Point cloud denoising method and system for unstructured scene of engineering machinery

Abstract

The invention relates to the technical field of point cloud data processing, and discloses a point cloud denoising method and system for an unstructured scene of engineering machinery, and the method comprises the steps: collecting original point cloud data of the unstructured scene of the engineering machinery, carrying out the preprocessing of the original point cloud data, and obtaining the preprocessed point cloud data; based on the function association features of the preprocessed point cloud data, performing functional region division on the scene, and dividing a functional region into a driving region, a driving boundary region and an operation region; dynamically generating a denoising threshold value to which each functional region belongs based on terrain fluctuation characteristics of the divided different functional regions; fusing the denoising threshold value, the elevation continuity and the reflection intensity of each functional area, screening out useful terrain points and engineering machinery equipment points, retaining the screened useful terrain points, and rejecting the engineering machinery equipment points to obtain a screened point cloud; and performing connectivity optimization and reverse verification on the screened point clouds, and outputting final de-noised point clouds. According to the method, the precision of point cloud terrain modeling can be improved, and reliable data support is provided for intelligentization of engineering machinery.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, and specifically to a point cloud denoising method and system for unstructured scenarios in engineering machinery. Background Technology

[0002] With the accelerating pace of intelligentization in construction machinery, virtual testing of construction machinery in unstructured scenarios (such as open-pit mining, construction sites, and comprehensive test areas) is becoming increasingly crucial. Point cloud data acquired by LiDAR scanning the area of ​​the construction machinery in these scenarios is the core basis for terrain modeling of the virtual testing scene. However, unstructured scenarios for construction machinery are characterized by complex and diverse terrain, dense and dynamically changing equipment, and numerous environmental interferences (such as dust obstruction and temporary accumulations). This often results in the collection of point clouds containing a large amount of noise, including equipment points of the construction machinery itself, measurement fly-by points, and environmental interference points. Since the construction machinery equipment is not part of the terrain, it needs to be removed in the terrain modeling; otherwise, the usability and effectiveness of the point cloud data will be severely affected.

[0003] The existing point cloud denoising technologies are as follows:

[0004] (1) The statistical outlier removal method is a classic filtering method based on spatial distance features. The core principle is to remove noise by analyzing the spatial distribution pattern of the local neighborhood of the point cloud. This method first calculates the average distance between each point and its K nearest neighbors, and then determines the standard deviation threshold based on the distance distribution of all points. When the distance of a single point exceeds the mean plus a certain number of standard deviations, it is identified as an outlier and removed. Its typical applications include removing measurement error flypoints in laser scanning, distinguishing isolated noise from continuous terrain points by density differences, and is suitable for processing scattered random noise.

[0005] However, statistical outlier removal methods have significant limitations: relying solely on spatial distance features, they cannot distinguish between "noisy points with elevation anomalies" and "useful terrain points with natural elevation abrupt changes." In unstructured scenarios involving construction machinery, terrain points such as those in the boundary areas of the driving zone and slopes of the work area, despite having significant elevation differences from their neighbors, are part of continuous terrain structures and are easily misjudged as outliers and deleted. Furthermore, fixed neighborhood sizes and standard deviation thresholds are difficult to adapt to complex scenarios. For example, point clouds in densely populated areas of construction machinery may be retained due to localized high density, while small undulations in the driving zone may be mistakenly deleted due to distance threshold triggers, leading to disruption of terrain continuity.

[0006] (2) The fixed elevation threshold filtering method is a filtering technique optimized for terrain scenes. The core idea is to distinguish between ground points and non-ground points by setting a preset elevation difference threshold. This method first establishes a local reference terrain surface, then calculates the elevation difference between each point and the reference surface. When the difference exceeds the fixed threshold, it is judged as an anomaly and removed. It removes non-terrain targets that exceed the elevation range, is suitable for noise filtering in areas with gentle terrain undulations, and is simple to operate and has high computational efficiency.

[0007] However, the core flaw of the fixed elevation threshold filtering method lies in the lack of adaptability of the threshold parameter, its failure to fully utilize material properties, and its poor scene adaptability. In unstructured construction machinery scenes, there may be natural elevation changes such as roadside boundaries and slopes, whose elevation characteristics highly overlap with those of construction machinery equipment (metallic materials) such as mining trucks and excavators. If the threshold is set too low, useful terrain points are easily deleted; if the threshold is set too high, a large number of construction machinery point clouds will be retained. In addition, a globally uniform threshold cannot adapt to regional differences. For example, small elevation changes in the construction machinery driving area (such as ore spillage, road surface bumps or depressions caused by vehicle rolling) may be misjudged as noise, and construction machinery equipment in the work area may also be missed due to elevation overlap that does not reach the threshold, resulting in unstable denoising effects.

[0008] In summary, most existing point cloud denoising methods are general algorithms that are not customized for the functional area characteristics of unstructured engineering machinery scenarios (the terrain structure of the driving area, driving boundary area, and work area are significantly different). The thresholds used in the point cloud denoising process are mostly empirical fixed values ​​that cannot be adaptively adjusted according to the terrain features of different areas. As a result, the denoised point cloud either retains a lot of noise or excessively smooths useful terrain details, which in turn affects the accuracy of point cloud-based terrain modeling.

[0009] In addition, existing methods mostly rely on single features and lack the fusion of multiple features such as "differences in reflection intensity between terrain material and equipment material and continuity of regional elevation" in engineering machinery scenarios. It is difficult to accurately distinguish between useful terrain points and engineering machinery equipment points. In particular, in dynamic operation scenarios, the removal effect of dynamic noise such as moving engineering machinery and temporary obstacles is not good. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides a point cloud denoising method and system for unstructured scenarios of engineering machinery, which can improve the accuracy of point cloud terrain modeling and provide reliable data support for the intelligentization of engineering machinery.

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] In a first aspect, this invention proposes a point cloud denoising method for unstructured engineering machinery scenes, comprising:

[0013] Collect raw point cloud data of unstructured scenes of construction machinery, preprocess the raw point cloud data, and obtain preprocessed point cloud data;

[0014] Based on the functional association features of the preprocessed point cloud data, the scene is divided into functional areas, namely driving area, driving boundary area and operation area.

[0015] Based on the terrain fluctuation characteristics of different functional areas, the denoising threshold of each functional area is dynamically generated.

[0016] By integrating the noise reduction threshold, elevation continuity, and reflection intensity of each functional area, useful terrain points and engineering machinery equipment points are selected. The selected useful terrain points are retained, and the engineering machinery equipment points are removed to obtain the filtered point cloud.

[0017] The selected point cloud is subjected to connectivity optimization and reverse verification to output the final denoised point cloud.

[0018] In conjunction with the first aspect, further, the process of collecting raw point cloud data from unstructured scenes of engineering machinery, and preprocessing the raw point cloud data to obtain preprocessed point cloud data includes:

[0019] Collect raw point cloud data of unstructured scenes of engineering machinery and convert it into a format that includes three-dimensional coordinates (X,Y,Z) and reflection intensity;

[0020] Voxel downsampling is performed on the original point cloud data to obtain downsampled point cloud data;

[0021] Calculate the number of K nearest neighbors for each point in the downsampled point cloud data; if the number of K nearest neighbors for a point is less than k, then the point is determined to be an isolated noise point; remove the points determined to be isolated noise points to obtain the preprocessed point cloud data. Preferably, k=5.

[0022] In conjunction with the first aspect, further, based on the functional association features of the preprocessed point cloud data, the scene is divided into functional regions, namely, a driving area, a driving boundary area, and a work area, including:

[0023] Based on the preprocessed point cloud data, a basic grid is divided into the preprocessed point cloud data.

[0024] Based on the division of the basic grid, the functional association features of each grid are calculated; the functional association features include road surface smoothness σ. L Boundary drop ∆H, functional gradient proportion G F ;

[0025] Based on the functional association features of each grid, the driving area, driving boundary area and work area are initially delineated respectively;

[0026] The initially defined driving area, driving boundary area, and work area are subjected to grid refinement inspection, regional continuity constraint, and material characteristic auxiliary inspection to obtain the final defined driving area, driving boundary area, and work area.

[0027] In conjunction with the first aspect, further, the step of performing basic grid division on the preprocessed point cloud data includes: performing basic grid division on the preprocessed point cloud data in the XY plane space outside the elevation Z coordinate dimension, based on the point cloud density and scene functional requirements.

[0028] In conjunction with the first aspect, further, the road surface smoothness σ L The elevation standard deviation of points within the base grid is used to reflect the characteristic index of the driving area meeting the flatness requirements for engineering machinery driving; the boundary drop ∆H is the difference between the maximum and minimum elevations within the base grid, used to reflect the characteristic index of the driving boundary area's ability to vertically define the driving range; the functional gradient ratio G F The percentage of points within the base grid that have a slope greater than or equal to the slope threshold of the steep boundary area is used to distinguish between the steep characteristics of the driving boundary area and the gentle slope characteristics of the work area.

[0029] In conjunction with the first aspect, the method for initially delineating the driving area, driving boundary area, and work area (i.e., the conditions for the initial division of each functional area) is as follows:

[0030] When the road surface smoothness σ of a certain grid L The requirements for the smoothness of the driving of construction machinery are met and its functional gradient ratio G F If the grid is below the preset threshold for distinguishing significantly steep structures (i.e., there are no significantly steep structures), then the grid is initially designated as a driving area.

[0031] When the boundary drop ∆H of a certain grid meets the vertical definition requirement of the driving range and its functional gradient ratio G F If the preset threshold for distinguishing the steepness of the boundary is reached (i.e., it meets the steepness characteristics of boundaries such as road edges), then the grid is initially defined as the driving boundary area.

[0032] The remaining grids that do not conform to the above two categories (including complex terrain such as slopes, stockpiles, and pits required for the operation) are initially designated as the operation area.

[0033] In conjunction with the first aspect, the method for further refining the grid is as follows: for the edge grid of the initially defined area, a secondary grid division verification is performed using a size half that of the basic grid. If more than half of the secondary grids do not meet the initial definition conditions of the functional area, the non-compliant grids are defined as mixed grids and assigned to the work area.

[0034] The method for constraining the continuity of the region is as follows: the driving area must satisfy "it and at least m adjacent grids around it are candidate driving areas" to ensure continuous vehicle passage; the driving boundary area must satisfy "at least n adjacent grids along the edge of the driving area are candidate driving boundary areas" to ensure that the boundary completely defines the driving range.

[0035] Preferably, the driving area must satisfy the condition that "itself and at least three consecutive adjacent grids are candidate driving areas"; the driving boundary area must satisfy the condition that "at least three consecutive grids along the edge of the driving area are candidate driving boundary areas".

[0036] The auxiliary verification method for material characteristics is as follows: the grid in the driving area and driving boundary area needs to have more than a% of the point reflection intensity within the R1 interval; where the R1 interval represents the reflection intensity interval of the corresponding road surface paving / soil and other terrain materials. Preferably, a=80.

[0037] Building upon the first aspect, further, after initially delineating functional regions based on the functional association characteristics of each grid, each functional region is ultimately determined and output through grid refinement verification, regional continuity constraints, and material characteristic-assisted verification (integrated quantitative indicators):

[0038] Driving area: The output of the corresponding area must meet the requirement of road surface flatness σ L ≤10~15cm (standard deviation of elevation of points within the basic grid), functional gradient percentage G F <30%~40% (no significant steep structure), and at least 3 consecutive adjacent grids of itself and its surroundings are candidate driving areas, while 80%~90% of the point reflection intensity within the grid is in a flat area of ​​the R1 interval corresponding to the road surface material.

[0039] Driving boundary zone: When outputting the corresponding area, the boundary elevation difference ∆H ≥ 0.5~0.8m (the maximum and minimum elevation difference within the basic grid) and the functional gradient ratio G are satisfied. F ≥70%~80% (meets the characteristics of steep boundaries), and at least 3 consecutive adjacent grids along the edge of the driving area are candidate driving boundary areas, while 80%~90% of the point reflection intensity within the grid is in the defined area of ​​the R1 interval corresponding to the curb-type boundary material;

[0040] Work area: When outputting the corresponding area, it is the work operation area outside the driving area and driving boundary area (such as excavation area, material stacking area, etc.), covering areas with complex terrain such as slopes, material stacks, and pits.

[0041] The functional area dynamic division and scene adaptation of this invention addresses the functional and terrain differences of the "driving area, driving boundary area, and work area" in unstructured scenarios of engineering machinery. It calculates features such as road surface smoothness and boundary drop through a function-oriented grid and combines them with regional continuity rules to achieve precise and mutually exclusive division of functional areas, thus solving the problem of "global uniform processing ignoring the functional heterogeneity of scene areas" in traditional methods.

[0042] In conjunction with the first aspect, further, the dynamic generation of denoising thresholds for each functional region based on the terrain fluctuation characteristics of the divided functional regions includes:

[0043] Based on the principle of "functional representativeness", preliminary screening is carried out in each functional area through neighborhood features or quantile features, and then secondary screening is carried out through statistical features such as median, standard deviation or fluctuation value. The set of topographic feature benchmarks that represent the inherent topographic morphology is extracted by secondary screening.

[0044] Calculate the elevation standard deviation σ of the set of topographic feature benchmarks for each functional area (the elevation standard deviation σ reflects the "reasonable range of topographic fluctuation" of the corresponding area) and quantify the elevation fluctuation of the functional area.

[0045] Combining the elevation standard deviation σ with the functional region's tolerance for anomalous point clouds L t (Quantized L in the range of 2.5 to 4.5) t (The lower the value, the lower the tolerance of the functional area to abnormal point clouds), generating adaptive denoising thresholds for each functional area.

[0046] In conjunction with the first aspect, further, based on the principle of "functional representativeness," preliminary screening is conducted within each functional area using neighborhood features or quantile features, followed by refined screening using statistical features such as median, standard deviation, or fluctuation value. This secondary screening process extracts a set of topographic feature benchmarks representing the inherent topographic morphology of each area, including:

[0047] Method for extracting the benchmark point set of terrain features in the driving area: First, randomly select m points from the driving area. d There are 10 candidate terrain points with a preset neighborhood radius r. d Perform a neighborhood search on the candidate terrain points in the driving area and calculate the standard deviation σ of the neighborhood elevation of each candidate point. h Then by preserving σ h ≤σ ht (σ) ht Candidate points are selected based on the elevation standard deviation threshold corresponding to the flatness of the driving area, resulting in a preliminary set of selected points; then, the elevation median μ is calculated for the preliminary set of selected points. h and elevation standard deviation σ hp By preserving the elevation at μ h ±σ hpThe points within the interval are carefully selected to form the benchmark point set of terrain features of the driving area.

[0048] Method for extracting the benchmark point set of terrain features in the driving boundary area: First, randomly select m points from within the driving boundary area. db There are 10 candidate terrain points with a preset neighborhood radius r. db Perform a neighborhood search on candidate terrain points in the driving boundary area, calculate the neighborhood slope p of each candidate point; then retain the slope p ≥ p t (p) t Candidate points are selected based on the slope threshold corresponding to the steep boundary characteristics, resulting in a preliminary set of selected points; then, the median slope μ is calculated for the preliminary set of selected points. p and slope standard deviation σ p Preserve slope at μ p ±σ p The points within the interval are carefully selected to form the final set of benchmark points for terrain features in the driving boundary area.

[0049] Method for extracting the benchmark point set of terrain features in the work area: First, randomly select m points from the work area. w 10 candidate terrain points were selected, and the 25th percentile Q of the work area elevation was calculated. 25 With the 75th percentile Q 75 Then, by preserving the elevation in Q... 25 ~Q 75 Candidate points within the interval are used to obtain a preliminary set of selected points; then, a preset neighborhood radius r is used... w Perform a neighborhood search on the initially selected points and calculate the neighborhood elevation fluctuation ΔH for each point. n (ΔH) n (The difference between the maximum and minimum elevations in the neighborhood), by retaining ΔH n ≤ΔH nt (ΔH) nt The points corresponding to the elevation fluctuation threshold of the work area terrain are finely screened to form the benchmark point set of the work area terrain features.

[0050] In conjunction with the first aspect, further, the calculation of the elevation standard deviation σ of the topographic feature benchmark set for each functional area, and the quantification of elevation fluctuations in the functional areas, includes:

[0051] Calculate the elevation standard deviation σ1 of the benchmark point set for terrain features in the driving area. σ1 is used to quantify the allowable range of small undulations in the road surface.

[0052] Calculate the elevation standard deviation σ2 of the benchmark point set for terrain features in the driving boundary area. σ2 is used to quantify the normal fluctuation range of the boundary structure.

[0053] Calculate the elevation standard deviation σ3 of the benchmark set of terrain features in the work area. σ3 is used to quantify the reasonable range of variation of the terrain in the work area.

[0054] In conjunction with the first aspect, further, the combination of elevation standard deviation σ and the functional region's tolerance for anomalous point clouds L... t (Quantized L in the range of 2.5 to 4.5) t (The lower the value, the lower the tolerance of the functional region to abnormal point clouds). Generate adaptive denoising thresholds for each functional region, including:

[0055] The method for generating the noise reduction threshold for the driving area is as follows: Because this functional area has extremely high requirements for terrain flatness and traffic stability, and the lowest tolerance for anomalous point clouds, the L value for this functional area is... t The value is then taken within the range of 2.5 to 3.0, and the calculated σ1 is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T1 for the driving area;

[0056] The method for generating the denoising threshold for the driving boundary region is as follows: Since this functional region needs to filter out interference points while preserving steep boundary features, its tolerance for anomalous point clouds is at a moderate level. The L value for this functional region... t The value is then taken within the range of 3.0 to 4.0, and the calculated σ² is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T2 for the driving boundary region;

[0057] The method for generating the denoising threshold for the work area is as follows: Since this functional area itself has elevation variations, it is necessary to retain terrain details relevant to the work, and it has the highest tolerance for anomalous point clouds. The L value for this functional area is... t The value is then taken within the range of 4.0 to 4.5, and the calculated σ3 is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T3 for the work area.

[0058] This invention sets an adaptive threshold specific to each functional area. Based on the elevation standard deviation of useful terrain points in each functional area, it customizes the denoising threshold according to differentiated needs, replacing the traditional "global single threshold". This significantly improves the denoising adaptability of various unstructured scenarios for construction machinery and avoids the accidental deletion of useful terrain points or the missed detection of equipment points on construction machinery.

[0059] In conjunction with the first aspect, further, by fusing the denoising threshold, elevation continuity, and reflection intensity of each functional region, useful terrain points and engineering machinery equipment points are selected. The selected useful terrain points are retained, and the engineering machinery equipment points are removed, resulting in a filtered point cloud, including:

[0060] Determine the functional area to which the point belongs based on its coordinates, and then call the noise reduction threshold of the corresponding functional area.

[0061] The terrain continuity of the test points based on the denoising threshold of each functional area;

[0062] Clustering of reflection intensity distinguishes between terrain materials and engineering machinery materials;

[0063] For points with blurred reflection intensity located within the work area, equipment integrity verification is performed;

[0064] The ambiguity of the reflection intensity refers to the reflection intensity being between R1 and R2, where R1 is the terrain reflection intensity range and R2 is the equipment reflection intensity range.

[0065] In conjunction with the first aspect, the method for verifying terrain continuity based on the denoising threshold test points of each functional region further includes:

[0066] Based on the point cloud density, select an appropriate neighborhood size, construct the neighborhood of the point, and calculate the elevation difference ∆H between the neighborhood points and the target point. p ;

[0067] For the target point, two indicators are statistically analyzed: the proportion of elevation difference compliance (A1) and the proportion of gradient consistency (A2). If A1 meets the preset requirement of functional area terrain continuity and A2 meets the preset requirement of functional area terrain trend, then the target point is determined to meet the terrain continuity.

[0068] Among them, the compliance rate of elevation difference is A1: within the neighborhood "|∆H p The percentage of points that are "≤ the regional threshold";

[0069] Gradient Consistency A2: The percentage of neighborhood points whose "elevation gradient direction (ascending / descending)" is consistent with the gradient direction of the target point.

[0070] In conjunction with the first aspect, further, the clustering of reflection intensity to distinguish between terrain materials and engineering machinery equipment materials includes:

[0071] K-means clustering was used to obtain the terrain reflection intensity range R1 (corresponding to terrain materials such as rock, soil, and road surface) and the equipment reflection intensity range R2 (corresponding to engineering machinery materials such as metal and rubber).

[0072] If the reflection intensity of a point is within the R1 interval and passes the above terrain continuity verification, it is determined to be a "useful terrain point" and retained.

[0073] If the reflection intensity of a point is within the R2 range, it is determined to be an engineering machinery equipment point and is eliminated.

[0074] If the reflection intensity of a point is between R1 and R2 (a fuzzy range, such as rubber tires, rusty metal, or non-metallic obstacles), and it is located in the work area (a high-frequency area for construction machinery and equipment), then equipment integrity verification is performed.

[0075] In conjunction with the first aspect, the method for verifying equipment integrity is as follows:

[0076] From the points whose reflection intensity is between R1 and R2 and which are the work area, select the points whose "reflection intensity meets the characteristics of the core material of the equipment, the number of similar points in the neighborhood meets the continuity of the equipment structure, and the elevation difference exceeds the range of terrain fluctuation in the work area" as the core seed points of the equipment.

[0077] Expand by "dynamic threshold of spatial distance ≤ equipment component width (adjusted according to equipment model and point cloud density)" and "dynamic threshold of reflection intensity fluctuation ≤ equipment material difference" to cover complete equipment;

[0078] The growth point set must conform to the size range (length and height are dynamically adjusted according to the equipment type) and geometric characteristics (such as tire cylindricity error, which are adjusted according to the equipment precision requirements) of the engineering machinery equipment. If the verification is passed, the entire equipment point set will be removed.

[0079] This invention achieves accurate identification through multi-feature collaboration, integrating multiple features such as "elevation continuity, reflection intensity clustering, and equipment morphology verification" to solve the pain point of traditional methods that "it is difficult to distinguish between low-reflection equipment components and terrain points with similar elevations". It enables accurate differentiation of engineering machinery equipment points, flying points, and useful terrain points, especially improving the identification rate of low-reflection equipment such as old tires and non-metallic parts.

[0080] In conjunction with the first aspect, further, the connectivity optimization and reverse verification of the filtered point cloud to output the final denoised point cloud includes:

[0081] Connectivity purification involves performing connectivity analysis on the filtered point cloud to remove isolated points remaining after filtering. When the number of connected points falls below the connectivity purification threshold C... jt When this happens, the connected component is determined to be noise and deleted entirely.

[0082] Connectivity purification involves performing connectivity analysis on the filtered point cloud to remove isolated points remaining after filtering. When the number of connected points falls below the connectivity purification threshold C... jt When this happens, the connected component is determined to be noise and deleted entirely.

[0083] Connectivity purification threshold C jt Method: First, quickly obtain the average number of points in the neighborhood of a single point. Randomly select 10-15 points from the set of terrain feature reference points of the functional area. Perform a fixed-radius neighborhood search for each point, count the number of points in the neighborhood of each terrain feature reference point, and calculate the average number of points N in the neighborhood. n Then, a correction coefficient is assigned to the average number of neighborhood points according to the functional area characteristics, and the average number of neighborhood points N is adjusted accordingly. n Correction coefficient C for adaptation region characteristics c (Driving Area C)c =0.3, Driving boundary area C c =0.4, Work Area C c =0.6) multiplied (C before rounding) jt =N n ×C c Then, round down to obtain the final connected component purification threshold C. jt ;

[0084] Functional area reverse verification: For each functional area, if the proportion of equipment points in a certain functional area exceeds the set fault tolerance ratio, the noise reduction threshold of that functional area is called for re-screening to remove elevation anomalies and ensure that there are no equipment point cloud residues in the core functional areas.

[0085] The output results in denoised point clouds for three functional areas, supporting 3D visualization.

[0086] This invention optimizes connectivity and performs reverse verification. It eliminates isolated misclassified points by purifying connected components and enhances the continuity and purity of classification results by combining reverse verification of functional regions. This further ensures the terrain modeling value of the denoised point cloud and its supporting value for intelligent applications such as engineering machinery path planning and operation control.

[0087] Secondly, this invention proposes a point cloud denoising system for unstructured engineering machinery scenes, implementing the aforementioned point cloud denoising method for unstructured engineering machinery scenes, including:

[0088] The point cloud preprocessing module is configured to collect raw point cloud data from unstructured scenes of engineering machinery, preprocess the raw point cloud data, and obtain preprocessed point cloud data.

[0089] The functional area division module is configured to divide the scene into functional areas based on the functional association features of the preprocessed point cloud data, namely, the driving area, the driving boundary area, and the work area.

[0090] A dedicated threshold calculation module is configured to dynamically generate denoising thresholds for each functional area based on the terrain fluctuation characteristics of the different functional areas.

[0091] The multi-feature point cloud filtering module is configured to integrate the denoising threshold, elevation continuity and reflection intensity of each functional area to filter out useful terrain points and engineering machinery equipment points. The useful terrain points are retained and the engineering machinery equipment points are removed to obtain the filtered point cloud.

[0092] The connectivity optimization and output module is configured to perform connectivity optimization and reverse verification on the filtered point cloud, and output the final denoised point cloud.

[0093] Thirdly, the present invention proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described point cloud denoising method for unstructured engineering machinery scenes.

[0094] Fourthly, the present invention provides a computer device comprising:

[0095] Memory, used to store computer programs;

[0096] A processor is used to execute the computer program to implement the steps of the above-described point cloud denoising method for unstructured engineering machinery scenarios.

[0097] Fifthly, the present invention proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described point cloud denoising method for unstructured engineering machinery scenarios.

[0098] This invention achieves precise differentiation between useful terrain points and equipment points, flying points, and environmental interference points in unstructured engineering machinery scenarios through a full-process design involving dynamic functional area division, adaptive threshold calculation, multi-feature collaborative screening, and connectivity optimization. During the denoising process, parameters are dynamically adapted based on point cloud density and scene functional characteristics, demonstrating strong compatibility with diverse unstructured scenarios such as construction sites and infrastructure sites. It solves the problems of erroneous deletion of useful terrain points or noise residue caused by unreasonable area division in traditional methods, and improves the recognition rate of low-reflectivity equipment components (such as rubber tires) through multi-feature fusion, ensuring stable and reliable denoising results. The final high-precision denoised point cloud output effectively supports high-precision terrain modeling in unstructured engineering machinery scenarios, providing data accuracy assurance for intelligent operations such as path planning, operation control, and environmental perception in complex unstructured scenarios, significantly improving operational efficiency and safety.

[0099] This invention can also be applied to point cloud denoising in scenarios such as dynamic monitoring of construction sites and topographic mapping of infrastructure projects, providing high-precision data support for environmental perception and construction planning.

[0100] Compared with the prior art, the present invention has the following beneficial effects:

[0101] (1) This invention solves the problems of "insufficient functional area targeting, poor adaptability of denoising threshold, and lack of multi-feature fusion in the denoising process" in the point cloud denoising of unstructured scenes of engineering machinery in the prior art. By dynamically dividing functional areas, adaptively calculating exclusive thresholds, collaboratively screening multiple features and optimizing connectivity, it provides a method and system for denoising point clouds in unstructured scenes of engineering machinery, which improves the accuracy of point cloud terrain modeling and provides reliable data support for the intelligentization of engineering machinery.

[0102] (2) This invention proposes a point cloud denoising scheme that integrates dynamic functional area division, adaptive calculation of exclusive threshold and multi-feature collaborative screening. Through functional-oriented grid division and continuity constraints, the mutually exclusive division of driving area, driving boundary area and operation area is achieved. Combined with functional area exclusive threshold and multi-feature screening of elevation continuity, reflection intensity and equipment shape, and with connected domain purification and functional area reverse verification optimization, the scheme achieves accurate removal of noise such as engineering machinery equipment points and flying points and complete preservation of useful terrain point clouds.

[0103] (3) The present invention is based on the functional area dynamic division method of functional guided grid and spatial continuity constraint to realize the mutual exclusion division of driving area, driving boundary area and working area, and adapt to the functional area characteristics of unstructured engineering machinery scenarios.

[0104] (4) The present invention provides a functional area-specific adaptive threshold calculation method based on the regional elevation standard deviation. It determines a specific denoising threshold for the driving area, driving boundary area and operation area respectively, replacing the global unified threshold, adapting to the differences in terrain fluctuations in each area, and improving the denoising accuracy and scene compatibility.

[0105] (5) This invention proposes a point cloud screening mechanism that combines multiple features such as elevation continuity, reflection intensity, and equipment shape to accurately distinguish useful terrain points from noise such as engineering machinery equipment points and flying points, thus solving the problems of incomplete identification of low-reflection equipment components and accidental deletion of useful terrain points.

[0106] (6) This invention proposes an optimization strategy that combines connected domain purification and functional region reverse verification to eliminate isolated misjudgment points, enhance the continuity and purity of functional regions, and ensure the engineering practicality of the noise reduction results. Attached Figure Description

[0107] Figure 1 This is a flowchart illustrating the point cloud denoising method for unstructured engineering machinery scenes in Embodiment 1 of the present invention.

[0108] Figure 2 This is a schematic diagram of the point cloud denoising system for unstructured engineering machinery scenes in Embodiment 2 of the present invention. Detailed Implementation

[0109] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0110] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0111] Example 1

[0112] like Figure 1 As shown, this invention proposes a point cloud denoising method for unstructured engineering machinery scenes, including: S1, point cloud preprocessing under preliminary noise filtering; S2, function-oriented dynamic sub-region division; S3, adaptive calculation of function region-specific thresholds; S4, point cloud screening with multi-feature collaboration; S5, connectivity optimization and result output.

[0113] In one specific implementation of this embodiment, step S1 includes the following steps:

[0114] S11. Collect raw point cloud data of unstructured engineering machinery scenes and convert it into PCD format (including three-dimensional coordinates (X,Y,Z) and reflection intensity).

[0115] S12. Perform voxel downsampling (preserve structural features while reducing data volume).

[0116] S13. Calculate the number of K nearest neighbors for each point, and remove isolated points with less than 5 neighbors (determined as isolated noise points) to obtain the preprocessed point cloud.

[0117] In one specific implementation of this embodiment, step S2 includes the following steps:

[0118] S21. Basic grid feature extraction;

[0119] S22. Based on the threshold values ​​of scene functional characteristics, three types of functional areas are initially divided;

[0120] S23. Optimize the continuity of functional areas through mesh refinement inspection, regional continuity constraints, and material feature assistance;

[0121] S24. The functional areas are finally determined, and three types of functional areas are output: driving area, driving boundary area, and work area.

[0122] In one specific implementation of this embodiment, step S21 includes the following steps:

[0123] S211. For the preprocessed point cloud in step S12, perform basic mesh division in the XY plane space outside the elevation Z coordinate dimension, based on the point cloud density and scene functional requirements.

[0124] S212, Associate features for each grid computing function:

[0125] (1) Road surface smoothness σ L Standard deviation of elevation of points within the grid (a characteristic indicator reflecting the flatness of the driving area for construction machinery);

[0126] (2) Boundary elevation difference ∆H: "maximum elevation - minimum elevation" within the grid (a characteristic index reflecting the ability of the driving boundary area to vertically define the driving range);

[0127] (3) Functional gradient proportion G F The percentage of points within the grid that have a slope greater than or equal to the slope threshold of the steep boundary area (an indicator that distinguishes between the steep characteristics of the driving boundary area and the gentle slope characteristics of the work area).

[0128] In one specific implementation of this embodiment, step S22 includes the following steps:

[0129] S221, When the road surface smoothness σ of a certain grid L The requirements for the smoothness of the driving of construction machinery are met and its functional gradient ratio G F If the grid is below the preset threshold for distinguishing significantly steep structures (i.e., there are no significantly steep structures), then the grid is initially designated as a driving area.

[0130] S222. When the boundary drop ∆H of a certain grid satisfies the vertical definition requirement of the driving range and its functional gradient ratio G F If the preset threshold for distinguishing the steepness of the boundary is reached (i.e., it meets the steepness characteristics of boundaries such as road edges), then the grid is initially defined as the driving boundary area.

[0131] S223. The remaining grids that do not conform to the above two categories (including complex terrains such as slopes, stockpiles, and pits required for the operation) are initially designated as the operation area.

[0132] In one specific implementation of this embodiment, step S23 includes the following steps:

[0133] S231. Mesh Refinement Verification: For the initially delineated area edge mesh, a secondary mesh division verification is performed using a size half that of the basic mesh. If more than half of the meshes do not meet the initial area delineation conditions in S22, then the mesh is designated as a mixed mesh and assigned to the work area (to avoid mixing of functional areas).

[0134] S232, Regional Continuity Constraint: The driving area must satisfy the condition that "it and at least three consecutive adjacent grids are candidate driving areas" to ensure continuous vehicle passage; the driving boundary area must satisfy the condition that "at least three consecutive adjacent grids along the edge of the driving area are candidate driving boundary areas" to ensure that the boundary completely defines the driving range.

[0135] S233, Material characteristic auxiliary inspection: More than 80% of the point reflection intensity of the grid in the driving area and driving boundary area needs to be within the R1 range (R1 corresponds to the reflection intensity range of road paving / soil and other terrain materials).

[0136] In one specific implementation of this embodiment, step S24 includes the following steps:

[0137] S241, Output Driving Area: When outputting the corresponding area, the road surface flatness σ must be satisfied. L ≤10~15cm (standard deviation of elevation of points within the basic grid), functional gradient percentage G F <30%~40% (no significant steep structure), and at least 3 consecutive adjacent grids of itself and its surroundings are candidate driving areas, while 80%~90% of the point reflection intensity within the grid is in a flat area of ​​the R1 interval corresponding to the road surface material.

[0138] S242, Output driving boundary area: When outputting the corresponding area, the boundary elevation difference ∆H ≥ 0.5~0.8m (the maximum and minimum elevation difference within the basic grid) and the functional gradient ratio G F ≥70%~80% (meets the characteristics of steep boundaries), and at least 3 consecutive adjacent grids along the edge of the driving area are candidate driving boundary areas, while 80%~90% of the point reflection intensity within the grid is in the defined area of ​​the R1 interval corresponding to the curb-type boundary material;

[0139] S243, Output work area: When outputting the corresponding area, it meets the requirements of the work operation area outside the driving area and driving boundary area (such as the excavation area, the stockpiling area, etc.), covering areas with complex terrain such as slopes, stockpiles, and pits.

[0140] In other embodiments, the functional area division in step S2 can replace the complex calculation of grid division with an "intuitive division method based on terrain slope and elevation," directly dividing according to slope and elevation intervals: the driving area is defined as a slope ≤ 5° and elevation fluctuation ≤ the maximum allowable value that meets the driving smoothness requirements; the driving boundary area is selected as a slope ≥ 50° and elevation difference ≥ the minimum critical value that meets the boundary drop characteristics; the remaining area is automatically assigned to the work area. This solution intuitively distinguishes areas through "slope + elevation fluctuation," eliminating the need for grid division and making it suitable for scenarios with uneven point cloud density.

[0141] In one specific implementation of this embodiment, step S3 includes the following steps:

[0142] S31. Based on the principle of "functional representativeness", preliminary screening is conducted in each functional area using neighborhood features or quantile features, followed by fine screening using statistical features such as median, standard deviation or fluctuation value, and a set of topographic feature benchmarks representing the inherent topographic morphology is extracted through a secondary screening method.

[0143] S32. Calculate the elevation standard deviation σ of the set of topographic feature benchmark points for each functional area (the elevation standard deviation σ reflects the "reasonable range of topographic fluctuation" of the corresponding area) and quantify the elevation fluctuation of the functional area.

[0144] S33. Combining the elevation standard deviation σ with the functional area's tolerance for anomalous point clouds L t (Quantized L in the range of 2.5 to 4.5) t (The lower the value, the lower the tolerance of the functional area to abnormal point clouds), generating adaptive denoising thresholds for each functional area.

[0145] In one specific implementation of this embodiment, step S31 includes the following steps:

[0146] S311. Extract the benchmark point set for terrain features in the driving area: First, randomly select m points from the driving area. d There are 10 candidate terrain points with a preset neighborhood radius r. d Perform a neighborhood search on the candidate terrain points in the driving area and calculate the standard deviation σ of the neighborhood elevation of each candidate point. h Then by preserving σ h ≤σ ht (σ) ht Candidate points are selected based on the elevation standard deviation threshold corresponding to the flatness of the driving area, resulting in a preliminary set of selected points; then, the elevation median μ is calculated for the preliminary set of selected points. h and elevation standard deviation σ hp By preserving the elevation at μ h ±σ hp The points within the interval are carefully selected to form the benchmark point set of terrain features of the driving area.

[0147] S312. Extract the benchmark point set for terrain features in the driving boundary area: First, randomly select m points from the driving boundary area. db There are 10 candidate terrain points with a preset neighborhood radius r. db Perform a neighborhood search on candidate terrain points in the driving boundary area, calculate the neighborhood slope p of each candidate point; then retain the slope p ≥ p t (p) t Candidate points are selected based on the slope threshold corresponding to the steep boundary characteristics, resulting in a preliminary set of selected points; then, the median slope μ is calculated for the preliminary set of selected points. p and slope standard deviation σ p Preserve slope at μ p ±σ p The points within the interval are carefully selected to form the final set of benchmark points for terrain features in the driving boundary area.

[0148] S313. Extract the benchmark point set for terrain features in the work area: First, randomly select m points from the work area. w10 candidate terrain points were selected, and the 25th percentile Q of the work area elevation was calculated. 25 With the 75th percentile Q 75 Then, by preserving the elevation in Q... 25 ~Q 75 Candidate points within the interval are used to obtain a preliminary set of selected points; then, a preset neighborhood radius r is used... w Perform a neighborhood search on the initially selected points and calculate the neighborhood elevation fluctuation ΔH for each point. n (ΔH) n (The difference between the maximum and minimum elevations in the neighborhood), by retaining ΔH n ≤ΔH nt (ΔH) nt The points corresponding to the elevation fluctuation threshold of the work area terrain are finely screened to form the benchmark point set of the work area terrain features.

[0149] In one specific implementation of this embodiment, step S32 includes the following steps:

[0150] S321. Calculate the elevation standard deviation σ1 of the benchmark point set for terrain features in the driving area. σ1 is used to quantify the allowable range of small undulations in the road surface.

[0151] S322. Calculate the elevation standard deviation σ2 of the benchmark point set of terrain features in the driving boundary area. σ2 is used to quantify the normal fluctuation range of the boundary structure.

[0152] S323. Calculate the elevation standard deviation σ3 of the benchmark point set for the terrain features of the work area. σ3 is used to quantify the reasonable range of variation of the terrain.

[0153] In one specific implementation of this embodiment, step S33 includes the following steps:

[0154] S331, Noise Reduction Threshold for the Driving Area: Due to the extremely high requirements for terrain flatness and traffic stability in this functional area, and the lowest tolerance for anomalous point clouds, the L value for this functional area is... t The value is then taken within the range of 2.5 to 3.0, and the calculated σ1 is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T1 for the driving area;

[0155] S332, Noise Reduction Threshold for Generating Driving Boundary Region: Since this functional region needs to filter out interference points while preserving steep boundary features, its tolerance for anomalous point clouds is at a moderate level. The L value for this functional region... t The value is then taken within the range of 3.0 to 4.0, and the calculated σ² is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T2 for the driving boundary region;

[0156] S333, Noise Reduction Threshold for the Generated Work Area: Since this functional area itself has elevation variations, it needs to retain terrain details relevant to the work. It has the highest tolerance for anomalous point clouds. The L value for this functional area is... t The value is then taken within the range of 4.0 to 4.5, and the calculated σ3 is then compared with the L of the functional region. t Multiply by the product to obtain the noise reduction threshold T3 for the work area.

[0157] In other embodiments, the calculation of each specific threshold in step S3 can be replaced by the standard deviation calculation using a "functional area experience threshold library". Basic thresholds for the driving area, driving boundary area, and work area are preset, and then dynamically fine-tuned based on real-time point cloud density (the threshold is reduced when the density is high to avoid over-screening; the threshold is increased when the density is low to avoid noise leakage). The functional area experience threshold library can be updated using measured data from engineering machinery scenarios, making it simple to operate and adaptable to common unstructured scenarios.

[0158] In one specific implementation of this embodiment, step S4 includes the following steps:

[0159] S41. Determine the region to which the point belongs based on its coordinates, and call the corresponding noise reduction threshold (T1, T2, T3) for the functional region.

[0160] S42. Terrain continuity based on regional threshold test points;

[0161] S43. Clustering of reflection intensity to distinguish between "terrain material" and "engineering machinery and equipment material";

[0162] S44. For "equipment components with blurred reflection intensity", conduct equipment integrity verification.

[0163] In one specific implementation of this embodiment, step S42 includes the following steps:

[0164] S421. Based on the point cloud density, select an appropriate neighborhood size, construct the neighborhood of the points, and calculate the elevation difference ∆H between the neighborhood points and the target point. p ;

[0165] S422. For the target point, calculate the compliance rate of elevation difference (A1) and gradient consistency (A2). If A1 meets the preset requirement for terrain continuity in the functional area and A2 meets the preset requirement for terrain trend in the functional area, then the target point is determined to meet the "terrain continuity" requirement. The compliance rate of elevation difference (A1) is calculated as follows: within the neighborhood, "|∆H p |≤Regional threshold” point percentage; Gradient consistency A2: percentage of neighboring points whose “elevation gradient direction (ascending / descending)” is consistent with the gradient direction of the target point.

[0166] In one specific implementation of this embodiment, the proportion of functional area terrain continuity requirement is preset to 70%, and the proportion of functional area terrain trend requirement is preset to 80%. That is, if the two indicators of elevation difference compliance ratio A1 and gradient consistency A2 simultaneously meet the requirements of A1≥70% and A2≥80%, the target point is determined to meet the terrain continuity.

[0167] In one specific implementation of this embodiment, step S43 includes the following steps:

[0168] S431. K-means algorithm is used to cluster the terrain reflection intensity range R1 (corresponding to terrain materials such as rock, soil, and road surface) and equipment reflection intensity range R2 (corresponding to engineering machinery materials such as metal and rubber).

[0169] S432. If the reflection intensity of a point is within the R1 interval and passes the terrain continuity verification in S422, it is determined to be a "useful terrain point" and retained.

[0170] S433. If the reflection intensity of a point is within the R2 range, it is determined to be an engineering machinery equipment point and is eliminated.

[0171] S434. If the reflection intensity of a point is between R1 and R2 (a fuzzy range, such as rubber tires, rusty metal, or non-metallic obstacles), and it is located in the work area (a high-frequency area for construction machinery and equipment), then proceed to "Equipment Integrity Verification" in S44.

[0172] In one specific implementation of this embodiment, step S44 includes the following steps:

[0173] S441. Seed point selection: Select points from S434 that meet the following criteria: "reflection intensity matches the characteristics of the core material of the equipment, the number of similar points in the neighborhood meets the continuity of the equipment structure, and the elevation difference exceeds the range of terrain fluctuations in the work area" as core seed points of the equipment.

[0174] S442, Regional growth, expands according to the dynamic threshold of "spatial distance ≤ equipment component width (adjusted according to equipment model and point cloud density)" and "reflection intensity fluctuation ≤ equipment material difference dynamic threshold" to cover the entire equipment;

[0175] S443. Morphological verification: The growth point set must conform to the size range (length and height are dynamically adjusted according to the equipment type) and geometric characteristics (such as tire cylindricity error, which are adjusted according to the equipment precision requirements) of the engineering machinery equipment. If the verification is passed, the entire equipment point set will be removed.

[0176] In other embodiments, the multi-feature screening in step S4 can replace reflection intensity analysis with a "combination of normal vector and spatial distribution method," which identifies objects by utilizing the differences in normal vector direction and spatial distribution: the normal vector in the driving area is close to horizontal (the angle between neighboring normal vectors tends to be in the horizontal angle range), the normal vector in the driving boundary area is close to vertical (the angle between neighboring normal vectors tends to be in the vertical angle range), and the normal vectors of engineering machinery equipment points are disordered (the angle between neighboring normal vectors has no fixed pattern and is discretely distributed). By calculating the angle between the normal vectors of each point and its neighboring points, terrain and engineering machinery equipment can be quickly distinguished without relying on point cloud reflection intensity.

[0177] In one specific implementation of this embodiment, step S5 includes the following steps:

[0178] S51. Connectivity purification: Perform connectivity analysis on the point cloud retained in S4, and remove isolated points remaining after filtering. When the number of connected points is lower than the connectivity purification threshold C... jt When the connected component is determined to be noise, it is deleted entirely, where the connected component purification determination threshold C is... jt Method: First, quickly obtain the average number of points in the neighborhood of a single point. Randomly select 10-15 points from the set of terrain feature reference points of the functional area. Perform a fixed-radius neighborhood search for each point, count the number of points in the neighborhood of each terrain feature reference point, and calculate the average number of points N in the neighborhood. n Then, a correction coefficient is assigned to the average number of neighborhood points according to the functional area characteristics, and the average number of neighborhood points N is adjusted accordingly. n Correction coefficient C for adaptation region characteristics c (Driving Area C) c =0.3, Driving boundary area C c =0.4, Work Area C c =0.6) multiplied (C before rounding) jt =N n ×C c Then, round down to obtain the final connected component purification threshold C. jt ;

[0179] S52. Functional area reverse inspection: For each functional area, if the proportion of equipment points exceeds a certain fault tolerance ratio, call the exclusive noise reduction threshold of the area to re-screen and remove elevation anomalies to ensure that there are no equipment point cloud residues in the core area.

[0180] S53. Output Results: Outputs denoised point clouds of three functional areas, supporting 3D visualization.

[0181] In other embodiments, the connectivity optimization in step S5 can be replaced by a "regional voting correction mechanism" instead of the complex algorithm based on connected component analysis and reverse correction. This mechanism performs "neighborhood voting" on the classification result of each point: if a point is initially identified as a driving area, but a majority of its surrounding neighboring points belong to the work area, then its classification is corrected to work area; conversely, the same applies. The regional voting correction mechanism corrects edge misjudgments through majority voting, is logically simple and computationally efficient, and is particularly suitable for real-time operation scenarios of construction machinery.

[0182] Example 2

[0183] like Figure 2 As shown, based on the same inventive concept as Embodiment 1, this embodiment introduces a point cloud denoising system for unstructured engineering machinery scenes, used to implement the point cloud denoising method for unstructured engineering machinery scenes in Embodiment 1, including:

[0184] The point cloud preprocessing module is used to perform preliminary filtering on the raw point cloud of unstructured engineering machinery scenes, remove isolated points and reduce data redundancy to obtain a preprocessed point cloud.

[0185] The functional area division module is used to divide the scene into three major functional areas: driving area, driving boundary area, and work area based on the functional association features of point cloud.

[0186] A dedicated threshold calculation module is used to dynamically generate a noise reduction threshold for each functional area based on the terrain fluctuation characteristics.

[0187] The multi-feature point cloud filtering module is used to integrate features such as functional area threshold, elevation continuity, and reflection intensity to accurately distinguish useful terrain points from engineering machinery equipment points.

[0188] The connectivity optimization and output module is used to optimize the connectivity of the filtered point cloud and perform reverse verification, ultimately outputting a high-precision denoised point cloud.

[0189] In one specific implementation of this embodiment, the region division module includes:

[0190] The feature extraction unit is used to calculate functionally related features such as road surface smoothness, boundary drop, and functional gradient ratio from the basic grid.

[0191] The area determination unit is used to initially divide the driving area, driving boundary area, and work area based on functional characteristic thresholds;

[0192] The continuity optimization unit is used to achieve precise division and optimization of functional areas through mesh refinement verification, regional continuity constraints, and material feature assistance.

[0193] In one specific implementation of this embodiment, the multi-feature filtering module includes:

[0194] The threshold matching unit is used to match the point cloud with the corresponding dedicated denoising threshold according to its functional region.

[0195] The elevation verification unit is used to filter out useful points with continuous elevation based on the terrain continuity of the regional threshold verification points.

[0196] Material clustering unit is used to cluster reflection intensity, distinguish between terrain material and engineering machinery material, and initially screen useful terrain points and engineering machinery points;

[0197] The equipment verification unit is used to verify the integrity of suspected engineering machinery equipment points with blurred reflection intensity, thereby eliminating the equipment points as a whole.

[0198] In one specific implementation of this embodiment, the optimized output module includes:

[0199] The connected component purification unit is used to remove isolated points remaining after screening, ensuring the continuity of the point cloud;

[0200] The reverse verification unit is used to verify equipment residue in the core functional area to ensure that the equipment point cloud rejection rate meets the accuracy requirements.

[0201] The result output unit is used to output the denoised point cloud of the three functional areas and supports 3D visualization.

[0202] Example 3

[0203] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described point cloud denoising method for unstructured engineering machinery scenes.

[0204] Example 4

[0205] Based on the same inventive concept as other embodiments, this embodiment introduces a computer device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described point cloud denoising method for unstructured engineering machinery scenes.

[0206] Example 5

[0207] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described point cloud denoising method for unstructured engineering machinery scenes.

[0208] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0209] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0210] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0211] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0212] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other modifications under the guidance of the present invention, and these modifications are all within the protection scope of the present invention.

Claims

1. A point cloud denoising method for unstructured engineering machinery scenes, characterized in that, include: Collect raw point cloud data of unstructured scenes of construction machinery, preprocess the raw point cloud data, and obtain preprocessed point cloud data; Based on the functional association features of the preprocessed point cloud data, the scene is divided into functional areas, namely driving area, driving boundary area and operation area. Based on the terrain fluctuation characteristics of different functional areas, the denoising threshold of each functional area is dynamically generated. By integrating the noise reduction threshold, elevation continuity, and reflection intensity of each functional area, useful terrain points and engineering machinery equipment points are selected. The selected useful terrain points are retained, and the engineering machinery equipment points are removed to obtain the filtered point cloud. The selected point cloud is subjected to connectivity optimization and reverse verification to output the final denoised point cloud.

2. The point cloud denoising method for unstructured engineering machinery scenes according to claim 1, characterized in that: The raw point cloud data of the unstructured scene of the engineering machinery is collected, and the raw point cloud data is preprocessed to obtain preprocessed point cloud data, including: Collect raw point cloud data from unstructured scenes of construction machinery; Voxel downsampling is performed on the original point cloud data to obtain downsampled point cloud data; Calculate the number of K nearest neighbors for each point in the downsampled point cloud data; if the number of K nearest neighbors for a point is less than k, then the point is determined to be an isolated noise point; remove the points determined to be isolated noise points to obtain the preprocessed point cloud data.

3. The point cloud denoising method for unstructured engineering machinery scenes according to claim 1, characterized in that: Based on the functional association features of the preprocessed point cloud data, the scene is divided into functional regions, namely, a driving area, a driving boundary area, and a work area, including: Based on the preprocessed point cloud data, a basic grid is divided into the preprocessed point cloud data. Based on the division of the basic grid, the functional association features of each grid are calculated; the functional association features include road surface smoothness σ. L Boundary drop ∆H, functional gradient proportion G F ; Based on the functional association features of each grid, the driving area, driving boundary area and work area are initially delineated respectively; The initially defined driving area, driving boundary area, and work area are subjected to grid refinement inspection, regional continuity constraint, and material characteristic auxiliary inspection to obtain the final defined driving area, driving boundary area, and work area.

4. The point cloud denoising method for unstructured engineering machinery scenes according to claim 1, characterized in that: The method of dynamically generating denoising thresholds for each functional region based on the terrain fluctuation characteristics of the divided functional regions includes: Based on the principle of "functional representativeness", preliminary screening is carried out in each functional area through neighborhood features or quantile features, and then secondary screening is carried out through median, standard deviation or fluctuation value statistical features. The set of topographic feature benchmarks that represent the inherent topographic morphology is extracted in the secondary screening method. Calculate the elevation standard deviation σ of the set of topographic feature benchmarks for each functional area, and quantify the elevation fluctuation of the functional area. Combining the elevation standard deviation σ with the functional region's tolerance for anomalous point clouds L t Generate adaptive denoising thresholds for each functional area.

5. The point cloud denoising method for unstructured engineering machinery scenes according to claim 1, characterized in that: The denoising threshold, elevation continuity, and reflection intensity of each functional area are integrated to distinguish between useful terrain points and engineering machinery equipment points. Useful terrain points are retained, while engineering machinery equipment points are removed, resulting in a filtered point cloud, including: Determine the functional area to which the point belongs based on its coordinates, and then call the noise reduction threshold of the corresponding functional area. The terrain continuity of the test points based on the denoising threshold of each functional area; Clustering of reflection intensity distinguishes between terrain materials and engineering machinery materials; For points with blurred reflection intensity located within the work area, equipment integrity verification is performed; The ambiguity of the reflection intensity refers to the reflection intensity being between R1 and R2, where R1 is the terrain reflection intensity range and R2 is the equipment reflection intensity range.

6. The point cloud denoising method for unstructured engineering machinery scenes according to claim 1, characterized in that: The process of optimizing connectivity and performing reverse verification on the filtered point cloud to output the final denoised point cloud includes: Connectivity purification involves performing connectivity analysis on the filtered point cloud to remove isolated points remaining after filtering. When the number of connected points falls below the connectivity purification threshold C... jt When this happens, the connected component is determined to be noise and deleted entirely. Functional area reverse verification: For each functional area, if the proportion of equipment points in a certain functional area exceeds the set fault tolerance ratio, the noise reduction threshold of that functional area is called for re-screening to remove elevation anomalies and ensure that there are no equipment point cloud residues in the core functional areas. The output results in denoised point clouds for three functional regions.

7. A point cloud denoising system for unstructured engineering machinery scenes, characterized in that, The point cloud denoising method for unstructured engineering machinery scenes as described in any one of claims 1 to 6 includes: The point cloud preprocessing module is configured to collect raw point cloud data from unstructured scenes of engineering machinery, preprocess the raw point cloud data, and obtain preprocessed point cloud data. The functional area division module is configured to divide the scene into functional areas based on the functional association features of the preprocessed point cloud data, namely, the driving area, the driving boundary area, and the work area. A dedicated threshold calculation module is configured to dynamically generate denoising thresholds for each functional area based on the terrain fluctuation characteristics of the different functional areas. The multi-feature point cloud filtering module is configured to integrate the denoising threshold, elevation continuity and reflection intensity of each functional area to filter out useful terrain points and engineering machinery equipment points. The useful terrain points are retained and the engineering machinery equipment points are removed to obtain the filtered point cloud. The connectivity optimization and output module is configured to perform connectivity optimization and reverse verification on the filtered point cloud, and output the final denoised point cloud.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the point cloud denoising method for unstructured engineering machinery scenes as described in any one of claims 1 to 6.

9. A computer device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the point cloud denoising method for unstructured engineering machinery scenes as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the point cloud denoising method for unstructured engineering machinery scenes as described in any one of claims 1 to 6.

Images