Coal mine underground space reconstruction method and device based on dynamic segmentation octree

Abstract

This disclosure proposes a method and apparatus for reconstructing underground coal mine space using dynamically segmented octrees, relating to the field of artificial intelligence technology. The method includes: acquiring a sequence of images of the underground coal mine space and motion estimation data of the image sequences; generating point cloud coordinates corresponding to the image information based on the image sequences and motion estimation data; determining the heterogeneous octree mesh to which each point cloud coordinate belongs; determining an update strategy for the currently constructed octree model based on the current state of each mesh; and executing the update strategy to determine the 3D reconstruction model corresponding to the current underground coal mine space. This method can effectively adapt to the narrow passages in underground coal mines. By using a dynamically segmented octree, unnecessary spatial computation is avoided, thereby reducing the number of computational meshes required for spatial reconstruction of narrow passages and reducing the computational load. This can meet the strong demand for accurate and rapid spatial information perception during mining and tunneling processes.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method and apparatus for reconstructing underground space in coal mines by dynamically segmenting an octree. Background Technology

[0002] The digitalization, automation, and intelligentization of mining faces are key aspects of smart coal mine construction. This involves deeply integrating next-generation information technologies such as the Internet of Things, cloud computing, big data, 5G, industrial internet, and artificial intelligence with modern mining technologies. This enables seamless data flow across all elements, processes, and lifecycles of the mining face, forming a data-driven unit-level and system-level digital twin intelligent closed-loop system characterized by ubiquitous sensing, interconnectivity, autonomous learning, comprehensive analysis, optimized decision-making, and precise execution. Ultimately, this promotes intelligent operation of mining faces, and achieving unmanned mining is the ultimate goal pursued by the coal industry.

[0003] In related technologies, using a fully segmented octree to reconstruct underground coal mine spatial data results in a massive computational load and consumes a huge amount of memory. Furthermore, the actual size of the underground spatial model is large and elongated, meaning that most of the computation is not helpful in obtaining the model. Therefore, how to accurately and reliably obtain a three-dimensional spatial model of underground coal mines is a pressing problem that needs to be solved. Summary of the Invention

[0004] This disclosure aims to at least partially address one of the technical problems in the related art.

[0005] According to the first aspect of this disclosure, a method for reconstructing underground space in coal mines using a dynamically segmented octree is proposed, comprising:

[0006] Acquire a sequence of images of the underground space in a coal mine and motion estimation data of the image sequence;

[0007] Based on the image sequence and the motion estimation data, generate point cloud coordinates corresponding to the image information;

[0008] Determine the non-homogeneous octree mesh to which each of the point cloud coordinates belongs;

[0009] Based on the current state of each grid, determine the update strategy for the currently constructed octree model;

[0010] The update strategy is executed to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine.

[0011] According to a second aspect of this disclosure, a coal mine underground space reconstruction device based on dynamically segmented octrees is proposed, comprising:

[0012] The acquisition module is used to acquire a sequence of images of the underground space in a coal mine and motion estimation data of the image sequence;

[0013] The generation module is used to generate point cloud coordinates corresponding to the image information based on the image sequence and the motion estimation data;

[0014] The first determining module is used to determine the non-homogeneous octree grid to which each of the point cloud coordinates belongs;

[0015] The second determining module is used to determine the update strategy for the currently constructed octree model based on the current state of each of the grids.

[0016] The third determining module is used to execute the update strategy to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine.

[0017] The method and apparatus for reconstructing underground space in coal mines using dynamically segmented octrees disclosed herein have the following beneficial effects:

[0018] In this embodiment, firstly, an image sequence of the underground space in a coal mine and motion estimation data of the image sequence are acquired. Then, based on the image sequence and motion estimation data, point cloud coordinates corresponding to the image information are generated. Next, the heterogeneous octree mesh to which each point cloud coordinate belongs is determined. Then, based on the current state of each mesh, an update strategy for the currently constructed octree model is determined. Finally, the update strategy is executed to determine the 3D reconstruction model corresponding to the current underground space in the coal mine. Therefore, compared to constructing the full octree model, the computational load is reduced, which can well adapt to the narrow passages in underground coal mines. By using a dynamically segmented octree, unnecessary spatial calculations are avoided, thereby reducing the number of computational meshes required for spatial reconstruction of narrow passages and reducing the computational load. This can meet the strong demand for accurate and rapid spatial information perception during mining and tunneling processes.

[0019] Additional aspects and advantages of this disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this disclosure will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, in which:

[0021] Figure 1 This is a flowchart illustrating the method for reconstructing underground space in a coal mine using a dynamically segmented octree, as provided in the first embodiment of this disclosure.

[0022] Figure 2 This is the minimum mesh partitioning diagram of the octree model provided in the first embodiment of this disclosure;

[0023] Figure 3 This is a diagram of the octree data organization provided in the first embodiment of this disclosure;

[0024] Figure 4 This is a flowchart illustrating the method for reconstructing underground space in a coal mine using a dynamically segmented octree, as provided in the second embodiment of this disclosure.

[0025] Figure 5 This is a structural block diagram of the coal mine underground space reconstruction device with dynamic segmentation of an octree provided in the third embodiment of this disclosure. Detailed Implementation

[0026] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0027] The following describes, with reference to the accompanying drawings, a method and apparatus for reconstructing underground space in coal mines using dynamically segmented octrees, according to embodiments of the present disclosure.

[0028] Figure 1 This is a flowchart illustrating the method for reconstructing underground space in coal mines using a dynamically segmented octree, as provided in an embodiment of this disclosure.

[0029] It should be noted that the executing entity of the dynamic octree segmentation coal mine underground space reconstruction method in this embodiment is the dynamic octree segmentation coal mine underground space reconstruction device. The following description will use the dynamic octree segmentation coal mine underground space reconstruction device as the executing entity to explain the dynamic octree segmentation coal mine underground space reconstruction method proposed in this disclosure, without any limitation.

[0030] like Figure 1 As shown, the method for reconstructing underground space in coal mines using a dynamically segmented octree may include the following steps:

[0031] Step 101: Obtain the image sequence of the underground space in the coal mine and the motion estimation data of the image sequence.

[0032] It should be noted that underground coal mines are mostly narrow and long passages, with passage lengths ranging from hundreds to thousands of meters. The underground space is dark, and color information is relatively limited when used. In addition, the computing resources of edge devices in underground coal mines are limited, and storage and transmission capabilities are constrained.

[0033] In this disclosure, an octree based on dynamic segmentation can be used to avoid unnecessary spatial calculations, thereby reducing the number of computational grids required for spatial reconstruction of narrow passages and reducing computational load. If a fully segmented octree is used to reconstruct underground coal mine spatial data, it will result in a huge amount of computation and memory consumption. Since the actual size of the underground space model is large and it is narrow, many calculations are difficult to obtain the underground space model.

[0034] It should be noted that the image sequence is a sequence of images containing color and depth information of underground coal mines, and the motion estimation data is the motion data corresponding to the image sequence.

[0035] Specifically, an octree can be pre-built.

[0036] Optionally, an initial octree can be constructed based on the bounding box size of the first image sequence, wherein the initial octree is pre-allocated with memory.

[0037] Specifically, an initial octree can be constructed using twice the bounding box size of the first image sequence.

[0038] At this point, we can first determine the highest resolution of spatial reconstruction. In this disclosure, 1 cm can be used as the highest resolution, and no further limitation is made here.

[0039] It should be noted that if 1cm is taken as the highest resolution, that is, the grid size is no less than 8 times that, i.e., the minimum grid size is 8cm, such as Figure 2 As shown, this is the minimum mesh division. The minimum octree node size is 8cm. According to the highest resolution requirement, it is divided into 8x8x8 voxel units.

[0040] It should be noted that runtime memory allocation may fail, thus affecting runtime efficiency. Therefore, this disclosure adopts a pre-allocation strategy, which means that the required memory can be allocated in advance. The pre-allocated octree occupies 4GB of memory by default, which is used for the node characteristic queue, child node index queue, leaf node index queue, and data queue.

[0041] It should be noted that the size of the octree 3D spatial model is allocated in memory before runtime, which facilitates the direct rendering and visualization of the 3D spatial model. This can be set by modifying the pre-allocated memory parameters.

[0042] When allocating the initial memory usage of the four octree-organized arrays, extreme scenarios can be considered. For example, if the current node is the root node and the network node corresponding to the highest resolution, the memory usage of the four arrays—the node feature queue, the child node index queue, the leaf node index queue, and the data queue—can be 20:32:4:5644 = 5:8:1:1411. Another extreme case is when the current node is an empty node, i.e., there is no actual data. In this case, the actual memory consumption of the data queue can be 0. Considering the large variation in the actual memory usage of the data array, 37.5% of the pre-allocated memory can be allocated to this array, with a default setting of 1.5GB. Of the remaining 2.5GB, 900MB is for the node feature queue, 1.2GB is for the child node index queue, and 400MB is for the leaf node index queue.

[0043] It should be noted that when initially building the octree, after pre-allocating memory, all bits in that memory location can be set to 1 (INT_MAX for unsigned integers) to distinguish whether the memory area is already in use.

[0044] Specifically, an initial octree can be constructed based on the bounding box of the first image sequence. During construction, an octree size that can enclose the initial bounding box twice its size is selected to avoid frequent octree growth.

[0045] This invention uses arrays to store octrees. There are four array queues related to octrees, including a node characteristic queue, a child node index queue, a leaf node index queue, and an octree data queue.

[0046] like Figure 3 As shown, Figure 3 An octree data organization method is shown.

[0047] The node characteristic queue contains the node's ID. The high 24 bytes represent the node's unique identifier. Of the remaining 8 bytes, 4 bits are used to distinguish the node's relative position in the overall octree, divided into three main categories: internal nodes, edge nodes, and corner nodes. One bit is used each to indicate whether it is a leaf node or an empty node. The remaining two bits are used to indicate whether it exists during traversal. The node data index stores the position of the leaf node index queue. For intermediate nodes that are not leaf nodes or the root node, this value is set to -1. The parent node ID has the same composition as the node ID and is used to quickly traverse adjacent nodes during traversal. If the current node is the root node, the parent node ID is set to -1, and the parent node data index is set to -1. The first child node index is used to record the index value of the child node. Based on the record value in the child node index queue, the node characteristic queue is re-indexed to obtain the node ID and other information. The purpose of setting up the child node index queue is to facilitate the calculation and management of child node data and to facilitate parallel processing during multi-threaded calculations. The index value of the leaf node index queue can be obtained through the node data index. This value records the position of the leaf node data in the data queue. If the current node is not a leaf node, the leaf node index is -1. If the current node is a leaf node, but the leaf node has not initialized actual data, the value recorded in the leaf node index queue is set to -1. The leaf node index queue facilitates the quick acquisition of the total number of currently active leaf nodes and the quick acquisition of actual node data based on node characteristics.

[0048] In addition, to facilitate the representation of different scale information, the actual data queue of the leaf nodes, in addition to the Q-value (data queue, 4 bytes) coupled with the leaf node index queue, also includes the actual node position P (integer representation, 4 bytes) and scale S (integer representation, 4 bytes). The remaining actual node data consists of TSDF (Truncated Signed Distance Function, single-precision floating-point representation, 4 bytes), weights (characterizing the node sampling quality, default is 1.0, using single-precision floating-point data representation, 4 bytes), R (color R channel, 1 byte), G (color G channel, 1 byte), and B (color B channel, 1 byte).

[0049] Step 102: Based on the image sequence and motion estimation data, generate the point cloud coordinates corresponding to the image information.

[0050] The image information includes image depth and color information about the underground space of the coal mine in the image sequence.

[0051] In this disclosure, the device can process image sequences and motion estimation data, and then convert the image information into point cloud coordinates in a global coordinate system, which can determine the coordinates of each three-dimensional point cloud in the scene.

[0052] Step 103: Determine the non-homogeneous octree mesh to which each point cloud coordinate belongs.

[0053] It should be noted that after determining the point cloud coordinates corresponding to each image sequence, the point cloud coordinates can be divided to determine the non-homogeneous octree mesh corresponding to each point cloud coordinate.

[0054] Step 104: Determine the update strategy for the currently constructed octree model based on the current state of each grid.

[0055] It should be noted that there may be various situations with the mesh. For example, if the mesh exists and has been divided into the minimum mesh size required for the highest resolution, then the TSDF (Truncated Signed Distance Function) value and weight of the mesh are updated, and the triangular facet coordinates and colors of the voxel cells formed in the neighborhood of the mesh are also updated.

[0056] Alternatively, if a grid exists but is empty, and the data belongs to the grid, then the data is considered dynamic data or an outlier and discarded.

[0057] Alternatively, if the mesh does not exist, a tree growth algorithm should be used to dynamically expand the octree itself, and after expansion, the TSDF and weights of the mesh should be updated, and the triangular facets of the corresponding voxel unit should be updated accordingly.

[0058] Step 105: Execute the update strategy to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine.

[0059] Specifically, updates can be performed using the following voxel unit TSDF usage formula:

[0060] w(p) = 1.0

[0061] tsdf(p)=max(-μ,min(μ,D(p)-voxel.z))

[0062]

[0063] W(p) = W(p) + w(p)

[0064] In the formula, p represents the center coordinate of the voxel mesh (i.e., the mesh with the highest resolution required for the division value, 1 cm in size), D(p) is the depth value, voxel.z represents the z-coordinate of the voxel mesh, and μ is the adjustment parameter.

[0065] Furthermore, based on the calculated TSDF values, the TSDF and weights of each voxel mesh can be obtained, and the RGB values ​​of the voxel mesh can be calculated using a linear interpolation method. Then, based on the Marching Cubes algorithm, the positions of the triangular faces are calculated, and the colors of the three vertices are calculated using linear interpolation, forming a PLY file format. This PLY format model is used as the model representation for 3D spatial reconstruction.

[0066] Among them, the PLY format is a simple three-dimensional data representation format that can be easily updated and directly interfaced with OpenGL for rendering and visualization.

[0067] In this embodiment, firstly, an image sequence of the underground space in a coal mine and motion estimation data of the image sequence are acquired. Then, based on the image sequence and motion estimation data, point cloud coordinates corresponding to the image information are generated. Next, the grid to which each point cloud coordinate belongs is determined. Then, based on the current state of each grid, an update strategy for the currently constructed octree model is determined. Finally, the update strategy is executed to determine the 3D reconstruction model corresponding to the current underground space in the coal mine. Therefore, compared to constructing the full octree model, the computational load is reduced, which can well adapt to the narrow passages in coal mines. By using an octree based on dynamic segmentation, unnecessary spatial calculations are avoided, thereby reducing the number of computational grids required for spatial reconstruction of narrow passages and reducing the computational load. This can meet the strong demand for accurate and rapid spatial information perception during mining and tunneling processes.

[0068] Figure 4 This is a flowchart illustrating the method for reconstructing underground space in coal mines using a dynamically segmented octree, as provided in an embodiment of this disclosure.

[0069] like Figure 4 As shown, the method for reconstructing underground space in coal mines using a dynamically segmented octree may include the following steps:

[0070] Step 201: Obtain a sequence of images of the underground space in the coal mine and motion estimation data of the image sequence.

[0071] Step 202: Generate point cloud coordinates corresponding to the image information based on the image sequence and the motion estimation data.

[0072] Step 203: Determine the non-homogeneous octree mesh to which each of the point cloud coordinates belongs.

[0073] It should be noted that the specific implementation methods of steps 201, 202, and 203 can refer to the above embodiments, and are not limited here.

[0074] Step 204: When the mesh exists and has been divided into the highest resolution, update the TSDF value and weight of the mesh, and update the triangular facets of the voxel elements formed by the mesh neighborhood.

[0075] Wherein, TSDF is the cutoff distance function, and weights are used to characterize the sampling quality of nodes.

[0076] Optionally, the current bounding box cell mesh queue can be obtained first, then the TSDF values ​​and weights of all voxel meshes in the current bounding box cell mesh queue can be calculated, and finally all current bounding box cell meshes can be placed into the octree model and the mesh queue can be updated, thereby ending the fusion.

[0077] In this disclosure, eight times the mesh corresponding to the highest resolution is considered as the minimum node of the octree, and the Marching Cubes algorithm is applied to implicitly calculate the triangular facets in this minimum leaf node. Since the sampling distance between the visual device and the actual object varies during actual sampling, and considering the characteristics of visual devices, the sampling accuracy generally increases with the sampling distance, this disclosure designs a scale-based minimum mesh: for example, when the sampling distance is less than 3 meters, it is considered a high-precision sampling, and the minimum mesh size is twice the specified highest resolution, which defaults to 1 cm. If the sampling distance is greater than 3 meters, the minimum mesh size will be 16 cm. The 3-meter threshold is set primarily because most depth cameras use a depth range below 3 meters for standard operation and better accuracy. However, most visual devices allow for actual sampling depths greater than 3 meters, but accuracy gradually decreases at this depth; therefore, this threshold is used.

[0078] Step 205: If a grid exists and is an empty grid, discard the point cloud coordinates.

[0079] Step 206: If the grid does not exist, expand the preset octree, and after expansion, update the TSDF value and weight of the grid, and update the voxel units formed by the grid neighborhood.

[0080] It should be noted that if the grid does not exist, it means that the octree cannot encompass the spatial point cloud data in the global coordinate system converted from the existing image frame. In this case, the octree growth algorithm can be executed, that is, the calculation can be performed dynamically.

[0081] Optionally, when the octree cannot encompass the spatial point cloud data in the global coordinate system derived from the current image frame, the octree growth algorithm is executed.

[0082] It should be noted that during growth, the size of the octree root node required for growth can be calculated first. For example, the current bounding box size currentDim and the maximum size limit limitDi can be used as input to obtain the minimum bounding box size ret that can enclose the current bounding box.

[0083] After the size of the root node of the octree is calculated, the relative position of the original octree can be determined first, and then the growth direction of the new octree can be determined.

[0084] Optionally, the current bounding box cell mesh queue can be obtained first, then the TSDF values ​​and weights of all voxel meshes in the current bounding box cell mesh queue can be calculated, and finally all current bounding box cell meshes can be placed into the octree model and the mesh queue can be updated, thereby ending the fusion.

[0085] Step 207: Execute the update strategy to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine.

[0086] In this embodiment, firstly, an image sequence of the underground space in a coal mine and motion estimation data of the image sequence are acquired. Then, based on the image sequence and motion estimation data, point cloud coordinates corresponding to the image information are generated. Next, the heterogeneous octree mesh to which each point cloud coordinate belongs is determined. Then, when the mesh exists and has been divided into the highest resolution, the TSDF value and weight of the mesh are updated, and the triangular facets of the voxel units formed by the mesh's neighborhood are updated. Then, if the mesh exists but is empty, the point cloud coordinates are discarded. Then, if the mesh does not exist, a preset octree is expanded, and after expansion, the TSDF value and weight of the mesh are updated, and the voxel units formed by the mesh's neighborhood are updated. Finally, an update strategy is executed to determine the 3D reconstruction model corresponding to the current underground space in the coal mine. This satisfies the computational grid requirements for spatial reconstruction of narrow passages, meeting the strong demand for accurate and rapid spatial information perception during mining and tunneling. By continuously fusing and updating the data based on newly acquired depth data, the model can be updated in a timely manner based on new information, resulting in a more accurate and reliable reconstructed 3D model.

[0087] Figure 5 This is a schematic diagram of the structure of the coal mine underground space reconstruction device with dynamic segmentation of an octree provided in the third embodiment of this disclosure.

[0088] like Figure 5 As shown, the coal mine underground space reconstruction device 500 with dynamic segmentation of octree may include: acquisition module 510, generation module 520, first determination module 530, second determination module 540, and third determination module 550.

[0089] The acquisition module is used to acquire a sequence of images of the underground space in a coal mine and motion estimation data of the image sequence;

[0090] The generation module is used to generate point cloud coordinates corresponding to the image information based on the image sequence and the motion estimation data;

[0091] The first determining module is used to determine the non-homogeneous octree grid to which each of the point cloud coordinates belongs;

[0092] The second determining module is used to determine the update strategy for the currently constructed octree model based on the current state of each of the grids.

[0093] The third determining module is used to execute the update strategy to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine.

[0094] Optionally, the second determining module includes:

[0095] The first update unit is used to update the TSDF value and weight of the mesh when the mesh exists and has been divided into the highest resolution, and to update the triangular facets of the voxel units formed by the mesh neighborhood.

[0096] A discard unit is used to discard the point cloud coordinates when the grid exists and is an empty grid.

[0097] The second update unit is used to expand a preset octree when the grid does not exist, and after expansion, update the TSDF value and weight of the grid, and update the voxel units formed by the grid neighborhood.

[0098] Optionally, the first update unit is specifically used for:

[0099] Get the current bounding box cell grid queue;

[0100] Calculate the TSDF values ​​and weights of all voxel meshes in the current bounding box cell mesh queue;

[0101] Update the bounding box cell grid queue.

[0102] Optionally, the acquisition module is further configured to:

[0103] An initial octree is constructed based on the bounding box size of the first image sequence, and the initial octree is pre-allocated with memory.

[0104] Optionally, the second update unit is further configured to:

[0105] When the octree cannot encompass the spatial point cloud data in the global coordinate system derived from the current image frame, the octree growth algorithm is executed.

[0106] In this embodiment, firstly, an image sequence of the underground space in a coal mine and motion estimation data of the image sequence are acquired. Then, based on the image sequence and motion estimation data, point cloud coordinates corresponding to the image information are generated. Next, the grid to which each point cloud coordinate belongs is determined. Then, based on the current state of each grid, an update strategy for the currently constructed octree model is determined. Finally, the update strategy is executed to determine the 3D reconstruction model corresponding to the current underground space in the coal mine. Therefore, compared to constructing the full octree model, the computational load is reduced, which can well adapt to the narrow passages in coal mines. By using an octree based on dynamic segmentation, unnecessary spatial calculations are avoided, thereby reducing the number of computational grids required for spatial reconstruction of narrow passages and reducing the computational load. This can meet the strong demand for accurate and rapid spatial information perception during mining and tunneling processes.

[0107] It should be understood that various parts of this disclosure can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0108] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0109] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0110] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A method for reconstructing underground space in coal mines using a dynamically segmented octree, characterized in that, include: Acquire a sequence of images of the underground space in a coal mine and motion estimation data of the image sequence; An initial octree is constructed based on the bounding box size of the first image sequence, and the initial octree is pre-allocated with memory. Based on the image sequence and the motion estimation data, generate point cloud coordinates corresponding to the image information; Determine the non-homogeneous octree mesh to which each of the point cloud coordinates belongs; Based on the current state of each grid, determine the update strategy for the currently constructed octree model; The update strategy is executed to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine; The step of determining the update strategy for the currently constructed octree model based on the current state of each of the grids includes: When the grid exists and has been divided into the highest resolution, update the TSDF value and weight of the grid, and update the triangular facets of the voxel units formed by the grid neighborhood. If the grid exists and is an empty grid, the point cloud coordinates are discarded; If the grid does not exist, expand the preset octree, and after expansion, update the TSDF value and weight of the grid, and update the voxel units formed by the grid neighborhood.

2. The method according to claim 1, characterized in that, The updating of the TSDF value and weight of the mesh includes: Get the current bounding box cell grid queue; Calculate the TSDF values ​​and weights of all voxel meshes in the current bounding box cell mesh queue; Update the bounding box cell grid queue.

3. The method according to claim 1, characterized in that, The expanded preset octree includes: When the octree cannot encompass the spatial point cloud data in the global coordinate system derived from the current image frame, the octree growth algorithm is executed.

4. A coal mine underground space reconstruction device for dynamically segmenting an octree, characterized in that, include: The acquisition module is used to acquire a sequence of images of underground space in a coal mine and motion estimation data of the image sequence. Based on the bounding box size of the first image sequence, an initial octree is constructed, and the initial octree is pre-allocated with memory. The generation module is used to generate point cloud coordinates corresponding to the image information based on the image sequence and the motion estimation data; The first determining module is used to determine the non-homogeneous octree grid to which each of the point cloud coordinates belongs; The second determining module is used to determine the update strategy for the currently constructed octree model based on the current state of each of the grids. The third determining module is used to execute the update strategy to determine the three-dimensional reconstruction model corresponding to the current underground space of the coal mine. The second determining module includes: The first update unit is used to update the TSDF value and weight of the mesh when the mesh exists and has been divided into the highest resolution, and to update the triangular facets of the voxel units formed by the mesh neighborhood. A discard unit is used to discard the point cloud coordinates when the grid exists and is an empty grid. The second update unit is used to expand a preset octree when the grid does not exist, and after expansion, update the TSDF value and weight of the grid, and update the voxel units formed by the grid neighborhood.

5. The apparatus according to claim 4, characterized in that, The first update unit is specifically used for: Get the current bounding box cell grid queue; Calculate the TSDF values ​​and weights of all voxel meshes in the current bounding box cell mesh queue; Update the bounding box cell grid queue.

6. The apparatus according to claim 4, characterized in that, The second update unit is further configured to: When the octree cannot encompass the spatial point cloud data in the global coordinate system derived from the current image frame, the octree growth algorithm is executed.

Images