and then the muck pile point cloud is processed, so that the muck pile volume and other morphological parameters are obtained quickly and accurately.

Description

Technical Field

[0001] The invention relates to the technical field of open-pit mine operation optimization, and in particular to a method for calculating the volume of an open-pit mine blast pile based on three-dimensional visual point cloud. Background Art

[0002] In the production of open-pit mines, blasting and loading and transportation are one of the links. The quality of blasting results directly affects the efficiency of subsequent shoveling and transportation processes and the comprehensive production cost of the project. The throwing distance, volume and other morphological parameters of the blast pile are important components for evaluating the blasting effect of the blast pile. It is very important to determine the volume and other morphological parameters of the blast pile before and after blasting in a timely and accurate manner.

[0003] At present, most of the methods for obtaining the morphological parameters of the explosion pile at home and abroad are to collect the information of the explosion pile through non-contact equipment such as laser radar, cameras, etc., and then obtain the volume information of the explosion pile by processing the laser point cloud or pictures of the explosion pile. Since 3D laser scanners are often expensive and have high requirements for operators, they cannot quickly and conveniently obtain the required data. With the continuous research in the field of computer vision and the continuous updating and improvement of visual 3D reconstruction technology, visual 3D reconstruction technologies such as motion structure recovery and neural radiation field are used in the reconstruction of real objects.

[0004] At present, the reconstruction of exploded piles at home and abroad is mostly completed through software and motion structure recovery and other technologies. However, whether it is through software or motion structure recovery algorithms, it has high requirements on computer equipment and takes a long time to process. Summary of the invention

[0005] In order to overcome the above technical problems, the purpose of the present invention is to provide a method for calculating the volume of explosive piles in open-pit mines based on three-dimensional visual point clouds. The method reconstructs the explosive piles in three dimensions through a neural radiation field and generates an explosive pile point cloud, and then processes the explosive pile point cloud, so as to quickly and accurately obtain the explosive pile volume and other morphological parameters.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for calculating the volume of an open-pit mine blast pile based on three-dimensional visual point cloud comprises the following steps:

[0008] Step 1: Extract frames from the video of the entire explosion pile taken by the drone to obtain the explosion pile picture, use colmap to process the collected pictures and videos, and obtain the camera's intrinsic parameters and pose information; then pass the picture and the camera's intrinsic parameters and pose information as input into the neural radiation field model to complete the 3D reconstruction of the explosion pile and obtain the point cloud data of the explosion pile;

[0009] Step 2: Process the point cloud data of the explosion pile to obtain the longitudinal section, volume, and throwing distance morphological parameters of the explosion pile.

[0010] The step 1 is specifically as follows:

[0011] Specifically:

[0012] Step 1.1: Use a drone to shoot the explosive pile. The captured data is images or videos. After collecting the video data, it is necessary to extract frames from the video to produce an image dataset. After obtaining the image dataset, the incremental Structure from Motion (SfM) module in colmap is used to process the collected two-dimensional images.

[0013] The structure from motion module recovers the 3D shape and camera trajectory from a series of 2D images. First, the SIFT algorithm is used to perform feature matching on the disordered images, which can be expressed as:

[0014] D=f(I)

[0015] Where D is the extracted feature descriptor, I is the input image, and f is the feature extraction function of the SIFT algorithm;

[0016] The outliers are filtered out through geometric verification to obtain the matching relationship of the image. Then, the pose of the camera that took the image is estimated based on the existing motion recovery structure module. The sparse structure of the point cloud is restored through triangulation. Then, the bundle adjustment (BA) method is used to optimize the camera pose C and the point cloud position P to reduce the cumulative error. The specific objective function E is:

[0017]

[0018] Where n is the number of cameras, m is i is the number of point clouds observed in the i-th camera view, ρ is a robust cost function, and u ij is the observed image coordinates of the j-th point cloud in the i-th camera view, π(K i ,R i ,t i ,X i ) is the reprojection function, which projects the 3D point onto the 2D plane through the camera parameters, where K i is the intrinsic parameter matrix of the i-th camera, R i is the rotation matrix of the i-th camera, t i is the translation vector of the i-th camera, X j is the coordinate of the jth point cloud;

[0019] Through the above steps, the intrinsic pose information of the camera is finally obtained;

[0020] Step 1.2: 3D reconstruction of the burst pile

[0021] After obtaining the intrinsic pose information of the camera, the burst image dataset and the parameter and pose information file in step 1.1 are input into the Instant-NGP (Instant Neural Graphics Primitives) neural radiation field model. The input involves the three-dimensional space coordinates x = (x, y, z) and the two-dimensional viewing direction The output includes the volume density σ and color c = (r, g, b) at the spatial position;

[0022] The expression is: F(x,d)→(c,σ); Instant-NGP uses Multiresolution Hash Encoding, which takes the three-dimensional space coordinate x and the two-dimensional viewing direction d as input, divides the three-dimensional space into multiple levels of voxels, and uses a neural network inside each voxel to predict features;

[0023] After the encoding is completed, the high-frequency three-dimensional spatial coordinate x and the two-dimensional viewing direction d are obtained, both expressed in vector form;

[0024] The five-dimensional coordinates are queried along the emitted light, and the obtained color and density information are mapped to the image using stereo rendering technology. The loss between the rendered image and the actual image is calculated. The loss is defined as the mean square error (MSE) loss between the color of the rendering result and the true color of the image. The network weights are optimized through back propagation, and the scene represented by the neural radiation field is repeatedly optimized and iterated to realize the construction of a three-dimensional model.

[0025] The step 2 is specifically as follows:

[0026] Step (1): Point cloud preprocessing: Radius filtering is performed on the point cloud model of the explosion pile obtained by the neural radiation field to remove the noise and outliers in the mining area point cloud, and the point cloud is restored to its actual size by calculating the ratio of the size of the truck in the explosion pile point cloud to the actual size of the truck;

[0027] The point clouds of the blast pile and its surrounding ground environment before and after the blast are intercepted and filtered again through radius filtering to further reduce the impact of noise on subsequent processing;

[0028] Step (2): Construct a complete explosion pile: After the point cloud preprocessing is completed, the preprocessed point cloud is processed to construct a complete explosion pile point cloud model to obtain the real volume information of the point cloud.

[0029] Step (3): Geometric reconstruction and volume calculation of the explosion pile: Use greedy triangulation reconstruction to geometrically reconstruct the explosion pile point cloud, obtain the mesh model, and finally calculate the volume information of the explosion pile.

[0030] The step (1) is specifically:

[0031] Set two key parameters: the radius of the query point and the minimum number of points in the neighborhood k. For each point p in the point cloud i , the neighborhood point set N(p i ) is expressed as:

[0032] N(p i )={p j ∈P|d(p i ,p j )≤radius}

[0033] Where P is the point cloud set, d(p i ,p j ) is point p i and point p j The Euclidean distance between

[0034] Then calculate the number of points in the neighborhood m = |N(p i )|, if m<k, then point p i Mark as outliers. This process needs to be repeated for all points in the point cloud to ensure that each point has been evaluated accordingly and finally all points marked as outliers are removed;

[0035] The scale of the filtered explosion pile point cloud is restored by calculating the ratio of the real length of the object in the point cloud scene to the length of the object in the point cloud, and then the scale of the entire point cloud scene is restored by this ratio, thereby restoring the scale information of the real explosion pile scene;

[0036] The point cloud of the explosive pile is intercepted by cloudcompare software, and the point cloud part of the explosive pile and its surrounding ground environment is intercepted, and finally the point cloud of the explosive pile and its surrounding ground environment is obtained.

[0037] The step (2) is specifically:

[0038] Step (2.1): Use the CSF algorithm to extract the ground from the point cloud of the explosive pile and its surrounding ground environment to obtain the ground point cloud;

[0039] Step (2.2): intercept the explosive pile again to obtain point cloud data of only the surface of the explosive pile, grow the surface point cloud of the explosive pile downward to the ground point cloud in (2.1), construct a closed point cloud of the explosive pile, obtain the bulge height, throwing distance, and width of the explosive pile through geometric calculation, and obtain the cross-sectional view of the explosive pile through the slicing method.

[0040] The step (2.1) is specifically:

[0041] First, the original point cloud data is inverted so that the original ground point is at the top. A virtual cloth grid is generated above the inverted point cloud to simulate the falling process of the cloth under the action of gravity. The cloth particle position update formula is as follows

[0042] X t+1 =X t +Δt·(F ext (X t ,t)+F int (X t ))

[0043] Where Δt is the time from t to t+1, X t is the position of the cloth particle at time t, F ext is the external force (such as gravity), F int are internal forces (such as elasticity and tension between cloth particles);

[0044] When the cloth reaches a stable state, points that are in contact with or very close to the cloth are considered to be ground points.

[0045] The step (2.2) is specifically:

[0046] After the bottom point cloud of the implosion pile is extracted, the point cloud is grown downward from the surface of the implosion pile point cloud to fill the blank area between the surface point cloud and the bottom point cloud of the implosion pile, thereby constructing a complete implosion pile point cloud model;

[0047] For each point in the explosion point cloud, move downward along its normal direction, each time moving a fixed step size step_size until reaching the ground point cloud;

[0048] After the complete point cloud of the explosive pile is constructed, the morphological parameter information of the bulge height and throwing distance of the explosive pile is calculated, and the longitudinal profile of the explosive pile is drawn.

[0049] The step (3) is specifically:

[0050] Greedy triangulation reconstruction is adopted to geometrically reconstruct the complete explosion pile point cloud data. First, the three-dimensional point cloud of the explosion pile is projected onto a two-dimensional plane and triangulated. Through the Delaunay triangulation method, a triangular patch is used as the initial step to expand the complete mesh surface and obtain the topological relationship between the points. Finally, the meshes are connected in three dimensions according to the topological relationship between the points to obtain the surface model of the complete explosion pile.

[0051] After obtaining the triangulated model of the explosion pile, for the triangulated mesh models of different explosion piles, first determine its lowest point O in the three-dimensional space; then, traverse each triangle in the mesh, extract the three vertices A, B, and C of the triangle, and calculate the position coordinates of these vertices relative to point O; use the determinant calculation method to obtain the volume of the tetrahedron formed by these three vertices and point O; and finally calculate the total volume V of the entire mesh model by continuously accumulating the volumes of these tetrahedrons.

[0052]

[0053] Among them, x A ,y A , z A They are the x, y, and z axis coordinate values ​​of vertex A respectively.

[0054] Beneficial effects of the present invention:

[0055] The present invention aims at the problem of three-dimensional reconstruction and morphological parameter information acquisition of open-pit mine blast piles. The data set of open-pit mine blast piles is collected by drone or camera, colmap is used to process the data set, and the blast piles are reconstructed in three dimensions through the Instant-NGP model. A three-dimensional reconstruction model of open-pit mine blast piles is constructed, which effectively solves the problem of three-dimensional reconstruction of open-pit mine blast piles. For the morphological parameters of the blast piles, the morphological parameters of the blast piles are acquired through the steps of blast pile point cloud preprocessing, construction of complete blast pile point cloud, blast pile geometric reconstruction, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of three-dimensional reconstruction of open-pit mine blast pile based on neural radiation field in the present invention.

[0057] Figure 2 It is a flow chart of open-pit mine blasting pile point cloud processing based on three-dimensional visual point cloud in the present invention. DETAILED DESCRIPTION

[0058] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0059] A method for calculating the volume of blast piles in open-pit mines based on three-dimensional visual point clouds comprises the following steps:

[0060] Step 1: First, the 3D reconstruction process of the explosion pile is as follows: Figure 1 As shown, the present invention proposes a 3D reconstruction technology for open-pit mine blast pile based on neural radiation field, and the implementation process of the technology mainly includes the following steps:

[0061] Step 1: Construction and processing of open-pit mine explosion image dataset

[0062] Using drones to photograph the explosive pile can capture the entire explosive pile and its surrounding environment. The captured data can be either images or videos. After collecting video data, it is necessary to extract frames from the video to create an image data set. This can greatly avoid the inability to obtain sufficient image feature information due to insufficient overlap. After obtaining the image data set, the incremental Structure from Motion (SfM) module in colmap is used to process the collected two-dimensional images. SfM can restore the three-dimensional shape and camera trajectory from a series of two-dimensional images. It first uses the scale-invariant feature transform (SIFT) algorithm to perform feature matching on unordered images. The main goal of the SIFT algorithm is to extract key points in the image and calculate its feature descriptor. It can be expressed as:

[0063] D=f(I)

[0064] Where D is the extracted feature descriptor, I is the input image, and f is the feature extraction function of the SIFT algorithm.

[0065] The image matching relationship is obtained by filtering out the external points through geometric verification, and then the pose of the camera that took the image is estimated based on the existing SfM model. The sparse structure of the point cloud is restored through triangulation, and then the bundle adjustment (BA) method is used to optimize the camera pose C and the point cloud position P to reduce the cumulative error. The specific objective function E is:

[0066]

[0067] Where n is the number of cameras, m is i is the number of point clouds observed in the i-th camera view, ρ is a robust cost function, and u ij is the observed image coordinates of the j-th point cloud in the i-th camera view, π(K i ,R i ,t i ,X i ) is the reprojection function, which projects the 3D point onto the 2D plane through the camera parameters. i is the intrinsic parameter matrix of the i-th camera, R i is the rotation matrix of the i-th camera, t i is the translation vector of the i-th camera, Xj are the coordinates of the jth point cloud.

[0068] Through the above steps, the intrinsic pose information of the camera is finally obtained.

[0069] Step 2: 3D reconstruction of the explosion pile

[0070] After obtaining the intrinsic pose information of the camera, the burst image dataset and the parameter and pose information file in step 1 are input into the Instant-NGP (Instant Neural Graphics Primitives) neural radiation field model. The input involves the three-dimensional space coordinates x = (x, y, z) and the two-dimensional viewing direction The output includes the volume density σ and color c = (r, g, b) of the spatial position, which is mathematically expressed as: F(x, d) → (c, σ). Due to the limitations of neural networks in processing low-frequency information, directly using spatial position and perspective as input may result in reduced resolution of the rendered image. Therefore, Instant-NGP improves efficiency by using multiresolution hash encoding, which takes the three-dimensional spatial coordinate x and the two-dimensional perspective direction d as input, divides the three-dimensional space into multiple levels of voxels, and uses neural networks inside each voxel to predict features. These features are obtained by querying the hash table. When querying, the spatial position of the point is used to find nearby voxels, and the features of the point are obtained by trilinear interpolation. These features are calculated at different resolution levels and finally spliced ​​together to form the final feature vector. This encoding structure makes a trade-off between quality, memory, and performance, and can be optimized by adjusting the size of the hash table, the size of the feature vector, and the number of resolutions.

[0071] After the encoding is completed, the high-frequency three-dimensional spatial coordinates x and the two-dimensional viewing direction d are obtained, both expressed in vector form. Instant-NGP uses a small-scale neural network to process these vectors. This method reduces the size of the network and speeds up training and rendering. Then, in the process of synthesizing the view, the five-dimensional coordinates are queried along the emitted light, and the obtained color and density information are mapped to the image using stereo rendering technology. By calculating the loss between the rendered image and the actual image, the loss is defined as the mean square error (MSE) loss between the color of the rendering result and the true color of the image, and optimizing the network weights through back propagation, the scene represented by the neural radiation field is repeatedly optimized and iterated to achieve the construction of a three-dimensional model.

[0072] Since the Instant-NGP model is an implicitly expressed 3D scene, it does not directly generate an explicitly expressed 3D point cloud or mesh model. Therefore, Instant-NGP uses the marching_cubes algorithm to extract the mesh of the 3D scene obtained in Instant-NGP, obtains the mesh data of the burst, and then obtains the point cloud data of the burst through the code provided by the Instant-NGP official, and finally realizes the construction of the 3D model.

[0073] After the three-dimensional reconstruction of the explosion pile is completed, the point cloud data of the explosion pile is obtained, and the next step is to process the point cloud data.

[0074] Step 2: If Figure 2 As shown, the present invention proposes an open-pit mine blasting pile point cloud parameter processing technology based on three-dimensional visual point cloud, and the implementation process of the technology mainly includes the following steps.

[0075] Step 1: Preprocessing of burst point cloud. The preprocessing mainly includes three steps: radius filtering, scale recovery and burst interception;

[0076] The first step is to perform radius filtering on the point cloud data. Radius filtering is a point cloud denoising technology. Its core is to carefully analyze the local neighborhood of each point in the point cloud to identify and remove isolated points that are far away from neighboring points. These isolated points are often noise caused by measurement errors or other factors, which has an adverse effect on the accuracy and subsequent processing of the point cloud. In order to implement radius filtering, two key parameters need to be set: the radius of the query point and the minimum number of points in the neighborhood k. These two parameters work together to determine which points will be judged as outliers by the system and removed from the point cloud data. For each point p in the point cloud i , the neighborhood point set N(p i ) is expressed as:

[0077] N(p i )={p j ∈P|d(p i , p j )≤radius}

[0078] Where P is the point cloud set, d(p i , p j ) is point p i and point p j The Euclidean distance between .

[0079] Then calculate the number of points in the neighborhood m = |N(p i )|, if n<k, then point p iMark as outliers. This process needs to be repeated for all points in the point cloud to ensure that each point has been evaluated accordingly and finally all points marked as outliers are removed.

[0080] Through radius filtering, the noise in the point cloud can be effectively removed and the quality of the point cloud data can be improved. Since the point cloud data reconstructed by Instant-NGP does not have real scale information, it is necessary to restore the scale of the filtered explosion pile point cloud. By calculating the ratio of the real length of the object in the point cloud scene to the length of the object in the point cloud, the scale ratio is obtained, and then the scale of the entire point cloud scene is restored by this ratio, thereby restoring the scale information of the real explosion pile scene. Since the point cloud of the explosion pile scene contains the point cloud of the explosion pile and its surrounding environment, it is necessary to intercept the explosion pile point cloud through the cloudcompare software, and intercept the point cloud part of the explosion pile and its surrounding ground environment, so as to facilitate the subsequent ground extraction and construction of a complete explosion pile and other steps. This step finally obtains the point cloud of the explosion pile and its surrounding ground environment.

[0081] Step 2: Ground extraction of point cloud

[0082] After the preprocessing is completed, the point cloud of the explosive pile and its surrounding ground environment is obtained, and step 2 processes the point cloud of the explosive pile and its surrounding ground environment. Since the point cloud of the explosive pile reconstructed by Instant-NGP only contains the surface data of the explosive pile, the real volume of the explosive pile cannot be calculated with the surface data of the explosive pile, so it is necessary to construct a complete point cloud of the explosive pile, which means that the bottom surface of the explosive pile after preprocessing needs to be extracted, and the point cloud of the explosive pile surface needs to be grown downward to complete the construction of the complete point cloud of the explosive pile.

[0083] The present invention uses the CSF (Cloth Simulation Filter) algorithm to extract the bottom surface of the explosion pile. The principle of the CSF algorithm is to simulate the point cloud data into a terrain model, and then cover the model with a layer of virtual cloth. By simulating the natural falling process of the cloth under the action of gravity, the cloth is closely attached to the terrain surface, thereby achieving effective separation of ground points. The specific steps are as follows:

[0084] First, invert the original point cloud data so that the original ground point is at the top. Generate a virtual cloth mesh above the inverted point cloud. The resolution of the cloth mesh is a key parameter that affects the accuracy and computational efficiency of the algorithm. This parameter is used as the input parameter of the csf function in the code. By continuously debugging the code, the optimal result of the ground simulation is obtained. Simulate the falling process of cloth under the action of gravity. This step involves many aspects of cloth physics simulation, including the interaction force between cloth particles, collision detection between cloth and point cloud, etc. The formula for updating the position of cloth particles is as follows

[0085] Xt+1 =X t +Δt·(F ext (X t , t)+F int (X t ))

[0086] Where Δt is the time from t to t+1, X t is the position of the cloth particle at time t, F ext is the external force (such as gravity), F int are internal forces (such as elasticity and tension between cloth particles).

[0087] When the cloth reaches a stable state, points that are in contact with or very close to the cloth are considered ground points. This process is determined by setting a classification threshold. In terms of specific implementation, the implementation of the CSF algorithm requires adjustment of five parameters: cloth resolution, rigidity, time step, class threshold, and iterations. For different burst piles and their surrounding ground point cloud data, these parameters need to be carefully adjusted to obtain the best ground extraction effect. After the bottom point cloud of the burst pile is extracted, the point cloud is grown downward from the surface of the burst pile point cloud to fill the blank area between the surface point cloud and the bottom point cloud of the burst pile, thereby constructing a complete burst pile point cloud model. For each point in the burst pile point cloud, move downward along its normal direction, each time moving a fixed step size step_size until the ground point cloud is reached.

[0088] After the complete point cloud of the blast pile is constructed, the morphological parameter information such as the bulge height and throwing distance of the blast pile can be calculated, and the longitudinal section of the blast pile can be drawn. This study calculates the bulge height of the blast pile by calculating the distance from the highest point of the blast pile to the ground of the blast pile, calculates the throwing distance of the blast pile by calculating the distance from the point with the maximum x-coordinate value of the blast pile to the back of the blast pile, and obtains the longitudinal section of the blast pile through the section method.

[0089] Step 3: Geometry reconstruction and volume calculation of the explosion pile

[0090] In step 2, the complete explosion pile point cloud data is constructed. In order to obtain the real volume information of the explosion pile, the present invention studies the reconstruction method of the explosion pile, and the present invention adopts greedy triangulation reconstruction to geometrically reconstruct the complete explosion pile point cloud data. The algorithm first projects the three-dimensional point cloud of the explosion pile onto a two-dimensional plane and triangulates it. Through the Delaunay triangulation method, a triangular facet is used as the initialization to expand the complete mesh surface and obtain the topological relationship between the points. Finally, the meshes are connected in three dimensions according to the topological relationship between the points to obtain a surface model of the complete explosion pile. In terms of the specific implementation of the PCL library in C++, for different explosion pile point cloud data, multiple parameters need to be adjusted to obtain the optimal reconstruction result.

[0091] After obtaining the triangulated model of the explosion pile, for the triangulated mesh models of different explosion piles, first determine its lowest point O in three-dimensional space. Then, traverse each triangle in the mesh, extract the three vertices A, B, and C of the triangle, and calculate the position coordinates of these vertices relative to point O. Using the determinant calculation method, the volume of the tetrahedron formed by these three vertices and point O is obtained. By continuously accumulating the volumes of these tetrahedrons, the total volume V of the entire mesh model is finally calculated.

[0092]

[0093] Among them, x A ,y A , z A They are the x, y, and z axis coordinate values ​​of vertex A respectively.

[0094] The present invention aims at the problem of three-dimensional reconstruction and morphological parameter information acquisition of open-pit mine blast piles. The data set of open-pit mine blast piles is collected by drone or camera, colmap is used to process the data set, and the blast piles are reconstructed in three dimensions through the Instant-NGP model. A three-dimensional reconstruction model of open-pit mine blast piles is constructed, which effectively solves the problem of three-dimensional reconstruction of open-pit mine blast piles. For the morphological parameters of the blast piles, the morphological parameters of the blast piles are acquired through the steps of blast pile point cloud preprocessing, construction of complete blast pile point cloud, blast pile geometric reconstruction, etc.

Claims

1. A method for calculating the volume of blast piles in open-pit mines based on three-dimensional visual point clouds, characterized in that: The steps include: Step 1: Extract the video frames from the entire explosion pile video taken by the drone to obtain the explosion pile picture, use colmap to process the collected pictures and videos, and obtain the camera's intrinsic parameters and pose information; then pass the picture and the camera's intrinsic parameters and pose information as input into the neural radiation field model to complete the 3D reconstruction of the explosion pile and obtain the point cloud data of the explosion pile; Step 2: Process the point cloud data of the explosion pile to obtain the longitudinal section, volume, throwing distance morphological parameters of the explosion pile, and the total volume V of the entire grid model.

2. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 1 is characterized in that: The step 1 is specifically as follows: Specifically: Step 1.1: Use a drone to shoot the explosive pile. The captured data is images or videos. After collecting the video data, it is necessary to extract frames from the video to create an image dataset. After obtaining the image dataset, the incremental motion recovery structure module in colmap is used to process the collected two-dimensional images. The structure from motion module recovers the 3D shape and camera trajectory from a series of 2D images. First, the scale-invariant feature transformation algorithm is used to perform feature matching on the disordered images, which can be expressed as: D=f(I) Where D is the extracted feature descriptor, I is the input image, and f is the feature extraction function of the SIFT algorithm; The outliers are filtered out through geometric verification to obtain the matching relationship of the image. Then, the pose of the camera that took the image is estimated based on the existing motion recovery structure module. The sparse structure of the point cloud is restored through triangulation. The bundle adjustment method is then used to optimize the camera pose C and the point cloud position P to reduce the cumulative error. The specific objective function E is: Where n is the number of cameras, m is i is the number of point clouds observed in the i-th camera view, ρ is a robust cost function, and u ij is the observed image coordinates of the j-th point cloud in the i-th camera view, π(K i ,R i ,t i ,X i ) is the reprojection function, which projects the 3D point onto the 2D plane through the camera parameters, where K i is the intrinsic parameter matrix of the i-th camera, R i is the rotation matrix of the i-th camera, t i is the translation vector of the i-th camera, X j is the coordinate of the jth point cloud; Through the above steps, the intrinsic pose information of the camera is finally obtained; Step 1.2: 3D reconstruction of the burst pile After obtaining the intrinsic pose information of the camera, the burst image dataset and the parameter and pose information file in step 1.1 are input into the Instant-NGP neural radiation field model. The input involves the three-dimensional spatial coordinates x = (x, y, z) and the two-dimensional viewing direction The output includes the volume density σ and color c = (r, g, b) at the spatial position; The expression is: F(x,d)→(c,σ); Instant-NGP uses multi-resolution hash coding, which takes the three-dimensional space coordinate x and the two-dimensional viewing direction d as input, divides the three-dimensional space into multiple levels of voxels, and uses a neural network inside each voxel to predict features; After the encoding is completed, the high-frequency three-dimensional spatial coordinate x and the two-dimensional viewing direction d are obtained, both expressed in vector form; The five-dimensional coordinates are queried along the emitted light, and the obtained color and density information are mapped to the image using stereo rendering technology. The loss between the rendered image and the actual image is calculated (the loss is defined as the mean square error loss between the color of the rendering result and the true color of the image). The network weights are optimized through back propagation, and the scene represented by the neural radiation field is repeatedly optimized and iterated to achieve the construction of a three-dimensional model.

3. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 1 is characterized in that: The step 2 is specifically as follows: Step (1): Point cloud preprocessing: Radius filtering is performed on the point cloud model of the explosion pile obtained by the neural radiation field to remove the noise and outliers in the mining area point cloud, and the point cloud is restored to its actual size by calculating the ratio of the size of the truck in the explosion pile point cloud to the actual size of the truck; The point clouds of the blast pile and its surrounding ground environment before and after the blast are intercepted and filtered again through radius filtering to further reduce the impact of noise on subsequent processing; Step (2): Constructing a complete explosion pile: After the point cloud preprocessing is completed, the preprocessed point cloud is processed to construct a complete explosion pile point cloud model to obtain the real volume information of the point cloud; Step (3): Geometric reconstruction and volume calculation of the explosion pile: Use greedy triangulation reconstruction to geometrically reconstruct the explosion pile point cloud, obtain the mesh model, and finally calculate the volume information of the explosion pile.

4. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 3 is characterized in that: The step (1) is specifically: Set two key parameters: the radius of the query point and the minimum number of points in the neighborhood k. For each point p in the point cloud i , the neighborhood point set N(p i ) is expressed as: N(p i )={p j ∈P|d(p i ,p j )≤radius} Where P is the point cloud set, d(p i ,p j ) is point p i and point p j The Euclidean distance between Then calculate the number of points m = |N(p i )| in the neighborhood. If n < k, then mark the point p i as an outlier. This process needs to be repeated for all points in the point cloud to ensure that each point has been evaluated accordingly. Finally, remove all the points marked as outliers; The scale of the filtered explosion pile point cloud is restored by calculating the ratio of the real length of the object in the point cloud scene to the length of the object in the point cloud, and then the scale of the entire point cloud scene is restored by this ratio, thereby restoring the scale information of the real explosion pile scene; The point cloud of the explosive pile is intercepted by cloudcompare software, and the point cloud part of the explosive pile and its surrounding ground environment is intercepted, and finally the point cloud of the explosive pile and its surrounding ground environment is obtained.

5. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 3 is characterized in that: The step (2) is specifically: Step (2.1): Use the CSF algorithm to extract the ground from the point cloud of the explosive pile and its surrounding ground environment to obtain the ground point cloud; Step (2.2): intercept the explosive pile again to obtain point cloud data of only the surface of the explosive pile, grow the surface point cloud of the explosive pile downward to the ground point cloud in (2.1), construct a closed point cloud of the explosive pile, obtain the bulge height, throwing distance, and width of the explosive pile through geometric calculation, and obtain the cross-sectional view of the explosive pile through the slicing method.

6. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 5 is characterized in that: The step (2.1) is specifically: First, the original point cloud data is inverted so that the original ground point is at the top. A virtual cloth grid is generated above the inverted point cloud to simulate the falling process of the cloth under the action of gravity. The cloth particle position update formula is as follows X t+1 =X t +Δt·(F ext (X t ,t)+F int (X t )) Where Δt is the time from t to t+1, X t is the position of the cloth particle at time t, F ext is the external force, F int is an internal force; When the cloth reaches a stable state, points that are in contact with or very close to the cloth are considered to be ground points.

7. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 5 is characterized in that: The step (2.2) is specifically: After the bottom point cloud of the implosion pile is extracted, the point cloud is grown downward from the surface of the implosion pile point cloud to fill the blank area between the surface point cloud and the bottom point cloud of the implosion pile, thereby constructing a complete implosion pile point cloud model; For each point in the explosion point cloud, move downward along its normal direction, each time moving a fixed step size step_size until reaching the ground point cloud; After the complete point cloud of the explosive pile is constructed, the morphological parameter information of the bulge height and throwing distance of the explosive pile is calculated, and the longitudinal profile of the explosive pile is drawn.

8. The method for calculating the volume of blast pile in open-pit mine area based on three-dimensional visual point cloud according to claim 3 is characterized in that: The step (3) is specifically: Greedy triangulation reconstruction is adopted to geometrically reconstruct the complete explosion pile point cloud data. First, the three-dimensional point cloud of the explosion pile is projected onto a two-dimensional plane and triangulated. Through the Delaunay triangulation method, a triangular patch is used as the initial step to expand the complete mesh surface and obtain the topological relationship between the points. Finally, the meshes are connected in three dimensions according to the topological relationship between the points to obtain the surface model of the complete explosion pile. After obtaining the triangulated model of the explosion pile, for the triangulated mesh models of different explosion piles, first determine its lowest point O in the three-dimensional space; then, traverse each triangle in the mesh, extract the three vertices A, B, and C of the triangle, and calculate the position coordinates of these vertices relative to point O; use the determinant calculation method to obtain the volume of the tetrahedron formed by these three vertices and point O; and finally calculate the total volume V of the entire mesh model by continuously accumulating the volumes of these tetrahedrons. Among them, x A ,y A , z A They are the x, y, and z axis coordinate values ​​of vertex A respectively.

Images