Thick vegetation coverage area point cloud filtering method and system based on implicit neural representation

Abstract

The invention discloses a thick vegetation coverage area point cloud filtering method and system based on implicit neural representation, and belongs to the technical field of point cloud filtering, and the method comprises the steps: obtaining point cloud data, building a self-adaptive irregular grid according to the point cloud data, segmenting the point cloud data based on the self-adaptive irregular grid, and obtaining the point cloud data of a plurality of sub-regions; inputting the point cloud data of each sub-region into an up-sampling framework based on an implicit neural network to obtain global point cloud data; and carrying out filtering processing on the global point cloud data based on a multi-scale feature point cloud filtering algorithm, and outputting a global point cloud ground point set. By setting an adaptive irregular grid, the density of the grid is automatically adjusted, so that the fine segmentation of the point cloud data of the thick vegetation coverage area is realized; an upsampling framework based on an implicit neural network is adopted to solve the sparsity of the point cloud; a multi-scale feature point cloud filtering algorithm is used for filtering global point cloud data, and the filtering effect on ground objects and vegetation at all levels is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud filtering, and particularly relates to a point cloud filtering method and system for thick vegetation-covered areas based on implicit neural representation. Background Art

[0002] Point cloud data, as a three-dimensional discrete point set directly derived from real-world scenes, aims to accurately capture and reconstruct the three-dimensional shape of objects. However, due to the disorderliness and other characteristics of point cloud data, significant challenges are brought to the subsequent construction of digital elevation models (DEMs) and three-dimensional reconstruction processes. In the data processing of original point clouds, a large number of scholars have conducted in-depth explorations on point cloud upsampling and filtering methods. How to generate non-abnormal and dense and uniform point clouds from sparse point clouds through sampling and filtering has become the research focus in recent years. In recent years, various three-dimensional point cloud processing devices have developed rapidly. Although good results have been achieved in sampling and filtering effects, in terms of accuracy, the point clouds obtained by sampling are relatively sparse, and the filtering effect in complex scene environments is not very satisfactory.

[0003] Traditional upsampling methods include interpolation upsampling and end-to-end upsampling. Among them, interpolation upsampling includes the nearest neighbor interpolation method, bilinear interpolation method, and bicubic interpolation method. End-to-end upsampling includes self-supervised point cloud upsampling and arbitrary-scale upsampling, but there are also corresponding problems. The nearest neighbor interpolation method has a fast calculation speed, but the image edges will be serrated; the bilinear interpolation method is smoother than the nearest neighbor interpolation method, but the calculation amount is larger; the bicubic interpolation method obtains a smoother and more delicate image, but the calculation amount is also larger; while self-supervised point cloud upsampling and arbitrary-scale upsampling are still limited to a fixed upsampling factor.

[0004] Commonly used filtering methods include difference-based filtering methods, slope filtering algorithms, and segmentation-based filtering algorithms. However, the above methods have good filtering effects in specific set scenarios, but they cannot handle point cloud data in complex scenarios well.

[0005] Therefore, how to provide an effective technical solution to solve the problems that the point clouds obtained by sampling are relatively sparse and the filtering effect in complex scene environments is not very satisfactory has become an urgent problem to be solved in the prior art. Summary of the Invention

[0006] The purpose of the present invention is to provide a point cloud filtering method and system for thick vegetation-covered areas based on implicit neural representation to solve the above problems existing in the prior art.

[0007] To achieve the above purpose, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a point cloud filtering method for thick vegetation-covered areas based on implicit neural representations, including: Obtain point cloud data, establish an adaptive irregular grid based on the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; Input the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; Perform filtering processing on the global point cloud data based on a multi-scale feature point cloud filtering algorithm, and output a global point cloud ground point set.

[0008] In a possible design, obtaining point cloud data, establishing an adaptive irregular grid based on the point cloud data, and segmenting the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions includes: Obtain point cloud data, traverse the point cloud data, and obtain the spatial position and point cloud density of the point cloud data; Construct an adaptive irregular grid based on the point cloud density, where the adaptive irregular grid based on the point cloud density includes multiple sub-region grids; Segment the point cloud data using the adaptive irregular grid based on the point cloud density, and assign the point cloud data to the corresponding sub-region grids according to the spatial position of the point cloud data to obtain point cloud data of multiple sub-regions.

[0009] In a possible design, after obtaining the point cloud data of multiple sub-regions, the following steps are further included: Step 1: Traverse the point cloud data of each sub-region to obtain the distribution of the point cloud data of each sub-region. The distribution includes the point cloud data within each sub-region and the boundary point cloud data on the sub-region boundary. Filter out the outliers in the boundary point cloud data on the sub-region boundary according to the normal vector consistency to obtain the boundary point cloud; Step 2: Establish a secondary subdivision grid, divide the sub-regions with boundary point clouds according to the secondary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent secondary subdivision grid, perform inverse distance weighting on the adjacent secondary subdivision grids according to the distance, and the adjacent secondary subdivision grids vote on each boundary point cloud according to the weight to obtain a first weight voting result. Assign each boundary point cloud to the secondary subdivision grid with the highest first weight voting result; Step 3: Traverse the point cloud data in the secondary subdivision grid again. If there are boundary point clouds in the secondary subdivision grid, establish a tertiary subdivision grid. Divide the secondary subdivision grid with boundary point clouds according to the tertiary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent tertiary subdivision grid, perform inverse distance weighting on the adjacent tertiary subdivision grids according to the distance, and the adjacent tertiary subdivision grids vote on each boundary point cloud according to the weights to obtain the second weight voting result. Assign each boundary point cloud to the tertiary subdivision grid with the highest second weight voting result; Repeat the above steps until there are no boundary point clouds on the grid boundary.

[0010] In a possible design, the upsampling framework based on the implicit neural network includes an encoder and a decoder. The encoder includes a position encoder, and the decoder includes a feature extractor and an upsampling module; the point cloud data of each sub-region includes point cloud coordinate data; input the point cloud data of each sub-region into the upsampling framework based on the implicit neural network to obtain global point cloud data, including: Input the point cloud data of each sub-region into the position encoder to preprocess the point cloud data of each sub-region to obtain encoded information of each sub-region, where the encoded information of each sub-region includes position attributes and color attributes; use the implicit neural function to map the point cloud coordinate data to the corresponding position attributes in the high-dimensional space, and output the high-dimensional feature vectors of each sub-region; Use the feature extractor to learn and extract the high-dimensional feature vectors of each sub-region to obtain the terrain geometric features of each sub-region, and input the terrain geometric features of each sub-region into the upsampling module; In the upsampling module, predict the local features of the missing hole part in the point cloud data of each sub-region based on the terrain geometric features of each sub-region to obtain a prediction result. Based on the prediction result, complete the missing hole part of the point cloud to obtain the point cloud of each sub-region after completion. Based on the point cloud of each sub-region after completion, adopt the principle of stepped sampling number, set different upsampling ratios for different sub-region point cloud densities, and perform upsampling to generate a high-resolution and density-uniform point cloud; Perform point cloud stitching based on the high-resolution and density-uniform point cloud to obtain global point cloud data.

[0011] In a possible design, predicting the local features of the missing hole part in the point cloud data of each sub-region based on the terrain geometric features of each sub-region to obtain a prediction result, and completing the missing hole part of the point cloud based on the prediction result to obtain the point cloud of each sub-region after completion, includes: Based on the point cloud coordinate data, use the implicit neural function to predict the density distribution of the missing hole part in the point cloud data of each sub-region, and generate new point clouds in the missing hole part based on a preset density threshold; For the position offset of the new point cloud computing, the new point cloud is corrected based on the position offset to obtain the point cloud of each sub-region after completion.

[0012] In a possible design, based on the point cloud coordinate data, the calculation expression for predicting the density distribution of the missing hole part in the point cloud data of each sub-region using the implicit neural function is: ; In the formula, is the point cloud coordinate data, is the position of the point in the point cloud in the x-axis direction, is the position of the point in the point cloud in the y-axis direction, is the position of the point in the point cloud in the z-axis direction, is the implicit function, is the density of the point cloud data of each sub-region, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the x-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the y-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the z-axis direction, is the predicted density distribution value of the missing hole part in the point cloud data of each sub-region, is the activation function.

[0013] In a possible design, the global point cloud data includes a ground point set; the global point cloud data is filtered based on a multi-scale multi-feature point cloud filtering algorithm, and the global point cloud ground point set is output, including: The spherical neighborhood method is used to perform multi-feature extraction on the global point cloud data to obtain multi-feature data, where the multi-feature data includes color information, roughness, and covariance-based features, and the covariance-based features include linearity, perpendicularity, flatness, and total variance features; The global point cloud data and the multi-feature data are feature-fused using a multi-feature fusion method based on a spherical neighborhood to obtain a fused feature point set; The large-scale grid is used to obtain the elevation information in the ground point set, and the small-scale grid is used to correct the elevation information to obtain the corrected elevation information. The fused feature point set is corrected according to the corrected elevation information to obtain the corrected fused feature point set; Generate a low - resolution DSM model and a high - resolution DSM model based on the corrected fused feature point set, construct a difference model according to the low - resolution DSM model and the high - resolution DSM model, calculate the elevation change degree between the low - resolution DSM model and the high - resolution DSM model according to the difference model, and calculate the average value and standard deviation of the elevation change degree. Calculate the elevation variation coefficient of the DSM model according to the average value and standard deviation of the elevation change degree, and determine the height difference threshold based on the elevation variation coefficient of the DSM model; Filter out the feature points of the global point cloud data according to the height difference threshold, and output the global point cloud ground point set.

[0014] In a possible design, the calculation expressions of the elevation change degree, elevation variation coefficient, and height difference threshold between the low - resolution DSM model and the high - resolution DSM model are as follows: ; In the formula, is the elevation change degree between the low - resolution DSM model and the high - resolution DSM model, is the high - resolution DSM model, is the low - resolution DSM model, is the elevation variation coefficient of the DSM model, is the standard deviation obtained according to the elevation change degree, is the average value obtained according to the elevation change degree, is the height difference threshold, and are empirical coefficients, The default value of is 0.5, and the default value of

[0015] In the second aspect, the present invention provides a point cloud filtering system for thick vegetation - covered areas based on implicit neural representation, including: An acquisition and segmentation module, configured to acquire point cloud data, establish an adaptive irregular grid according to the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub - regions; An implicit neural up - sampling module, configured to input the point cloud data of each sub - region into an up - sampling framework based on an implicit neural network to obtain global point cloud data; A multi - scale feature point cloud filtering module, configured to perform filtering processing on the global point cloud data based on a multi - scale feature point cloud filtering algorithm, and output a global point cloud ground point set.

[0016] In a third aspect, the present invention provides a computer program product comprising instructions that, when executed on a computer, cause the computer to perform the method for filtering point clouds in a thick vegetation-covered area based on implicit neural representations as described in the first aspect above.

[0017] The beneficial effects of the present invention are as follows: The present invention discloses a method and system for filtering point clouds in a thick vegetation-covered area based on implicit neural representations, including obtaining point cloud data, establishing an adaptive irregular grid based on the point cloud data, segmenting the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions, inputting the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; filtering the global point cloud data using a multi-scale feature point cloud filtering algorithm and outputting a global point cloud ground point set. The present invention sets an adaptive irregular grid, which can process the point cloud data in a thick vegetation-covered area and automatically adjust the density of the grid according to the characteristics of the scale and density of the input point cloud data, thereby realizing fine segmentation of the point cloud data in a thick vegetation-covered area; adopting an upsampling framework based on an implicit neural network can reshape the ground objects of the point cloud and complete the data of the missing parts, and adopting the method of differential upsampling to solve the sparsity of the point cloud; using a multi-scale feature point cloud filtering algorithm to filter the global point cloud data to improve the filtering effect on ground objects and vegetation at all levels. Description of the Drawings

[0018] Figure 1 It is a flowchart of a method for filtering point clouds in a thick vegetation-covered area based on implicit neural representations provided in the first aspect of this embodiment; Figure 2 It is a block diagram of a system for filtering point clouds in a thick vegetation-covered area based on implicit neural representations provided in the second aspect of this embodiment. Detailed Embodiments

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the present invention in combination with the drawings and the descriptions of the embodiments or the prior art. Obviously, the following descriptions of the structures of the drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. It should be noted here that the descriptions of these embodiment modes are used to help understand the present invention, but do not constitute a limitation to the present invention.

[0020] It should be understood that although terms such as first and second may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object may be referred to as the second object, and similarly, the second object may be referred to as the first object, without departing from the scope of the exemplary embodiments of the present invention.

[0021] Embodiment: As Figure 1 shown, in the first aspect of this embodiment, a point cloud filtering method for thick vegetation-covered areas based on implicit neural representation is provided, which can but is not limited to being executed by a computer device or virtual machine with certain computing resources, such as an electronic device like a personal computer or a smartphone, or a virtual machine; as Figure 1 shown, the point cloud filtering method for thick vegetation-covered areas based on implicit neural representation may but is not limited to include the following steps: S1. Obtain point cloud data, establish an adaptive irregular grid based on the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; Specifically, the point cloud data can be obtained by airborne LiDAR (Light Detection and Ranging), which is the abbreviation of a laser detection and ranging system, and is a space measurement system integrating modern laser scanning technology, computer technology, high-dynamic carrier attitude measurement technology, and high-precision dynamic GPS (Global Positioning System) differential positioning technology.

[0022] Specifically, in step S1, obtaining point cloud data, establishing an adaptive irregular grid based on the point cloud data, and segmenting the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions includes: S101. Obtain point cloud data, traverse the point cloud data, and obtain the spatial position and point cloud density of the point cloud data; S102. Construct an adaptive irregular grid based on the point cloud density, where the adaptive irregular grid based on the point cloud density includes multiple sub-region grids; S103. Segment the point cloud data using the adaptive irregular grid based on the point cloud density, and assign the point cloud data to the corresponding sub-region grids according to the spatial position of the point cloud data to obtain point cloud data of multiple sub-regions.

[0023] In this embodiment, a dictionary is also established to collect information on sub-region point cloud data.

[0024] Further, after obtaining the point cloud data of multiple sub-regions, the following steps are also included: S104. Traverse the point cloud data of each sub-region to obtain the distribution of the point cloud data of each sub-region. The distribution includes the point cloud data within each sub-region and the boundary point cloud data on the boundary of the sub-region. Filter out the outlier points in the boundary point cloud data on the boundary of the sub-region according to the normal vector consistency to obtain the boundary point cloud; Specifically, the role of normal vector consistency is to identify and filter out noise points or outlier points. In three-dimensional space, a normal vector can be calculated for each point cloud, and this normal vector is perpendicular to the surface where the point is located. The basic idea is that for a region with a smooth surface, the normal vectors of adjacent points should be similar; if the normal vector of a certain point is very different from the normal vectors of its adjacent points, then this point may not be on the same surface and can be determined as an outlier point or noise. Therefore, the outlier points are filtered out to improve the accuracy of the data.

[0025] S105. Establish a secondary subdivision grid, divide the sub-regions with boundary point clouds according to the secondary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent secondary subdivision grid, perform inverse distance weighting on the adjacent secondary subdivision grids according to the distance, and the adjacent secondary subdivision grids make a voting decision on each boundary point cloud according to the weight to obtain the first weight voting result. Assign each boundary point cloud to the secondary subdivision grid with the highest first weight voting result; Specifically, inverse distance weighting reflects the inverse proportional relationship between the weight and the distance; the voting decision is a voting method that takes into account the weight, which is equivalent to the weighted average method. Multiply the number of votes obtained after voting by the weight to get the weighted number of votes, and then sum up all the weighted number of votes obtained by each boundary point cloud to get the first weight voting result.

[0026] S106. Traverse the point cloud data in the secondary subdivision grid again. If there are boundary point clouds in the secondary subdivision grid, establish a tertiary subdivision grid, divide the secondary subdivision grids with boundary point clouds according to the tertiary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent tertiary subdivision grid, perform inverse distance weighting on the adjacent tertiary subdivision grids according to the distance, and the adjacent tertiary subdivision grids make a voting decision on each boundary point cloud according to the weight to obtain the second weight voting result. Assign each boundary point cloud to the tertiary subdivision grid with the highest second weight voting result; Specifically, both the first weight voting result and the second weight voting result are voting scores, and each boundary point cloud is assigned to the subdivision grid with the highest voting score.

[0027] S107. Repeat steps S104 - S106 until there are no boundary point clouds on the grid boundary.

[0028] Specifically, when performing grid division, set the maximum recursion depth. When performing recursive division, the level of the divided grid is less than or equal to the maximum recursion depth.

[0029] Furthermore, only the case where the boundary point cloud is in two adjacent grid regions is considered in steps S105 to S107. In this embodiment, preferably, when the boundary point cloud is in the grid region at the intersection of three or four neighbors, for the attribution of the boundary point cloud, it includes: calculating the distance from the boundary point cloud to the centroid of the adjacent grid, and performing inverse distance weighting on the distance to obtain a weight value, obtaining a multi-dimensional weight based on the weight value and the feature similarity, normalizing the multi-dimensional weight, and selecting the corresponding grid with the largest multi-dimensional weight value as the attribution according to a preset determination threshold.

[0030] S2. Input the point cloud data of each sub-region into the upsampling framework based on an implicit neural network to obtain the global point cloud data; Furthermore, the upsampling framework based on an implicit neural network includes an encoder, a decoder, an MLP (Multi-Layer Perceptron) layer, and an output layer. Among them, the encoder includes a position encoder, and the decoder includes a feature extractor based on Transformer and an upsampling module; the point cloud data of each sub-region includes point cloud coordinate data.

[0031] Specifically, in step S2, inputting the point cloud data of each sub-region into the upsampling framework based on an implicit neural network to obtain the global point cloud data includes: S201. Input the point cloud data of each sub-region into the position encoder to preprocess the point cloud data of each sub-region to obtain the encoding information of each sub-region. Among them, the encoding information of each sub-region includes position attributes; use the implicit neural function to map the point cloud coordinate data to the corresponding position attributes in the high-dimensional space, and output the high-dimensional feature vectors of each sub-region; In this embodiment, the implicit neural function, as a continuous representation of terrain geometry, can capture complex spatial relationships in sparse point clouds and output the high-dimensional feature vectors of each sub-region, enhancing the interpretability of the data and the accuracy and efficiency of analysis; the encoding information of each sub-region includes, but is not limited to, position attributes and color attributes.

[0032] S202. Use the feature extractor to learn and extract the high-dimensional feature vectors of each sub-region to obtain the terrain geometry features of each sub-region, and input the terrain geometry features of each sub-region into the upsampling module; Furthermore, the feature extractor is constructed based on Transformer.

[0033] S203. In the upsampling module, based on the topographic geometric features of each sub-region, predict the local features of the missing hole parts in the point cloud data of each sub-region to obtain a prediction result. Based on the prediction result, complete the missing hole parts of the point cloud to obtain the point clouds of each sub-region after completion. Based on the point clouds of each sub-region after completion, adopt the principle of stepped sampling numbers, and set different upsampling magnification factors for different sub-region point cloud densities and point cloud scales to perform upsampling and generate high-resolution and uniformly dense point clouds; In this embodiment, feature prediction is performed on the missing part through topographic geometric features, and the missing part is completed based on the prediction result, avoiding the loss of features at a single scale.

[0034] Specifically, in step S203, based on the topographic geometric features of each sub-region, predict the local features of the missing hole parts in the point cloud data of each sub-region to obtain a prediction result. Based on the prediction result, complete the missing hole parts of the point cloud to obtain the point clouds of each sub-region after completion, including: S2031. Based on the point cloud coordinate data, use an implicit neural function to predict the density distribution of the missing hole parts in the point cloud data of each sub-region, and generate new point clouds in the missing hole parts based on a preset density threshold; Furthermore, the calculation expression for using an implicit neural function to predict the density distribution of the missing hole parts in the point cloud data of each sub-region based on the point cloud coordinate data is: ; In the formula, is the point cloud coordinate data, is the position of the point in the point cloud in the x-axis direction, is the position of the point in the point cloud in the y-axis direction, is the position of the point in the point cloud in the z-axis direction, is the implicit function, is the density of the point cloud data of each sub-region, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the x-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the y-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the z-axis direction, is the predicted density distribution value of the missing hole part in the point cloud data of each sub-region, is the activation function.

[0035] S2032. Calculate the position offset for the new point cloud, and correct the new point cloud based on the position offset to obtain the point clouds of each sub-region after completion.

[0036] S204. Stitch the point clouds based on the high-resolution and density-uniform point clouds to obtain the global point cloud data.

[0037] Specifically, the global point cloud data includes all three-dimensional information such as ground point sets, objects, buildings, vegetation, terrain, etc.

[0038] Furthermore, after the upsampling of the point clouds is completed, construct a K-d tree, establish an upsampling quality evaluation mechanism, and compare the global point cloud data with the initially obtained point cloud data. Among them, the upsampling quality evaluation mechanism can select methods such as point cloud density evaluation and DTM model visual discrimination evaluation. Point cloud density evaluation refers to the quantitative analysis of the distribution density of points in the point cloud data; DTM model visual discrimination evaluation refers to evaluating the quality and accuracy of the DTM through visual inspection. The above evaluation mechanisms are all existing technologies and will not be elaborated here.

[0039] S3. Filter the global point cloud data based on the multi-scale and multi-feature point cloud filtering algorithm, and output the global point cloud ground point set.

[0040] Specifically, in step S3, filter the global point cloud data based on the multi-scale and multi-feature point cloud filtering algorithm, and output the global point cloud ground point set, including: S301. Use the spherical neighborhood method to perform multi-feature extraction on the global point cloud data to obtain multi-feature data. Among them, the multi-feature data includes color information, roughness, and covariance-based features. The covariance-based features include linearity, perpendicularity, flatness, and total variance features; Specifically, the spherical neighborhood method is a method for defining neighborhoods in a metric space, mainly used to analyze the relationship between points in space and their surrounding points.

[0041] Furthermore, add features such as perpendicularity and flatness in the building feature extraction to improve the extraction integrity; add features such as roughness and total variance in the process of classifying vegetation at all levels in complex mountains to improve the classification accuracy.

[0042] S302. Use the multi-feature fusion method based on the spherical neighborhood to fuse the global point cloud data and the multi-feature data to obtain the fused feature point set; S303. Use the large-scale grid to obtain the elevation information in the ground point set, and use the small-scale grid to correct the elevation information to obtain the corrected elevation information. Correct the fused feature point set according to the corrected elevation information to obtain the corrected fused feature point set; Specifically, after obtaining the fused point set, the fused point set is input into two-dimensional and three-dimensional grids for presentation in various forms. Then, the elevation information in the ground point set is obtained using large-scale grids, and the elevation information is corrected using small-scale grids to improve the authenticity and reliability of obtaining local details.

[0043] S304. Generate a low-resolution DSM model (Digital Surface Model) and a high-resolution DSM model based on the corrected fused feature point set. Construct a difference model according to the low-resolution DSM model and the high-resolution DSM model. Calculate the elevation change degree between the low-resolution DSM model and the high-resolution DSM model based on the difference model, and calculate the average value and standard deviation of the elevation change degree. Calculate the elevation variation coefficient of the DSM model based on the average value and standard deviation of the elevation change degree, and determine the height difference threshold based on the elevation variation coefficient of the DSM model. Further, the calculation expressions for the elevation change degree, elevation variation coefficient, and height difference threshold between the low-resolution DSM model and the high-resolution DSM model are as follows: ; In the formula, is the elevation change degree between the low-resolution DSM model and the high-resolution DSM model, is the high-resolution DSM model, is the low-resolution DSM model, is the elevation variation coefficient of the DSM model, is the standard deviation obtained according to the elevation change degree, is the average value obtained according to the elevation change degree, is the height difference threshold, and are empirical coefficients, The default value of is 0.5, and the default value of

[0044] S305. Filter out the feature points of the ground objects in the global point cloud data according to the height difference threshold, and output the global point cloud ground point set.

[0045] Further, after outputting the global point cloud ground point set, establish a filtering performance evaluation system to evaluate the effect of the filtering algorithm.

[0046] In this embodiment, preferably, the multi-scale feature point cloud filtering algorithm is selected for filtering processing, but other common filtering algorithms can also be used. Its application scenario will not be limited to the thick vegetation-covered area described in this article, but can be extended to more complex scenarios for use.

[0047] This embodiment discloses a point cloud filtering method for thick vegetation coverage areas based on implicit neural representations, including obtaining point cloud data, establishing an adaptive irregular grid according to the point cloud data, segmenting the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; inputting the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; filtering the global point cloud data based on a multi-scale feature point cloud filtering algorithm and outputting a global point cloud ground point set. In this embodiment, the point cloud data is segmented using an adaptive irregular grid, and the density of the point cloud data is dynamically adjusted, enabling the processing of point cloud data in various complex scenarios, thereby achieving fine segmentation of the point cloud data; in the upsampling framework based on an implicit neural network, an improved encoding-decoding result is used to extract ground object features and complete the point cloud data, and a preset upsampling coefficient is used to better solve the density difference between different blocks and solve the sparsity problem of the point cloud; at the same time, features such as verticality and flatness are added during the filtering process, improving the integrity of building extraction, and features such as roughness and total variance are added, enhancing the classification effect of vegetation at different levels in the mountains.

[0048] As Figure 2 shown, the second aspect of this embodiment provides a point cloud filtering system for thick vegetation coverage areas based on implicit neural representations, including: An acquisition and segmentation module, configured to obtain point cloud data, establish an adaptive irregular grid according to the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; An implicit neural upsampling module, configured to input the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; A multi-scale feature point cloud filtering module, configured to filter the global point cloud data based on a multi-scale feature point cloud filtering algorithm and output a global point cloud ground point set.

[0049] This embodiment discloses a point cloud filtering system for thick vegetation coverage areas based on implicit neural representation. The acquisition and segmentation module is used to obtain point cloud data, establish an adaptive irregular grid based on the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; the implicit neural upsampling module inputs the point cloud data of each sub-region into the upsampling framework based on the implicit neural network to obtain global point cloud data; the multi-scale feature point cloud filtering module uses the multi-scale feature point cloud filtering algorithm to filter the global point cloud data and outputs the global point cloud ground point set. The adaptive irregular grid can automatically adjust the density of the grid according to the characteristics of the scale and density of the input point cloud data, so as to achieve fine segmentation of the point cloud data in thick vegetation coverage areas; the multi-scale feature point cloud filtering algorithm is used to filter the global point cloud data to improve the filtering effect on ground objects and vegetation at all levels.

[0050] In the third aspect of this embodiment, a computer program product is provided, including a computer program or instruction, and when the computer program or the instruction is executed by the computer, it implements the method for filtering point clouds in thick vegetation coverage areas based on implicit neural representation as described in the first aspect.

[0051] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A point cloud filtering method for thick vegetation covered areas based on implicit neural representation, characterized in that, Including: Obtain point cloud data, establish an adaptive irregular grid based on the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; Input the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; Perform filtering processing on the global point cloud data based on a multi-scale feature point cloud filtering algorithm, and output a global point cloud ground point set.

2. The method for filtering point clouds in a thick vegetation-covered area based on implicit neural representation according to claim 1, wherein Obtain point cloud data, establish an adaptive irregular grid based on the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions, including: Obtain point cloud data, traverse the point cloud data, and obtain the spatial position and point cloud density of the point cloud data; Construct an adaptive irregular grid based on point cloud density, where the adaptive irregular grid based on point cloud density includes multiple sub-region grids; Segment the point cloud data using the adaptive irregular grid based on point cloud density, and assign the point cloud data to the corresponding sub-region grid according to the spatial position of the point cloud data to obtain point cloud data of multiple sub-regions.

3. A point cloud filtering method for thick vegetation-covered areas based on implicit neural representation according to claim 2, characterized in that, After obtaining the point cloud data of multiple sub-regions, the following steps are also included: Step 1, traverse the point cloud data of each sub-region to obtain the distribution of the point cloud data of each sub-region. The distribution includes the point cloud data in each sub-region and the boundary point cloud data on the sub-region boundary. Filter out the outlier points in the boundary point cloud data on the sub-region boundary according to the normal vector consistency to obtain the boundary point cloud; Step 2, establish a secondary subdivision grid, divide the sub-region with boundary point cloud according to the secondary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent secondary subdivision grid, perform inverse distance weighting on the adjacent secondary subdivision grids according to the distance, and the adjacent secondary subdivision grids vote on each boundary point cloud according to the weight to obtain the first weight voting result, and assign each boundary point cloud to the secondary subdivision grid with the highest first weight voting result; Step 3, traverse the point cloud data in the secondary subdivision grid again. If there is boundary point cloud in the secondary subdivision grid, establish a tertiary subdivision grid, divide the secondary subdivision grid with boundary point cloud according to the tertiary subdivision grid, calculate the distance from each boundary point cloud to the centroid of the adjacent tertiary subdivision grid, perform inverse distance weighting on the adjacent tertiary subdivision grids according to the distance, and the adjacent tertiary subdivision grids vote on each boundary point cloud according to the weight to obtain the second weight voting result, and assign each boundary point cloud to the tertiary subdivision grid with the highest second weight voting result; Repeat the above steps until there is no boundary point cloud on the grid boundary.

4. A point cloud filtering method for thick vegetation covered areas based on implicit neural representation according to claim 3, characterized in that, The upsampling framework based on an implicit neural network includes an encoder and a decoder. The encoder includes a position encoder, and the decoder includes a feature extractor and an upsampling module; the point cloud data of each sub-region includes point cloud coordinate data; Input the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data, including: Input the point cloud data of each sub-region into the position encoder, preprocess the point cloud data of each sub-region to obtain the encoding information of each sub-region, where the encoding information of each sub-region includes position attributes; use an implicit neural function to map the point cloud coordinate data to the corresponding position attributes in the high-dimensional space, and output the high-dimensional feature vectors of each sub-region; Use a feature extractor to learn and extract the high-dimensional feature vectors of each sub-region to obtain the terrain geometric features of each sub-region, and input the terrain geometric features of each sub-region into the upsampling module; In the upsampling module, based on the terrain geometric features of each sub-region, predict the local features of the missing hole parts in the point cloud data of each sub-region to obtain the prediction result, based on the prediction result, complete the missing hole parts of the point cloud, and obtain the point clouds of each sub-region after completion. Based on the point clouds of each sub-region after completion, adopt the principle of stepped sampling number, set different upsampling magnification ratios for different sub-region point cloud densities, and perform upsampling to generate high-resolution and density-uniform point clouds; Perform point cloud stitching based on the high-resolution and density-uniform point clouds to obtain the global point cloud data.

5. A method for filtering point clouds in thick vegetation-covered areas based on implicit neural representations according to claim 4, characterized in that, Based on the terrain geometric features of each sub-region, predict the local features of the missing hole parts in the point cloud data of each sub-region to obtain the prediction result, and based on the prediction result, complete the missing hole parts of the point cloud to obtain the point clouds of each sub-region after completion, including: Based on the point cloud coordinate data, use an implicit neural function to predict the density distribution of the missing hole parts in the point cloud data of each sub-region, and based on a preset density threshold, generate new point clouds in the missing hole parts; Calculate the position offset for the new point clouds, and correct the new point clouds based on the position offset to obtain the point clouds of each sub-region after completion.

6. A point cloud filtering method for thick vegetation covered areas based on implicit neural representation according to claim 5, characterized in that, The calculation expression for using an implicit neural function to predict the density distribution of the missing hole parts in the point cloud data of each sub-region based on the point cloud coordinate data is: ； Wherein, is the point cloud coordinate data, is the position of the point in the point cloud in the x-axis direction, is the position of the point in the point cloud in the y-axis direction, is the position of the point in the point cloud in the z-axis direction, is the implicit function, is the density of the point cloud data of each sub-region, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the x-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the y-axis direction, is the distance between the point cloud and the adjacent point cloud in the point cloud data of the missing hole part in each sub-region in the z-axis direction, is the density distribution value of the missing hole part in the predicted point cloud data of each sub-region, is the activation function.

7. A point cloud filtering method for thick vegetation covered areas based on implicit neural representation according to claim 1, characterized in that, The global point cloud data includes a ground point set; perform filtering processing on the global point cloud data based on a multi-scale and multi-feature point cloud filtering algorithm, and output the global point cloud ground point set, including: Adopt a spherical neighborhood method to perform multi-feature extraction on the global point cloud data to obtain multi-feature data, where the multi-feature data includes color information, roughness, and covariance-based features, and the covariance-based features include linearity, perpendicularity, flatness, and total variance features; Use a multi-feature fusion method based on a spherical neighborhood to fuse the global point cloud data and the multi-feature data to obtain a fused feature point set; Use a large-scale grid to obtain the elevation information in the ground point set, and use a small-scale grid to correct the elevation information to obtain the corrected elevation information. Correct the fused feature point set according to the corrected elevation information to obtain the corrected fused feature point set; Generate a low-resolution DSM model and a high-resolution DSM model based on the corrected fused feature point set, construct a difference model according to the low-resolution DSM model and the high-resolution DSM model, calculate the elevation change degree between the low-resolution DSM model and the high-resolution DSM model according to the difference model, and calculate the average value and standard deviation of the elevation change degree. Calculate the elevation variation coefficient of the DSM model according to the average value and standard deviation of the elevation change degree, and determine the height difference threshold based on the elevation variation coefficient of the DSM model; Filter out the feature points of the global point cloud data according to the height difference threshold, and output the global point cloud ground point set.

8. A method for filtering point clouds in a thick vegetation-covered area based on implicit neural representation according to claim 7, characterized in that, The calculation expressions of the elevation change degree, elevation variation coefficient, and height difference threshold between the low-resolution DSM model and the high-resolution DSM model are as follows: ； In the formula, is the elevation change degree between the low-resolution DSM model and the high-resolution DSM model, is the high-resolution DSM model, is the low-resolution DSM model, is the elevation variation coefficient of the DSM model, is the standard deviation value obtained according to the elevation change degree, is the average value obtained according to the elevation change degree, is the elevation difference threshold, and are empirical coefficients, The default value of is 0.5, and the default value of 9. A point cloud filtering system for thick vegetation covered areas based on implicit neural representation, characterized in that, Include: An acquisition segmentation module, configured to acquire point cloud data, establish an adaptive irregular grid according to the point cloud data, and segment the point cloud data based on the adaptive irregular grid to obtain point cloud data of multiple sub-regions; An implicit neural upsampling module, configured to input the point cloud data of each sub-region into an upsampling framework based on an implicit neural network to obtain global point cloud data; A multi-scale feature point cloud filtering module, configured to perform filtering processing on the global point cloud data based on a multi-scale feature point cloud filtering algorithm, and output a global point cloud ground point set.

10. A computer program product comprising a computer program or instructions, characterized in that, The computer program or the instruction, when executed by a computer, implements the point cloud filtering method for thick vegetation coverage areas based on implicit neural representation according to any one of claims 1 to 8.

Images