Ground laser radar point cloud normalization processing method applied to moso bamboo forest after snow disaster

Abstract

The invention relates to the technical field of ground laser radar point cloud data processing, in particular to a ground laser radar point cloud normalization processing method applied to a moso bamboo forest after a snow disaster. The method mainly solves the technical problems that when moso bamboo points with form variation such as bending, breaking and lodging caused by snow disasters are processed through an existing normalization technology, due to incomplete ground point extraction and bamboo pole form mutation, seriously discontinuous points such as sawteeth and faults occur in normalized point clouds. The method comprises the following steps: firstly, calculating an initial relative elevation by constructing an elevation reference model; carrying out local elevation correction on the point cloud by using an overlapped window along the elevation direction by adopting a layered sliding window median smoothing technology; for missing points which are not covered by the window, carrying out bottom-carrying processing through global K-nearest neighbor median smoothing; and finally, a result is fused based on the smoothing times, translation return-to-zero is carried out, and normalized point cloud is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ground laser radar point cloud data processing, and particularly relates to a ground laser radar point cloud normalization processing method applied to a bamboo forest after a snow disaster. BACKGROUND

[0002] The ground laser radar technology has become an important means in the forestry resource investigation, and the ground laser radar technology obtains high-precision three-dimensional point cloud data to provide a basis for forest parameter inversion and form analysis. In the point cloud data processing flow, point cloud normalization is a key preprocessing step, and the point cloud normalization aims to eliminate the influence of terrain undulations on the height information and convert the point cloud height value to a relative height system with the ground as a reference. The existing technology usually generates a digital elevation model by using an algorithm based on an irregular triangle net or cloth simulation filtering, and then realizes point cloud height normalization.

[0003] These traditional normalization methods have significant limitations in processing disaster forest scenes. In particular, after a snow disaster, the bamboo forest will have complex morphological variations such as large-scale bamboo bending, breaking, lodging and uprooting, which will cause the structure of the undergrowth terrain to be extremely complex. In such a scene, the traditional method often causes the extraction of the ground points to be incomplete due to the vegetation coverage and terrain mutation, and the constructed height model has deviations.

[0004] The existing method only focuses on the elimination of the terrain influence and fails to consider the destruction of the point cloud structure caused by the morphological mutation of the disaster bamboo. When the traditional normalization processing is directly applied to the point cloud containing the bent and broken bamboo, the structure distortion of the point cloud data will be aggravated, and obvious point cloud discontinuity will be generated at the bamboo morphological mutation position, which is manifested as a jagged distribution and a height fault, and the morphological continuity of the bamboo is seriously damaged, which brings great difficulties to the subsequent analysis. SUMMARY

[0005] The present application aims to provide a ground laser radar point cloud normalization processing method applied to a bamboo forest after a snow disaster to solve the problems in the background art.

[0006] The object of the present application can be achieved by the following technical solutions. A ground laser radar point cloud normalization processing method applied to a bamboo forest after a snow disaster, comprising the following steps: Step 1: extracting the entire ground by using the Trimble RealWorks software, performing ground point extraction on the original point cloud by using an improved progressive triangulation network filtering algorithm, and generating classified ground point cloud; Step 2: constructing a space index model by using the two-dimensional plane coordinates of the classified ground point cloud, and generating an elevation reference model with efficient query capability; Step three: initial relative elevation calculation is performed on the complete sample point cloud through the elevation reference model to obtain initial relative elevation values of the points; Step four: stratified sliding window median smoothing is performed on the initial relative elevation values according to preset layer height and sliding step to generate point cloud elevation cumulative values including smoothed elevation values and smoothing times; Step five: global K-nearest neighbor median smoothing is performed on the uncovered points after stratified sliding window median smoothing according to two-dimensional plane coordinates to output final relative elevation values of the missing points; Step six: result fusion and translation zero processing are performed on the smoothed elevation values according to the smoothing times in the elevation cumulative values to output normalized final point cloud data set.

[0007] As preferably, the step one specifically includes: extracting the entire ground area from the original point cloud by using a preprocessing software; refining accurate ground points from the extracted ground area by using an improved progressive triangulated irregular network filtering algorithm; loading three-dimensional spatial coordinate information and reflectivity intensity information on the complete sample point cloud; The improved progressive triangulated irregular network filtering algorithm improves the terrain fitting accuracy by iteratively optimizing the ground seed points.

[0008] As preferably, the step two includes: The space index module adopts a KD tree data structure for construction; the KD tree data structure organizes ground point data by recursively dividing the two-dimensional coordinate space; the KD tree data structure realizes fast nearest neighbor query and provides acceleration support for elevation reference query.

[0009] As preferably, the step four specifically includes: a plurality of sliding windows with fixed layer height intervals are set along the relative elevation direction; the sliding step is set as half of the layer height interval to form a continuous and overlapping sliding window sequence; K-nearest neighbor search is performed on the point set falling into each sliding window based on the two-dimensional plane coordinates thereof; the median statistical quantity of the relative elevation values of the K nearest neighbor points of each point is calculated; the median statistical quantity is used to update the relative elevation value of the center point; the elevation cumulative value of each point after smoothing and the total number of times of smoothing thereof are recorded.

[0010] As preferably, the layer height interval is in the range of 0.01 meters to 0.2 meters; The layer height interval is dynamically adjusted according to the point cloud density and terrain complexity; a smaller layer height interval value is used for a complex terrain area, and a larger layer height interval value is used for a flat terrain area.

[0011] Preferably, the K value in the K nearest neighbor search is dynamically adjusted according to the local point cloud density. The K value is set to a range of 5 to 25; a larger K value is used in a dense point cloud area, and a smaller K value is used in a sparse point cloud area; the optimal K value parameter is automatically determined by local point cloud density estimation.

[0012] Preferably, the calculation formula of the median statistical quantity is: , wherein, represents the relative elevation value of K nearest neighbors of a certain point, and Median is the median function; the median statistical quantity has robustness to abnormal elevation values and can effectively maintain the original morphological characteristics of the point cloud.

[0013] Preferably, in step five, the global K nearest neighbor median smoothing uses the same K value parameter as the layered sliding window median smoothing. The same median statistical algorithm is used to ensure consistency of the processing results; the global K nearest neighbor median smoothing is specially used to process the missing points not covered by any sliding window; through global search, 100% of the point cloud data is ensured to be smoothed.

[0014] Preferably, in step six, the result fusion and translation to zero specifically includes: For points smoothed at least once, divide the accumulated elevation value by the number of times of smoothing to obtain the average smoothed elevation value; The final relative elevation values of all points are subjected to overall translation processing, so that the minimum elevation value in the data set is zero; The translation processing is realized by the following formula: , wherein, is the final processed elevation value of the point, is the minimum elevation value in all points, is the normalized final elevation value.

[0015] Preferably, after step six, a post-processing step is further included: Statistical filtering processing is performed on the normalized final point cloud data set to remove possible outlier noise points; The filtering threshold parameter is set based on statistical analysis of the point cloud density; The processed point cloud data is output as a general laser radar point cloud format file; The format file includes LAS format and XYZ format, which is used for subsequent point cloud analysis and feature extraction.

[0016] Advantages of the present application: The present application can significantly improve the data discontinuity problem caused by the morphological variation of bamboo bending and breaking caused by snow disaster in point cloud through the innovative hierarchical sliding window and global bottom-up double smoothing mechanism. This method effectively smooths the "sawtooth" fluctuation and "fault" phenomenon while eliminating the influence of terrain, so that the normalized point cloud can more truly and continuously reflect the actual three-dimensional morphology of the disaster bamboo, laying a reliable data foundation for subsequent accurate parameter inversion.

[0017] The elevation smoothing strategy based on two-dimensional spatial proximity adopted by the present application reduces the absolute dependence on the integrity of ground point extraction. Even in the case of complex terrain and poor forest environment leading to partial loss or deviation of ground point extraction, this method can still correct the elevation anomaly through its unique smoothing algorithm to obtain stable and reliable normalization results. This feature makes it particularly suitable for special application scenarios such as post-disaster forest structure disorder and difficulty in accurately obtaining ground points. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a ground laser radar point cloud normalization processing method applied to post-disaster bamboo forest according to the present application; Figure 2 A diagram for extracting the entire ground in a ground laser radar point cloud normalization processing method applied to post-disaster bamboo forest according to the present application.

[0019] Figure 3 A diagram for cutting the ground size in a ground laser radar point cloud normalization processing method applied to post-disaster bamboo forest according to the present application.

[0020] Figure 4 A diagram for extracting ground points in a ground laser radar point cloud normalization processing method applied to post-disaster bamboo forest according to the present application.

[0021] Figure 5 A diagram for point cloud normalization in a ground laser radar point cloud normalization processing method applied to post-disaster bamboo forest according to the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0023] REFERENCE Figures 1-5As shown, in the present embodiment, a ground laser radar point cloud normalization processing method applied to a bamboo forest after a snow disaster includes the following steps: Step one: extract the entire ground surface by Trimble RealWorks software, use the improved progressive encryption triangulation filter algorithm to perform ground point extraction on the original point cloud, and generate classified ground point cloud; Step two: use the two-dimensional plane coordinates of the classified ground point cloud to construct a spatial index model, and generate an elevation reference model with high query capability; Step three: calculate the initial relative elevation of each point by the elevation reference model to obtain the initial relative elevation value of each point; Step four: according to the preset layer height and sliding step, implement layered sliding window median smoothing on the initial relative elevation value to generate point cloud elevation cumulative value containing smoothed elevation value and smoothing times; Step five: according to the two-dimensional plane coordinates, perform global K nearest neighbor median smoothing on the uncovered points after layered sliding window median smoothing, and output the final relative elevation value of the missing points; Step six: according to the smoothing times in the elevation cumulative value, perform result fusion and translation zero processing on the smoothed elevation value, and output the normalized final point cloud data set.

[0024] In the embodiment of the present application, the entire ground surface is extracted by Trimble RealWorks software, and the improved progressive encryption triangulation filter algorithm is used to perform ground point extraction on the original point cloud to generate classified ground point cloud, and the specific implementation includes: In the data preprocessing and initial ground point extraction stage, first, the original point cloud data collected by the ground laser radar is preprocessed to remove obvious noise and outliers. Then, based on the region growing algorithm and morphological filtering principle, the professional tool provided by Trimble RealWorks software is used to quickly extract the complete ground area from the original point cloud. This step uses the advanced algorithm built-in the software to effectively identify large-scale terrain features, providing reliable basic data for subsequent processing.

[0025] After obtaining the ground area extracted by the software, grid seed point selection is implemented. The extracted ground area is divided into regular grid units, and the lowest point in each grid is calculated as the initial seed point. These seed points constitute the basis for constructing the initial irregular triangulation network, and the initial terrain model established by the grid method can reflect the general outline of the terrain, laying the foundation for subsequent iterative encryption processing. Although the initial triangulation model has limited accuracy, it can effectively exclude most non-ground object interference and provide a robust starting point for the algorithm.

[0026] In the iterative encryption filtering stage of the core, based on the initial triangular net model, all data points not classified as ground points are traversed. The algorithm calculates the spatial geometric relationship of each point with the nearest triangle in the triangular net, using a double judgment criterion: both the vertical distance of the point to the triangle plane and the spatial angle between the point and the triangle normal vector are evaluated. For the special environment of the bamboo forest after the snow disaster, a relatively loose initial angle threshold and distance threshold are set. Only when the data point meets the double conditions of distance threshold and angle threshold at the same time, it is determined as a candidate ground point and added to the ground point set, while the local irregular triangular net model is dynamically updated.

[0027] Dynamic threshold optimization and terrain fitting are implemented. After each complete iteration process ends, the angle threshold and distance threshold are simultaneously reduced according to the preset shrinkage coefficient. This dynamic adjustment mechanism enables the algorithm to adapt to complex terrain fluctuations caused by snow, fallen bamboo rods, etc. in the early stage, accurately capturing macro terrain features; in the later stage, it uses stricter standards to filter real ground points in detail, effectively avoiding misclassification of bamboo rods or residual vegetation close to the ground as ground. Through this coarse-to-fine progressive strategy, the algorithm can adaptively fit an accurate ground surface model.

[0028] After completing the iterative encryption, post-processing and result optimization are performed. Manual inspection and morphological filtering are performed on the preliminary obtained ground point set to remove a small number of possible residual non-ground points. The optimized high-precision ground point cloud is integrated with the non-ground points in the original point cloud to generate a complete sample point cloud data set containing accurate classification. The high-quality ground point cloud generated in this stage is used as a direct data source for constructing an elevation reference model, and its accuracy has a decisive influence on the final effect of the entire normalization process, which is the key foundation for the successful implementation of the method.

[0029] In the embodiment of the application, a spatial index model is constructed using the two-dimensional plane coordinates of the classified ground point cloud to generate an elevation reference model with efficient query capability. The specific implementation includes: Data structure initialization and spatial division are performed. The classified ground point cloud data is loaded into the processing system, and the two-dimensional plane coordinates (x, y) and the corresponding elevation value z of each ground point are extracted. Based on these two-dimensional coordinates, a KD tree spatial index structure is constructed. The construction process uses a recursive partitioning strategy, selects the dimension with the maximum data variance as the partitioning dimension, and uses the median point as the partitioning point to ensure that the generated binary tree remains balanced. Each tree node stores the statistical information of the point set in its corresponding spatial range, including the number of points, coordinate range, and elevation distribution characteristics. Through this hierarchical spatial division, an optimized data structure foundation is established for subsequent efficient spatial queries.

[0030] KD tree construction and optimization is performed. The construction process starts from the root node, recursively dividing the two-dimensional space into two non-overlapping sub-regions. The partitioning hyperplane is perpendicular to the current partition dimension coordinate axis and passes through the selected median point. Each internal node stores the partition dimension and partition threshold, and the leaf node stores the set of ground points falling within the space partition. To improve query efficiency, a leaf node capacity threshold is set, and when the number of points in the node exceeds the threshold, the partitioning is continued. After the construction is completed, tree structure optimization is performed, including subtree balancing adjustment and node data compression, to minimize the tree depth and optimize the query path. The optimized KD tree can complete the nearest neighbor search in average time complexity.

[0031] Further establish an elevation interpolation model based on the KD tree index structure to build an elevation reference model suitable for the entire sampling area. For any query point p(x, y), the leaf node region it belongs to is quickly located by traversing the KD tree. The inverse distance weighted interpolation algorithm is used to calculate the elevation reference value of the point. The interpolation formula is: where the weight , represents the Euclidean distance from the query point p to the adjacent ground point , the weight value of the i-th adjacent point is , and the elevation value of the i-th adjacent point is . The k nearest neighbor ground points around the query point are efficiently retrieved by the KD tree, and the elevation values of these adjacent points are used for weighted averaging to generate a continuous and smooth elevation reference surface.

[0032] Query optimization and performance improvement are implemented. To improve the query efficiency of the elevation reference model, various optimization strategies are implemented based on the KD tree structure. These include pre-computing the elevation values of commonly used query grids, establishing a query cache mechanism, implementing batch query processing to reduce tree traversal overhead, and using approximate nearest neighbor search algorithms to improve query speed while ensuring accuracy. For large-scale point cloud data processing needs, a parallel query architecture is designed, which divides the query area into multiple sub-regions and uses multi-threading technology to process multiple query requests simultaneously. These optimization measures significantly improve the query performance of the elevation reference model when faced with massive point cloud data.

[0033] Model verification and accuracy evaluation are performed. The accuracy and reliability of the elevation reference model are evaluated through cross-validation. Some ground points with known elevations are randomly selected as validation points, and the difference between the model predicted elevation and the actual measured value is compared. The root mean square error is calculated to evaluate the model accuracy, where n is the number of validation points, is the model predicted elevation value of the i-th point, Actual measured elevation value of the i th point. Meanwhile, analyze the spatial distribution characteristics of the error, and perform local model optimization in the area with larger error, including increasing the local sampling density, adjusting the interpolation parameters, etc. The qualified elevation reference model will be used as the reference for subsequent point cloud normalization processing.

[0034] In the embodiment of the application, the initial relative elevation of each point is obtained by calculating the initial relative elevation of the complete sample point cloud based on the elevation reference model, and the specific implementation includes: Data preparation and model loading are performed, and the complete sample point cloud data containing all ground points and non-ground points are loaded into the processing system. At the same time, the elevation reference model is loaded, which is constructed based on the KD tree spatial index structure and the inverse distance weighted interpolation method. The system establishes the coordinate correspondence between the point cloud data and the elevation reference model to ensure that they are in the same plane coordinate system and elevation reference. The initial relative elevation calculation result storage structure is initialized, and storage space is allocated for each point to record its initial relative elevation value and related calculation state.

[0035] Point cloud traversal and reference elevation query are performed, and each point in the complete sample point cloud is processed in a sequential traversal manner. For the current processing point , its two-dimensional plane coordinates are extracted as query inputs. Through the query interface provided by the elevation reference model, the elevation reference value corresponding to the plane position is obtained. The query process uses KD tree spatial index to realize efficient nearest neighbor search, locates the spatial partition containing the point and retrieves the adjacent ground points, and calculates the accurate reference elevation value based on the inverse distance weighted algorithm.

[0036] Relative elevation calculation and result recording are performed, and the reference elevation value of the point is obtained, and then the initial relative elevation value of the point is calculated. The calculation formula is: , wherein represents the original absolute elevation value of the point , and represents the elevation reference value of the corresponding position, and is the initial relative elevation value calculated. The calculation process converts the absolute elevation of the point into the relative elevation with the local terrain as the reference, eliminating the influence of the terrain undulation on the point cloud elevation value. The system stores the calculation result in the output data set together with the point identification information.

[0037] Outlier detection and result verification are performed, with data quality control implemented concurrently during the calculation process. A reasonable threshold range for relative elevation values ​​is set, and outliers exceeding this threshold are marked and reviewed. Outlier review includes checking the accuracy of the original point cloud coordinates, verifying the reliability of the elevation benchmark model query results, and confirming the correctness of the coordinate transformation parameters. For confirmed outliers, their abnormal states are recorded, allowing for selective exclusion and repair using interpolation methods in subsequent processing stages.

[0038] After processing all points, the calculation results are integrated and output. The initial relative elevation calculation results are then combined to form a complete dataset. This dataset retains the planar coordinates and intensity information of the original point cloud, while also including the calculated initial relative elevation values. The output data is stored in a standardized format to ensure compatibility with subsequent processing modules. A processing report is generated, summarizing key indicators from the calculation process, including the number of processed points, the number of outliers, and calculation accuracy assessments, providing a quality reference for subsequent processing steps.

[0039] In this embodiment of the invention, the initial relative elevation value is subjected to layered sliding window median smoothing according to a preset layer height and sliding step size, generating a cumulative point cloud elevation value that includes the smoothed elevation value and the number of smoothing times. Specific implementation methods include: Perform parameter initialization and data structure preparation, and set the core parameters for layered sliding window processing: layer height interval. With sliding step size The sliding step size is half of the floor height interval, i.e. Establish a storage structure for cumulative point cloud elevation values, allocating storage space for each point to record the cumulative smoothed elevation value and the number of smoothing iterations. Initialize a sliding window sequence, starting from the minimum relative elevation value along the relative elevation direction, and sequentially create sliding windows covering different elevation intervals in increments of the sliding step size. Each sliding window records the elevation range it covers. and the index of the point set it contains.

[0040] Construct a sliding window sequence and assign point sets. Based on the distribution range of the initial relative elevation values, calculate the required number of sliding windows N: ,in and Let represent the maximum and minimum initial relative elevation values ​​in the point cloud, respectively. The elevation range of each sliding window is generated sequentially, with the coverage area of ​​the i-th window being... Iterate through each point in the point cloud, determine the sliding window it falls into based on its initial relative elevation value, and add the point index to the point set of the corresponding window. Establish a many-to-many mapping relationship between points and windows, and record which sliding windows cover each point.

[0041] Perform K-nearest neighbor search and median calculation on the point set within the window. For each point in the sliding window, perform the K-nearest neighbor search point by point. Based on its two-dimensional plane coordinates Search for the K nearest neighbors in the current sliding window's point set. Calculate the median statistic of the relative elevation values ​​of these nearest neighbors: ,in arrive This represents the relative elevation values ​​of the K nearest neighbors, with Median being the median function. The value of K is dynamically determined based on the point cloud density; a larger K value is used for regions with high density, and a smaller K value is used for regions with low density, ensuring statistical stability.

[0042] Implement elevation value updates and cumulative records, and calculate the median statistic. As a point Smoothing results in the current window. Update point. The cumulative elevation value is calculated by adding the current value to its original cumulative value. At the same time, add some points. A smoothing counter is used. A point status marking mechanism is established to record the processing status of each point in each window, preventing duplicate counting. For each sliding window, the number of processed points and processing progress are recorded to ensure processing continuity between windows.

[0043] To address window boundary effects and point set overlap, since overlapping areas exist between sliding windows, some points may be covered by multiple windows. A strategy for handling overlapping points is designed to ensure that these points can participate independently in the smoothing calculation within each covering window. The point set overlap rate at window boundaries is calculated to verify the rationality of the overlap settings. For points located at the window edges, the completeness of their K-nearest neighbor search is specifically checked to ensure that nearest neighbor points are not missed due to window segmentation.

[0044] Implement quality control and anomaly handling, monitoring data quality in real time during the smoothing process. Detect anomalies in median calculation, including insufficient nearest neighbors and abnormally dispersed elevation values. For cases with excessively large variance in the elevation values ​​of the K nearest neighbors, activate alternative processing schemes, such as mean filtering or expanding the search range. Record the number of outliers encountered during processing and the processing results, generating a quality control report.

[0045] After completing the result integration and data output, integrate the cumulative elevation values ​​of the point cloud after all sliding windows have been processed. Generate a complete dataset containing the cumulative smoothed elevation value and the number of smoothing iterations for each point. Verify the integrity of the data processing to ensure that all points falling within the sliding window have been correctly processed. Output a detailed processing report for layered sliding window median smoothing, including window parameters, statistics on the number of processed points, quality assessment indicators, and other information. Provide a standardized data interface for subsequent processing steps.

[0046] To optimize processing efficiency and resource management, a parallel processing strategy is adopted for large-scale point cloud data. Sliding windows are distributed across multiple processing threads, processing point sets within multiple windows simultaneously. An efficient memory management mechanism is designed to load window point set data on demand, reducing memory consumption. A processing progress monitoring system is established to display the processing status of each window and the overall processing progress in real time. The K-nearest neighbor search algorithm is optimized, utilizing a spatial index structure to improve search efficiency, ensuring the completion of the massive point cloud processing task within the specified time.

[0047] Parameter sensitivity analysis and effect verification were conducted, and the floor height spacing was analyzed through experiments. The influence of the K-value on the smoothing effect is investigated. Typical regions are selected to test the processing results under different parameter combinations, evaluating the impact of parameter changes on point cloud smoothness and detail preservation. Parameter optimization selection criteria are established to provide parameter configuration suggestions for different point cloud features. The elevation distribution characteristics of point clouds before and after smoothing are compared to verify the effectiveness of layered sliding window median smoothing in eliminating noise and maintaining morphological continuity.

[0048] In this embodiment of the invention, based on two-dimensional planar coordinates, global K-nearest neighbor median smoothing is performed on the uncovered points after layered sliding window median smoothing, and the final relative elevation value of the missing points is output. Specific implementation methods include: An uncovered point identification mechanism is established. The system traverses each data point in the point cloud dataset and identifies all points not covered by the layered sliding window by querying their smoothing count record. These uncovered points are mainly distributed in: elevation gaps in the sliding window sequence, areas of bamboo morphological variation, and areas where morphological variations intersect or mix with surrounding bamboo. The system establishes an independent set of uncovered points, recording the spatial location information and initial relative elevation value of each point, while retaining its index mapping relationship with the original point cloud data.

[0049] A global processing environment is constructed, and a spatial index structure is rebuilt based on the two-dimensional planar coordinates of the complete sample point cloud. This index covers all data points, both those that have undergone hierarchical sliding window processing and those that have not. A balanced KD-tree algorithm is used to organize the spatial data, ensuring efficient nearest neighbor query performance when processing massive point clouds. The index structure includes the spatial coordinates, processing status flags, and elevation information for each point, providing data support for subsequent global searches.

[0050] A global nearest neighbor search and elevation calculation are implemented. For each uncovered point, the system performs a K-nearest neighbor search in its two-dimensional plane coordinate space. The search range is not limited by elevation layering; it finds the K spatially closest points from the entire point cloud data. The relative elevation values ​​of these nearest neighbors are collected, sorted in ascending order, and the median is taken as the final relative elevation value of the uncovered point. The calculation process employs robust statistical methods to ensure reasonably smooth results even when the point cloud density is uneven.

[0051] Perform data processing and state update, assign the calculated median elevation result directly as the final relative elevation value of the uncovered point. At the same time, update the processing state flag of the point, mark it as completed smoothing processing. The system maintains a special processing log to record the original elevation of each uncovered point, the processed elevation and the distribution characteristics of its neighboring points, which facilitates subsequent quality tracing and effect evaluation.

[0052] Perform quality verification and abnormal processing, the system performs automatic quality check on the global smoothing result, verifies the elevation continuity of the uncovered point after processing and the surrounding point cloud. For abnormal situations that occur during processing, such as insufficient number of neighboring points or abnormal discrete elevation values, the system starts the backup processing mechanism. By gradually expanding the search range or introducing elevation constraint conditions, it ensures that all uncovered points can obtain reasonable elevation assignment.

[0053] Complete data integration and output preparation, after all the uncovered points are processed, the system integrates the global smoothing result with the layered sliding window processing result to form a unified point cloud elevation data set. Verify the integrity and consistency of the data set to ensure that each point has a final relative elevation value. Output preparation includes data format standardization, metadata generation and processing report compilation to provide complete and reliable data input for subsequent processing stages.

[0054] In the embodiment of the present application, according to the smoothing times in the elevation accumulation value, the smoothed elevation value is subjected to result fusion and translation zero processing, and a normalized final point cloud data set is output, and the specific implementation includes: The system reads the point cloud elevation accumulation data set after layered sliding window smoothing and global smoothing processing. Verify data integrity to ensure that each point contains smoothed elevation accumulation value and smoothing times record. Establish a point cloud state classification mechanism to divide points into three processing categories: points that have undergone multiple smoothing, points that have undergone only one smoothing, and points that have undergone global smoothing but not layered smoothing. Design corresponding result fusion strategies for each category.

[0055] For points with smoothing times greater than zero, calculate their average smoothed elevation value. Use the formula: Wherein represents the elevation value obtained by the jth smoothing, represents the total smoothing times of the point, represents the calculated average smoothed elevation value. For points that have only undergone single smoothing, their average elevation value is equal to the elevation value of this smoothing. Establish an average elevation value quality control mechanism to range check and detect outliers of the calculation result.

[0056] Implement translation zero processing, scan all points' average smoothed elevation values, determine the minimum elevation value in the data set The translation transformation formula is executed: This transformation ensures that all elevation values in the normalized point cloud dataset are non-negative, with the minimum elevation value being zero. The mathematical correctness of the translation transformation is verified, confirming that the transformation process does not change the spatial relative relationships and morphological characteristics of the point cloud. The translation transformation parameters, including their numerical values and transformation timestamps, are recorded.

[0057] Data quality assessment and post-processing are performed to conduct a comprehensive quality check on the normalized point cloud dataset. The distribution characteristics of the elevation values are analyzed to verify their compliance with the expected statistical rules. Possible outliers, including negative elevation values, abnormally large values, and elevation values that do not obviously conform to spatial continuity, are detected and processed. Data consistency verification is performed to ensure the correct correspondence between planar coordinates and elevation values.

[0058] Data formatting and output are completed to convert the processed normalized point cloud data into a standard format. The data output includes the three-dimensional coordinate information of each point, with the planar coordinates remaining as the original values and the elevation coordinates being the normalized values. Processing metadata, including processing parameters, quality assessment results, and transformation records, are also output. A data version management mechanism is established to ensure the traceability of the output data and the processing process. Various data interfaces are provided for subsequent applications to support compatibility requirements for different platforms.

[0059] It should be noted that the relational terms such as first and second, and the like, are used solely to distinguish one from another entity or action, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0060] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements, and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A ground laser radar point cloud normalization processing method applied to a post-snow disaster Phyllostachys edulis forest, characterized by, The method comprises the following steps: Step 1: using an improved progressive triangulated irregular network filtering algorithm to perform ground point extraction on the original point cloud to generate a classified ground point cloud; Step 2: using the two-dimensional plane coordinates of the classified ground point cloud to construct a spatial index model to generate an elevation reference model with high query efficiency; Step 3: using the elevation reference model to perform initial relative elevation calculation on the complete sample point cloud to obtain initial relative elevation values of each point; Step 4: performing hierarchical sliding window median smoothing on the initial relative elevation values according to a preset layer height and a sliding step to generate point cloud elevation cumulative values containing smoothed elevation values and smoothing times; Step 5: performing global K-nearest neighbor median smoothing on the uncovered points after hierarchical sliding window median smoothing according to the two-dimensional plane coordinates to output final relative elevation values of the missing points; Step 6: performing result fusion and translation zero processing on the smoothed elevation values according to the smoothing times in the elevation cumulative values to output a normalized final point cloud dataset.

2. The method according to claim 1, characterized in that, The step 1 specifically comprises: extracting the entire ground area from the original point cloud using a preprocessing software; refining accurate ground points from the extracted ground area using an improved progressive triangulated irregular network filtering algorithm; loading three-dimensional spatial coordinate information and reflectivity intensity information on the complete sample point cloud; the improved progressive triangulated irregular network filtering algorithm improves the terrain fitting accuracy by iteratively optimizing the ground seed points.

3. The method according to claim 1, characterized in that, The step 2 comprises: the spatial index model is constructed using a KD-tree data structure; the KD-tree data structure organizes ground point data by recursively dividing the two-dimensional coordinate space; the KD-tree data structure is used to realize fast nearest neighbor query and provide acceleration support for elevation reference query.

4. The method according to claim 1, characterized in that, In the step 4, the hierarchical sliding window median smoothing specifically comprises: setting a plurality of sliding windows with a fixed layer height interval along the relative elevation direction; setting the sliding step to be half of the layer height interval to form a series of continuously overlapping sliding windows; performing K-nearest neighbor search on the point set falling within each sliding window based on its two-dimensional plane coordinates; calculating the median statistics of the relative elevation values of the K nearest neighbors of each point; updating the relative elevation value of the center point using the median statistics; recording the smoothed elevation cumulative value of each point and the total number of times it has been smoothed.

5. The method according to claim 4, characterized in that, The layer height interval ranges from 0.01 meters to 0.2 meters; the layer height interval is dynamically adjusted according to the point cloud density and terrain complexity; a smaller layer height interval value is used for complex terrain areas and a larger layer height interval value is used for flat terrain areas.

6. The ground laser radar point cloud normalization processing method for a post-snow disaster Phyllostachys edulis forest according to claim 4, characterized in that, The K value in the K-nearest neighbor search is dynamically adjusted according to the local point cloud density; the K value ranges from 5 to 25; a larger K value is used in dense point cloud areas and a smaller K value is used in sparse point cloud areas; the optimal K value parameter is automatically determined by local point cloud density estimation.

7. The method according to claim 4, characterized in that, The calculation formula of the median statistic is: wherein, represents the relative elevation value of K neighboring points of a certain point, and Median is a median function; the median statistic has robustness to abnormal elevation values and can effectively maintain the original morphological characteristics of the point cloud.

8. The method according to claim 1, characterized in that, In the step 5, the global K-nearest neighbor median smoothing uses the same K value parameter as the hierarchical sliding window median smoothing. The same median filtering algorithm is used to ensure consistency of the processing results; the global K-NN median smoothing specifically handles the missing points that are not covered by any sliding window; and a global search is used to ensure that 100% of the point cloud data is smoothed.

9. The method according to claim 1, characterized in that, The result fusion and translation to zero in step six specifically includes: For points that are smoothed at least once, divide the accumulated elevation value by the number of times it is smoothed to obtain the average smoothed elevation value; Translate the final relative elevation values of all points as a whole to make the minimum elevation value in the data set zero; The translation process is achieved by the following equation: wherein, is the final processed elevation value for a point, is the minimum elevation value among all points, is the normalized final elevation value.

10. The method according to claim 1, characterized in that, The post-processing step after step six includes: Perform statistical filtering on the normalized final point cloud data set to remove possible outlier noise points; Set the filtering threshold parameter based on statistical analysis of the point cloud density; Output the processed point cloud data in a general laser radar point cloud format file; The format file includes LAS format and XYZ format, which is used for subsequent point cloud analysis and feature extraction.

Images