Manifold-based construction environment point cloud denoising method and system for excavator

Abstract

The application discloses a manifolds-based excavator construction environment field point cloud denoising method and system, and belongs to the technical field of excavator environment perception. The application proposes an end-to-end adaptive denoising network based on point-by-point noise level in manifolds under unknown scenes and unknown noise types, extracts manifold features in a local manifold constructed by K-Nearest Neighbor (KNN), and inputs the manifold features to predict noise offset of each point under different noise levels through a residual network. According to a joint offset function, noise offsets of K neighbor points around each point in the manifold are fitted, and the fitted offset is raised through gradient until the noise point returns to a clean surface to complete denoising. The application can denoise irregular and disordered high-noise point clouds and retain detailed features, and solves the problem that detailed features are easily lost in point cloud data denoising in the prior art.

Description

Technical Field

[0001] This invention relates to the field of excavator environmental perception technology, and in particular to a manifold-based method and system for denoising point clouds in excavator construction environments. Background Technology

[0002] The statements in this section merely refer to the background art related to this invention and do not necessarily constitute prior art.

[0003] To ensure the safety of excavators in complex construction environments such as riverbanks, steep slopes, and depressions, and to continuously improve the accuracy and efficiency of construction operations, intelligent and autonomous operation has become the main development trend of excavators. Traditionally, excavators operate in complex environments and rely on lidar to perceive the surrounding environment in real time to ensure personnel safety. However, due to the influence of the construction environment and the sensor itself, the raw point cloud data collected by lidar inevitably contains noise and outliers.

[0004] Point cloud denoising plays an indispensable role in high-precision environmental perception under complex construction environments. The irregularity and disorder of point clouds make point cloud denoising more difficult than that of two-dimensional images. Traditional denoising methods rely heavily on geometric or noise priors, and when processing high-noise point clouds, they exhibit over-smoothing and loss of detailed features, which is extremely detrimental to environmental perception in complex excavator construction scenarios.

[0005] With the emergence of PointNet, the first deep learning framework to directly process point clouds, a large number of excellent point cloud denoising methods have been implemented. Early deep learning-based point cloud denoising methods typically denoised by predicting the offset between noisy points and potentially clean surfaces. However, these methods often produce two types of artifacts, which are caused by inaccurate offset estimation at high noise levels: overestimation of point cloud shrinkage and underestimation of outlier residue. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a manifold-based point cloud denoising method, system, electronic device, and computer-readable storage medium for excavator construction environments. Under unknown scenarios and unknown noise types, point cloud denoising is performed based on an end-to-end adaptive denoising network that measures the noise level at each point in the manifold.

[0007] In a first aspect, the present invention provides a method for denoising point clouds in the construction environment of excavators based on manifolds;

[0008] A manifold-based point cloud denoising method for excavator construction environments includes:

[0009] Acquire point cloud data of the excavator's construction environment, divide the point cloud data of the excavator's construction environment into two-dimensional curved manifolds, and construct a local manifold with each point in the two-dimensional curved manifold as the center.

[0010] Based on the two-dimensional curved manifold, multi-scale graph features and noise level scores are extracted for each point. Based on the local manifold, edge features and density features of the center point are extracted.

[0011] Based on the multi-scale plot features, noise level fraction, edge features, and density features of the center point in the local manifold, calculate the noise offset of all points in the local manifold;

[0012] Based on the noise offset, a joint offset function is constructed, and the noise offset of the local manifold center point is obtained by fitting the joint offset function. The denoised excavator construction environment field point cloud data is obtained through multiple rounds of iteration.

[0013] Furthermore, the extraction of multi-scale graph features and noise level scores for each point based on the two-dimensional curved manifold includes:

[0014] For each point, a dense stack of dynamic graph convolutions is constructed at different scales. A two-dimensional curved surface is input into the stack of dynamic graph convolutions to extract output features at different scales. The output features of the current layer are used as the input of the next layer.

[0015] The output features at different scales are concatenated to obtain the multi-scale map features of each point;

[0016] The multi-scale graph features of each point in the two-dimensional curved manifold are mapped to one dimension to obtain one-dimensional feature values; the one-dimensional feature values ​​are then normalized to obtain the noise level score of each point.

[0017] Furthermore, the extraction of edge features and density features of the center point based on the local manifold includes:

[0018] Calculate the relative coordinates of the center point of the local manifold and its K neighboring points. Based on the relative coordinates, extract the edge features of the center point using a multilayer perceptron.

[0019] Calculate the distances and average distances between the center point of the local manifold and its K neighboring points. Based on the average distances, obtain the density of the center point using a multilayer perceptron.

[0020] Furthermore, the noise offset of all points in the local manifold is expressed as:

[0021]

[0022] in, The predicted noise offset. For the residual network used to estimate the noise offset, The noise level fraction at the local manifold center point. Density characteristics at the local manifold center points Multi-scale plot features of local manifold centers These are the edge features of the local manifold center point.

[0023] Furthermore, the noise offset of the local manifold center point is obtained by fitting based on the joint offset function, and the denoised excavator construction environment point cloud data is obtained through multiple rounds of iteration, including:

[0024] Based on the joint offset function, the mean noise offset of multiple neighboring points of the local manifold center point is obtained;

[0025] The mean of the noise offsets of adjacent points is used to replace the noise offset of the center point. By performing gradient ascent on the noise offsets of the center points of all local manifolds, the position of each point in the two-dimensional curved manifold is updated, and the denoised excavator construction environment field point cloud data is obtained.

[0026] Furthermore, the joint offset function is

[0027]

[0028] in, This represents the mean noise offset of the K neighboring points of the local manifold center. These are the K neighboring points of the center point. The noise offset of the K neighboring points of the center point.

[0029] Furthermore, it also includes:

[0030] Calculate the noise offset of each point in the local manifold to the ground truth clean point cloud, and construct a loss function to make the noise offset consistent with the ground truth noise offset.

[0031] Preferably, the loss function is

[0032]

[0033] in, This represents the loss between the predicted and actual offsets at each point in the local manifold. Concentrated in the three-dimensional point cloud space The distribution Point True value cleaning point cloud to the ground The noise offset.

[0034] Secondly, the present invention provides a point cloud denoising system for excavator construction environment based on manifold;

[0035] A manifold-based point cloud denoising system for excavator construction environments includes:

[0036] The local manifold construction module is used to acquire point cloud data of the excavator construction environment, divide the point cloud data of the excavator construction environment into two-dimensional surface manifolds, and construct a local manifold with each point in the two-dimensional surface manifold as the center.

[0037] The manifold feature extraction module is used to extract multi-scale graph features and noise level scores for each point based on a two-dimensional curved manifold, and to extract edge features and density features for the center point based on a local manifold.

[0038] The manifold adaptive denoising module is used to calculate the noise offset of all points in the local manifold based on the multi-scale map features, noise level scores, edge features, and density features of the center point in the local manifold. Based on the noise offset, a joint offset function is constructed, and the noise offset of the center point of the local manifold is obtained by fitting the joint offset function. The denoised excavator construction environment field point cloud data is obtained through multiple rounds of iteration.

[0039] Thirdly, the present invention provides an electronic device;

[0040] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, the computer instructions perform the steps of the above-described manifold-based point cloud denoising method for excavator construction environment.

[0041] Fourthly, the present invention provides a computer-readable storage medium;

[0042] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described manifold-based point cloud denoising method for excavator construction environment.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. The technical solution provided by this invention, given a noisy point cloud as input, first uses the KMeans point cloud clustering and segmentation algorithm to transform the point cloud into a set of two-dimensional curved manifolds, extracting multi-scale map features and noise level scores for each point. A local manifold is constructed centered on each point in the curved manifold, and the edge features and density features of the center point are extracted. Using the four manifold features as input, a residual network is used to predict the noise offset of points with different noise levels in the local manifold. Our network predicts and learns the noise offset point-by-point in the local manifold, rather than the point cloud itself, thereby reducing the potential space of the 3D geometric representation, improving the network's prediction accuracy and generalization ability for noise levels of point clouds with different shapes, ensuring that downstream tasks denoise while preserving clean point clouds, and retaining the original point cloud information to the greatest extent.

[0045] 2. The technical solution provided by the present invention extracts the multi-scale graph features, noise level fraction, edge features and density features of the local manifold center point. These four features describe the noisy local manifold from different perspectives to ensure accurate estimation of noise offset.

[0046] 3. The technical solution provided by this invention uses a joint offset function, that is, the mean of the noise offset of adjacent points, to replace the displacement of the current local manifold center point. Finally, the noise is removed by multiple iterations of all local manifold center points. Unlike previous methods that predict the noise offset based on a single point, this method enhances robustness and reduces estimation bias.

[0047] 4. The technical solution provided by this invention calculates the noise offset of each point in the local manifold to the ground true clean point cloud, and constructs a loss function to make the predicted noise offset consistent with the ground true noise offset.

[0048] Unlike previous methods that only considered the loss at a single point, this invention proposes replacing the center point loss with the loss of the entire local manifold near the center point. This makes the method of this invention more robust to training estimation for noise offsets. (See attached figures.)

[0049] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0050] Figure 1 This is a schematic diagram of the adaptive denoising network structure provided in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of multi-scale graph feature extraction provided in an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of manifold edge feature extraction provided in an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of the adaptive noise reduction process provided in an embodiment of the present invention;

[0054] Figure 5 A visual comparison diagram of the denoising results of the method of this embodiment and other algorithms on the PU-Net dataset affected by 1%, 2%, and 3% Gaussian noise, provided for the embodiments of the present invention;

[0055] Figure 6 A visual comparison diagram of the denoising results of the method of this embodiment and other algorithms on the PU-Net dataset subject to 10%, 15%, and 20% discrete noise interference, provided for the embodiments of the present invention;

[0056] Figure 7Visual comparison diagram of the method of this embodiment and other algorithms on the PU-Net dataset denoising algorithms with uniform noise interference of 5%, 10%, and 15% provided for the embodiments of the present invention;

[0057] Figure 8 Visual comparison diagram of the method of this embodiment and other algorithms on the PU-Net dataset with 2.5%, 3.75%, and 5% Lapses noise interference, provided for the embodiments of the present invention;

[0058] Figure 9 A visual comparison diagram of the method of this embodiment of the invention and other algorithms for denoising the Paris-rue-Madame dataset;

[0059] Figure 10 A visual comparison diagram of the method of this embodiment and other algorithms on the KITTI autonomous driving dataset denoising method, provided for this embodiment of the invention.

[0060] Figure 11 Example figures illustrating the quantitative comparison results of the method of this embodiment with other algorithms for denoising under Gaussian noise, provided in this embodiment of the invention;

[0061] Figure 12 Example figures illustrating the quantitative comparison results of the method of this embodiment with other algorithms in denoising discrete noise, provided for the purposes of this invention.

[0062] Figure 13 Example figures illustrating the quantitative comparison results of the method of this embodiment with other algorithms in denoising uniform noise, provided for the purposes of this invention.

[0063] Figure 14 Example diagram showing the quantitative comparison results of the method of this embodiment and other algorithms for denoising under Laplace noise, provided for the embodiments of the present invention. Detailed Implementation

[0064] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0065] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0066] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0067] Example 1

[0068] Existing point cloud denoising methods typically denoise by predicting the offset between noise points and the potentially clean surface, which can lead to excessive point cloud shrinkage or outlier retention. Therefore, as... Figure 1 As shown, this invention provides a manifold-based point cloud denoising method for excavator construction environments, transforming the point cloud denoising task into the reconstruction of curved manifolds.

[0069] Next, combined Figures 1-14 This embodiment provides a detailed description of the manifold-based point cloud denoising method for excavator construction environments. The manifold-based point cloud denoising method for excavator construction environments includes the following steps:

[0070] S1. Obtain point cloud data of the excavator construction environment. Segment the point cloud data of the excavator construction environment into two-dimensional surface manifolds. Construct a local manifold with each point in the two-dimensional surface manifold as the center. Specifically, use the point cloud clustering and segmentation algorithm K-means to transform the point cloud data of the excavator construction environment into a set of two-dimensional surface manifolds. Then, using each point in the surface manifold as the center, use the KNN (K-Nearest Neighbor) algorithm to construct a local manifold containing K nearby points centered at each point in the surface manifold.

[0071] S2. Based on the two-dimensional curved manifold, extract multi-scale graph features and noise level scores for each point. Based on the local manifold, extract edge features and density features for the center point. These four features describe the noisy local manifold from different perspectives. Specific steps include:

[0072] S201. Construct a four-layer dense dynamic graph convolution stack for each point at different scales. Input the two-dimensional curved surface manifold into the dynamic graph convolution stack to extract output features at different scales; wherein, the output features of the current layer become the input of the next layer. Specifically, as follows... Figure 2As shown, from a set of manifolds containing noise Point-by-point learning of multi-scale graph features; for each point, a four-layer dense Dynamic Graph Convolutional (DGCNN) stack is constructed at different scales using the KNN algorithm, with the output features of the current layer... This is the input for the next layer. Feature calculation is as follows:

[0073]

[0074] in, This represents the output features of the current layer. Indicates the current floor number, with values ​​ranging from 1 to 4. For DGCNN, For a noisy two-dimensional curved manifold, ReLU represents the maximum activation function, max is the max pooling function, and N(i) represents... The neighborhood, For a densely connected multilayer perceptron (MLP), θ represents the learnable weight and bias parameter matrix.

[0075] To extract features with richer contextual information, multiple DGCNNs are densely connected:

[0076]

[0077] in, This indicates that features are output layer by layer. This indicates a multi-layered dense link.

[0078] To obtain information at different scales, the number of neighbors aggregated in KNN is varied, and the above operation is repeated to extract features at different scales. .

[0079] Finally, the features from multiple scales are concatenated to obtain multi-scale map features:

[0080]

[0081] in, Represents multi-scale map features. This represents the output features of densely linked multi-layer graph convolutions at different scales. This represents feature links at different scales.

[0082] S202. Map the multi-scale graph features of each point in the two-dimensional curved manifold to one dimension to obtain one-dimensional feature values; normalize the one-dimensional feature values ​​to obtain the noise level score of each point.

[0083] Specifically, the multi-scale graph features of point-by-point two-dimensional curved manifolds. Used as input to calculate noise level score .

[0084] First, the eigenvalues ​​are mapped to one dimension using an MLP, and then the mapped eigenvalues ​​are normalized, as shown below:

[0085]

[0086] in, For point-by-point noise level fractions, From arrive Feature mapping, Feature normalization can improve training speed and model accuracy.

[0087] S203. Calculate the relative coordinates of the center point of the local manifold and its adjacent points, and obtain the edge features of the center point based on the relative coordinates.

[0088] Specifically, such as Figure 3 As shown, calculate the relative coordinates between the local manifold center point and its neighboring points. Then, the edge features of the center point are obtained based on the relative coordinates MLP:

[0089]

[0090] in, The edge features of the center point of the local manifold The center point of the local manifold These are the adjacent points of the local manifold center.

[0091] S204. Calculate the distance between the center point of the local manifold and its adjacent points, and obtain the density characteristics of the center point based on the distance between the center point of the local manifold and its adjacent points.

[0092] Specifically, firstly, KNN is used to calculate the local manifold center point and its nearest neighbor. Neighboring points The distance D between them is calculated as follows:

[0093]

[0094]

[0095]

[0096] in, For K nearest neighbors, The number of adjacent points, The distance between the center point and its adjacent points. The density at the center point.

[0097] S3. Calculate the noise offset of all points in the local manifold based on the multi-scale graph features, noise level fraction, edge features, and density features of the center point in the local manifold.

[0098] Specifically, to ensure that the noise point moves to a clean underlying surface, the noise offset is calculated as follows:

[0099]

[0100] in, The predicted noise offset. For the residual network used to estimate the noise offset, The noise level score at the center point. Density characteristics at the center point Multi-scale map features centered on the central point Edge features of the center point.

[0101] S4. Based on the noise offset, construct a joint offset function, and obtain the noise offset of the local manifold center point by fitting the joint offset function. Obtain the denoised excavator construction environment point cloud data through multiple iterations.

[0102] To enhance robustness and reduce estimation bias, a joint offset function is used for denoising. The joint offset function is as follows:

[0103]

[0104] in, It is the noise offset of the K neighboring points of the center point. This is the average noise offset of the K neighboring points of the center point, i.e., the noise offset of the center point.

[0105] like Figure 4 As shown, point cloud denoising is equivalent to updating the position of each point by the gradient of the noise offset until it rises to the clean underlying surface:

[0106]

[0107] in, Indicates the number of rounds of gradient ascent. Indicates the first Step length, Point At the position after gradient ascent t times. Point The position after the gradient rises for t-1 times.

[0108] Furthermore, in order to make the network's predicted noise offset consistent with the actual ground noise offset, a loss function is constructed.

[0109] First, calculate the noise offset from each point in the local manifold to the ground truth point cloud:

[0110]

[0111] in, This represents the noise offset from the noise point to the ground truth point cloud. Indicates clean point clouds. express Mid-range The nearest point.

[0112] Then, construct the loss function:

[0113]

[0114]

[0115] in, For the K neighboring points near the center of the local manifold, adjacent points True value cleaning point cloud to the ground The noise offset. The final loss of the local manifold is the average loss of the K neighboring points near the center point.

[0116] To better verify the advancement of the method in this embodiment, experiments were conducted and qualitative and quantitative comparisons were made with existing denoising methods.

[0117] For quantitative testing, the PU-Net test set (20 shapes) was used. First, point clouds were obtained by sampling each grid at 10K and 50K resolutions. Then, the K-means algorithm was used to segment the point clouds into a set of manifolds. Finally, the performance of our model was evaluated using various noise models, including Gaussian, Laplacian, uniform, and discrete noise. The final denoising effect is shown below. Figures 11-14 As shown, Figure 11 For a quantitative comparison of denoising algorithms under Gaussian noise, CD is multiplied by P2M multiplied by , Figure 12 For a quantitative comparison of denoising algorithms under discrete noise, CD multiplied by P2M multiplied by , Figure 13 For quantitative comparison of denoising algorithms for uniform noise, CD is multiplied by P2M multiplied by , Figure 14 For a quantitative comparison of denoising algorithms under Laplace noise, CD is multiplied by P2M multiplied by .

[0118] The results of this embodiment outperform deep learning-based methods and traditional methods in most cases. Unlike previous work, this embodiment not only compares the results at low Gaussian noise levels with a standard deviation of 1%-3%, but also conducts detailed comparative experiments under conditions where other noise models are subjected to higher noise interference. The experimental results demonstrate that the performance remains significant as the noise level increases, proving that the manifold-based adaptive denoising network can learn more complex features and has stronger robustness in denoising at high noise levels.

[0119] For qualitative testing, in addition to the various synthetic noises mentioned above, this embodiment also used the Paris street dataset and the KITTI autonomous driving dataset for visual evaluation. These datasets were obtained from the real world using lasers. The denoising results of the synthetic noise are as follows: Figures 5-8 As shown, the proposed method in this embodiment, compared with competing algorithms in Gaussian noise, Laplace noise, uniform noise, and discrete noise, demonstrates that the results of this embodiment are clearer and more visually appealing. It is worth noting that, compared with other algorithms, whether deep learning or traditional algorithms, the method in this embodiment better preserves details during denoising and removes outliers more thoroughly. The denoising results for real-world noise are shown below. Figures 9-10 As shown, Figure 9 For denoising the Paris-rue-Madame street dataset, compared to Score, the method in this embodiment achieves more complete noise removal while preserving detailed features. Compared to DMR, edges, corners, and other detailed geometric features in this embodiment are better preprocessed and not overly smoothed. Figure 10 To denoise the KITTI dataset for autonomous driving, this embodiment compares the method used in this embodiment with the Score dataset. The results are consistent with those of the Paris-rue-Madame dataset, and the method in this embodiment achieves more perfect denoising.

[0120] Example 2

[0121] This embodiment discloses a manifold-based point cloud denoising system for excavator construction environments, including:

[0122] The local manifold construction module is used to acquire point cloud data of the excavator construction environment, divide the point cloud data of the excavator construction environment into two-dimensional surface manifolds, and construct a local manifold with each point in the two-dimensional surface manifold as the center.

[0123] The manifold feature extraction module is used to extract multi-scale graph features and noise level scores for each point based on a two-dimensional curved manifold, and to extract edge features and density features for the center point based on a local manifold.

[0124] The manifold adaptive denoising module is used to calculate the noise offset of all points in the local manifold based on the multi-scale map features, noise level scores, edge features, and density features of the center point. Based on the noise offset, a joint offset function is constructed, and the noise offset of the center point of the local manifold is obtained by fitting the joint offset function. The denoised excavator construction environment point cloud data is then obtained through multiple iterations. It should be noted that the aforementioned local manifold construction module, manifold feature extraction module, and manifold adaptive denoising module correspond to the steps in Embodiment 1. The examples and application scenarios implemented by these modules and their corresponding steps are the same, but not limited to the content disclosed in Embodiment 1. It should also be noted that these modules, as part of the system, can be executed in a computer system, such as a set of computer-executable instructions.

[0125] Example 3

[0126] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-mentioned manifold-based point cloud denoising method for excavator construction environment.

[0127] Example 4

[0128] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described manifold-based point cloud denoising method for excavator construction environment.

[0129] 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.

[0130] 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.

[0131] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed 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.

[0132] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0133] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A point cloud denoising method for excavator construction environment based on manifolds, characterized in that, include: Acquire point cloud data of the excavator's construction environment, divide the point cloud data of the excavator's construction environment into two-dimensional curved manifolds, and construct a local manifold with each point in the two-dimensional curved manifold as the center; Based on the two-dimensional curved manifold, multi-scale graph features and noise level scores are extracted for each point. Based on the local manifold, edge features and density features of the center point are extracted. Based on the multi-scale plot features, noise level fraction, edge features, and density features of the center point in the local manifold, calculate the noise offset of all points in the local manifold; Based on the noise offset, a joint offset function is constructed, and the noise offset of the local manifold center point is obtained by fitting the joint offset function. The denoised excavator construction environment field point cloud data is obtained through multiple rounds of iteration.

2. The point cloud denoising method for excavator construction environment based on manifold as described in claim 1, characterized in that, The extraction of multi-scale graph features and noise level scores for each point based on a two-dimensional curved manifold includes: For each point, a dense stack of dynamic graph convolutions is constructed at different scales. A two-dimensional curved surface is input into the stack of dynamic graph convolutions to extract output features at different scales. The output features of the current layer are used as the input of the next layer. The output features at different scales are concatenated to obtain the multi-scale map features of each point; The multi-scale graph features of each point in the two-dimensional curved manifold are mapped to one dimension to obtain one-dimensional feature values; the one-dimensional feature values ​​are then normalized to obtain the noise level score of each point.

3. The point cloud denoising method for excavator construction environment based on manifold as described in claim 1, characterized in that, The extraction of edge and density features of the center point based on the local manifold includes: Calculate the relative coordinates of the center point of the local manifold and its K neighboring points. Based on the relative coordinates, extract the edge features of the center point using a multilayer perceptron. Calculate the distances and average distances between the center point of the local manifold and its K neighboring points. Based on the average distances, obtain the density of the center point using a multilayer perceptron.

4. The point cloud denoising method for excavator construction environment based on manifold as described in claim 1, characterized in that, The noise offset of all points in the local manifold is expressed as: in, The predicted noise offset. For the residual network used to estimate the noise offset, The noise level fraction at the local manifold center point. Density characteristics at the local manifold center points Multi-scale plot features of local manifold centers These are the edge features of the local manifold center point.

5. The point cloud denoising method for excavator construction environment based on manifold as described in claim 1, characterized in that, The noise offset of the local manifold center point is obtained by fitting based on the joint offset function, and the denoised excavator construction environment point cloud data is obtained through multiple rounds of iteration, including: Based on the joint offset function, the mean noise offset of multiple neighboring points of the phase local manifold center point is obtained; The mean of the noise offset of adjacent points is used to replace the noise offset of the center point. By performing gradient ascent on the noise offset of the center point of all local manifolds, the position of each point in the two-dimensional curved manifold is updated to obtain the denoised point cloud data of the excavator construction environment. Furthermore, the joint offset function is in, This represents the mean noise offset of the K neighboring points of the local manifold center. These are the K neighboring points of the center point. The noise offset of the K neighboring points of the center point.

6. The point cloud denoising method for excavator construction environment based on manifold as described in claim 1, characterized in that, Calculate the noise offset of each point in the local manifold to the ground truth clean point cloud, and construct a loss function to make the noise offset consistent with the ground truth noise offset.

7. The point cloud denoising method for excavator construction environment based on manifold as described in claim 6, characterized in that, The loss function is: in, This represents the loss between the predicted and actual offsets at each point in the local manifold. For three-dimensional point cloud space concentrated The distribution, Point True value cleaning point cloud to the ground The noise offset.

8. A point cloud denoising system for excavator construction environment based on manifolds, characterized in that, include: The local manifold construction module is used to acquire point cloud data of the excavator construction environment, divide the point cloud data of the excavator construction environment into two-dimensional surface manifolds, and construct a local manifold with each point in the two-dimensional surface manifold as the center. The manifold feature extraction module is used to extract multi-scale graph features and noise level scores for each point based on a two-dimensional curved manifold, and to extract edge features and density features for the center point based on a local manifold. The manifold adaptive denoising module is used to calculate the noise offset of all points in the local manifold based on the multi-scale map features, noise level scores, edge features, and density features of the center point in the local manifold. Based on the noise offset, a joint offset function is constructed, and the noise offset of the center point of the local manifold is obtained by fitting the joint offset function. The denoised excavator construction environment field point cloud data is obtained through multiple rounds of iteration.

9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the steps of the manifold-based point cloud denoising method for excavator construction environment as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps of the manifold-based point cloud denoising method for excavator construction environment as described in any one of claims 1-7.

Images