Three-dimensional point cloud high-density reconstruction method based on gaussian process regression and related device

Abstract

The application discloses a three-dimensional point cloud high-density reconstruction method based on Gaussian process regression and related devices, comprising: preprocessing the original point cloud data set by using a random sample consensus algorithm to obtain point cloud data with ground points removed; using a region growing method to extract line segment features from the point cloud data with ground points removed to obtain point cloud data with longitudinal line bundles separated; performing densification processing on the point cloud data with longitudinal line bundles separated to obtain newly constructed point cloud data; and using the newly constructed point cloud data to build a laser odometer to generate a point cloud relative pose map and a point cloud map. Through preprocessing of the point cloud, longitudinal line feature segmentation and longitudinal Gaussian process processing, the effect of enriching the density of the longitudinal sparse point cloud is achieved, the key technology of high-precision positioning and mapping of the sparse point cloud laser is broken through, the efficient and accurate scanning matching and mapping task is completed, and higher-precision navigation and obstacle avoidance are realized.

Description

Technical Field

[0001] This application belongs to the field of lidar technology and relates to a method and related apparatus for high-density reconstruction of three-dimensional point clouds based on Gaussian process regression. Background Technology

[0002] LiDAR, also known as laser radar, is a type of optical radar. It measures information such as the target's distance, speed, and vibration state by emitting laser light towards the target and receiving the reflected signal. With the advancement of technology, lidar has been widely used in many fields, including topographic mapping, atmospheric physics research, environmental science, astronomical observation, and autonomous vehicles.

[0003] Over the past few decades, with the rapid development of laser technology and computing power, the resolution and measurement speed of lidar have been greatly improved. Various types of lidar, such as fixed-wavelength, modulated-wavelength, pulsed, and phase-difference lidar, have demonstrated their respective advantages in specific applications.

[0004] While modern lidar demonstrates superior performance in terms of measurement accuracy and range, its high cost remains a significant challenge. Furthermore, environmental factors such as rain, fog, and dust can also affect lidar measurement results. To overcome these challenges, researchers are actively exploring new laser sources, detector technologies, and advanced signal processing methods.

[0005] While lidar technology, especially entry-level or low-cost devices, has provided numerous conveniences for many applications, it still faces some technical and practical challenges. Specifically, these devices are often significantly limited in terms of data accuracy and measurement range. For example, when dealing with complex terrain or distant targets, low-cost lidar may not be able to provide the required data quality and completeness.

[0006] Furthermore, these devices are particularly sensitive to various environmental factors. Severe weather conditions, such as heavy rain, snow, or dense fog, can severely interfere with the data acquisition of lidar, thereby affecting its performance in practical applications. Summary of the Invention

[0007] The purpose of this application is to solve the problems in the prior art and provide a method and related apparatus for high-density reconstruction of 3D point clouds based on Gaussian process regression.

[0008] To achieve the above objectives, this application adopts the following technical solution:

[0009] One of the innovative aspects of this application is:

[0010] Gaussian process is used to perform high-density reconstruction of the vertical point cloud of 3D point cloud data. This increases the density of the sparse point cloud in the vertical direction without increasing the point cloud in the horizontal direction. This achieves high-precision mapping and positioning by increasing the number of point clouds while avoiding real-time computation difficulties caused by excessively large point cloud data.

[0011] The second innovative aspect of this application is:

[0012] The region growing method is used to determine the longitudinal line features of the point cloud, breaking up longitudinal point clouds that do not belong to the same object, and adding points along the longitudinal direction of the same object. Specifically:

[0013] In a first aspect, the present invention provides a method for high-density reconstruction of three-dimensional point clouds based on Gaussian process regression, comprising the following steps:

[0014] The original point cloud dataset was preprocessed using a random sampling consensus algorithm to obtain point cloud data with ground points removed.

[0015] The feature extraction of line segments is performed on the point cloud data with ground points removed using the region growing method to obtain point cloud data with separated longitudinal line bundles;

[0016] The point cloud data of the separated longitudinal line bundles is densified to obtain newly constructed point cloud data;

[0017] A laser odometry system was built using the newly constructed point cloud data to generate a point cloud relative pose graph and a point cloud map.

[0018] Secondly, the present invention provides a high-density reconstruction system for three-dimensional point clouds based on Gaussian process regression, comprising:

[0019] The preprocessing module is used to preprocess the original point cloud dataset using a random sampling consensus algorithm to obtain point cloud data with ground points removed.

[0020] The feature extraction module is used to extract line segment features from point cloud data with ground points removed using the region growing method, so as to obtain point cloud data with separated longitudinal line bundles.

[0021] The densification module is used to densify the point cloud data of the separated longitudinal line bundles to obtain newly constructed point cloud data.

[0022] The laser odometry construction module is used to build a laser odometry using newly constructed point cloud data, and generate a point cloud relative pose graph and a point cloud map.

[0023] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0024] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0025] Compared with the prior art, this application has the following beneficial effects:

[0026] This application focuses on how to perform high-density reconstruction of sparse data acquired by a low-cost 16-line LiDAR using a Gaussian process, addressing existing technologies. This Gaussian process-based algorithm not only effectively enhances the density of point cloud data but also significantly improves its accuracy, making it more suitable for high-precision applications. The scanning range of a 16-line LiDAR is roughly within a vertical range of 15' above and below the LiDAR's horizontal height. When the scene is 100 meters away from the LiDAR, the error between adjacent points in the vertical range is 1.1 meters; when the scene is 200 meters away, the error between adjacent points in the vertical range is 2.2 meters. The farther the distance, the sparser the LiDAR point cloud becomes. However, the density of the horizontal point cloud depends on the horizontal angular resolution of the LiDAR; at low resolution, the horizontal point cloud is dense. To address the problem of large map reconstruction errors and inaccurate positioning caused by vertically sparse point clouds in long-distance scanning, a method for high-density reconstruction of 3D sparse point clouds based on a Gaussian process is proposed. By preprocessing the point cloud, segmenting the vertical line features, and performing vertical Gaussian processing, the density of the sparse point cloud in the vertical direction is enriched. This breakthrough enables high-precision positioning and mapping of sparse point cloud laser SLAM (Simultaneous Localization and Mapping), completing efficient and accurate scanning matching and mapping tasks, and achieving higher-precision navigation and obstacle avoidance. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the method described in this application.

[0029] Figure 2 This is a schematic diagram of the system in this application.

[0030] Figure 3 This is a flowchart of the data processing part of an embodiment of this application.

[0031] Figure 4 To obtain the original point cloud image of the indoor environment.

[0032] Figure 5 This is a rendering of the ground segmentation using RANSAC.

[0033] Figure 6 This is a diagram illustrating the effect of the region growing method in an embodiment of this application.

[0034] Figure 7 This is a schematic diagram of the region growth effect in an embodiment of this application, where (a) is the unprocessed interpolation effect and (b) is the effect of interpolation after interruption.

[0035] Figure 8 This is a schematic diagram illustrating the effect of Gaussian interpolation on 100 sets of vertical points after vertical region growing.

[0036] Figure 9 This is the original point cloud image of the environment (29323 points).

[0037] Figure 10 This is the final result after processing a single frame of point cloud data. (Processed result - *Predicted point values ​​(117554 points))

[0038] Figure 11 The result of a complete process for a single point cloud frame (117,554 points). Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0040] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0041] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0042] In the description of the embodiments of this application, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In addition, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0043] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0044] In the description of the embodiments of this application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0045] Definitions:

[0046] RANSAC, short for Random Sample Consensus, is an algorithm that calculates mathematical model parameters for a dataset containing outliers to obtain valid sample data. First proposed by Fischler and Bolles in 1981, this algorithm is frequently used in computer vision, such as simultaneously solving the matching point problem between a pair of cameras and calculating the fundamental matrix in stereo vision.

[0047] Region growing is a common pixel or voxel classification method used in image processing or computer vision, based on adjacency relationships.

[0048] The present application will now be described in further detail with reference to the accompanying drawings:

[0049] See Figure 1 This application discloses a method for high-density reconstruction of 3D point clouds based on Gaussian process regression, including the following steps:

[0050] S1 uses the random sampling consensus algorithm to preprocess the original point cloud dataset to obtain point cloud data with ground points removed;

[0051] Furthermore, in a feasible embodiment of this application, for the original point cloud dataset, based on the point normal form equation of the plane, a random sampling consensus algorithm is used to extract as few initial data as possible that satisfy the feasible conditions, and the dataset conforming to the model is iteratively expanded using the consensus dataset.

[0052] Furthermore, in another feasible embodiment of this application, the random sampling consensus algorithm includes the following steps:

[0053] Step 1-1: Randomly select three points in the initial point cloud dataset and calculate the corresponding plane equations;

[0054] Step 1-2: Calculate the algebraic distance d from all points to the plane, select a threshold, filter out points within the model, and record the number of points;

[0055] Steps 1-3: Iteration, selecting the best fitting parameters; during the iteration process, determine whether to terminate the iteration based on the termination condition; the iteration termination condition is one or a combination of several of the following: expected error rate, optimal number of interior points, total number of samples, or iteration count evaluation factor.

[0056] Furthermore, in another feasible embodiment of this application, the plane equation is the equation corresponding to all points in space that lie on the same plane, as follows:

[0057] Ax + By + Cz + D = 0 (1)

[0058] Given the plane normal vector (A,B,C) and a point p0(x0,y0,z0) on the plane, the point normal form equation of the plane is as follows:

[0059] A(x-x0)+B(y-y0)+C(z-z0)=0 (2)

[0060] Expanding equation (2), we get:

[0061] D=-(Ax0+By0+Cz0) (3)

[0062] Where D is a constant term.

[0063] S2 uses the region growing method to extract line segment features from the point cloud data after removing ground points, and obtains point cloud data with separated longitudinal line bundles;

[0064] Furthermore, in another feasible embodiment of this application, the feature extraction of line segments from the point cloud data after removing ground points using the region growing method includes:

[0065] Step 2-1: Select one or more seed points in the point cloud as the starting point for growth;

[0066] Step 2-2: Set criteria for determining whether two points should be classified into the same region;

[0067] Steps 2-3: Starting from the seed point, gradually find and add neighboring points that meet the growth conditions to the same area; the newly added point will also become a new seed point, and the search for its neighboring points will continue; this process is repeated until no new points can be added, then growth stops; the growth conditions include distance, angle, color, and density characteristics;

[0068] Steps 2-4: When a region stops growing, select a new unclassified seed point to start growing a new region.

[0069] S3 performs densification processing on the point cloud data of the separated longitudinal line bundles to obtain newly constructed point cloud data;

[0070] Furthermore, in another feasible embodiment of this application, the Gaussian process regression method is used to perform vertical interpolation on the processed point cloud to densify the vertical point cloud data.

[0071] Furthermore, in another feasible embodiment of this application, the Gaussian process regression method is based on the Gaussian process, and for a given input point, finds its corresponding output value, that is, makes a prediction.

[0072] The Gaussian process includes a mean function and a covariance function, as detailed below:

[0073] m(t)=E[f(t)] (4)

[0074] K(t,t')=E[(f(t)-m(t))(f(t')-m(t'))] (5)

[0075] f(t)~gp(m(t),K(t,t')) (6)

[0076] Where m(t) is the mean function, f(t) is the true function, E is the sign of the mean, K(t,t') is the covariance function, gp is the Gaussian process, t is the true function parameter, and t' is the first derivative of the true function parameter.

[0077] The prediction specifically involves: given a set of training data and the observed values ​​of the training data, calculating the posterior distribution of the output value at a new input point using Gaussian processes and Bayesian rules; obtaining the prediction based on the posterior distribution, where the mean of the prediction is the most likely output value, and the variance of the prediction represents the uncertainty of the prediction, as detailed below:

[0078]

[0079]

[0080] Where L represents the measurement dataset observed by the lidar in the Cartesian coordinate system. For the observed azimuth information, r t i The distance between observed points is given by n, where n is the number of points in a single laser scan frame, and r is the distance between points. i For the observations in the training set, ε i To reduce noise that affects the observed values, To conform to a positive distribution, The method uses Gaussian noise, where i is the number of points acquired in one frame of laser data, and r is the distance of the original point cloud measurement data. * φ represents the test distance in the point cloud, and φ represents the azimuth angle of the original measurement data of the point cloud. * Let be the test azimuth angle in the point cloud, T be the transformed rank of the matrix, and I be the identity matrix with appropriate dimensions.

[0081] S4 uses the newly constructed point cloud data to build a laser odometry system and generate a point cloud relative pose map and a point cloud map.

[0082] like Figure 2 As shown in the figure, this application discloses a high-density reconstruction system for three-dimensional point clouds based on Gaussian process regression, including:

[0083] The preprocessing module is used to preprocess the original point cloud dataset using a random sampling consensus algorithm to obtain point cloud data with ground points removed.

[0084] The feature extraction module is used to extract line segment features from point cloud data with ground points removed using the region growing method, so as to obtain point cloud data with separated longitudinal line bundles.

[0085] The densification module is used to densify the point cloud data of the separated longitudinal line bundles to obtain newly constructed point cloud data.

[0086] The laser odometry construction module is used to build a laser odometry using newly constructed point cloud data, and generate a point cloud relative pose graph and a point cloud map.

[0087] The principle of this application

[0088] This application aims to utilize Gaussian processes to perform algorithmic-level densification processing on data acquired by existing low-cost 16-line LiDAR, in order to develop a high-density 3D point cloud reconstruction algorithm based on Gaussian processes. The specific data processing flow is as follows: Figure 3As shown, a mobile robot equipped with a LiDAR collects point clouds of the indoor environment, constructs a radar distance image, generates an original point cloud, preprocesses the original point cloud using the Random Sample Consensus (RANSAC) method to remove ground points, extracts line segment features using the region growing method, separates longitudinal line bundles, and then performs point cloud densification processing. Finally, a LiDAR odometry system is built using the newly constructed point cloud data to generate a point cloud relative pose map and a point cloud map, thus realizing the function of new 3D scene reconstruction.

[0089] Example:

[0090] This embodiment discloses a method for high-density reconstruction of 3D point clouds based on Gaussian process regression. The specific steps are as follows:

[0091] Step 1: Point cloud preprocessing based on the Random Sample Consensus (RANSAC) method

[0092] Ground segmentation of point clouds is an important step in many applications. Separating the ground helps reduce computation, remove redundant noise, perceive the environment more effectively, and extract features and identify targets in the parts of interest.

[0093] The equation of a plane is the equation corresponding to all points in space that lie on the same plane. Its general form is shown in Equation 1. Given the plane normal vector (A,B,C) and a point p0(x0,y0,z0) on the plane, the point normal form equation of the plane is shown in Equation 2.

[0094] Ax + By + Cz + D = 0 (1)

[0095] A(x-x0)+B(y-y0)+C(z-z0)=0 (2)

[0096] Expanding it, we get the constant term D, as shown in Equation 3:

[0097] D=-(Ax0+By0+Cz0) (3)

[0098] According to the above formula, the Random Sampling Consensus Algorithm (RANSAC) is used to extract as few initial data as possible that meet the feasible conditions for the dataset, and the dataset that conforms to the model is iteratively expanded using the consistent dataset.

[0099] The RANSAC plane fitting process is as follows:

[0100] (1) Randomly select three points in the initial point cloud and calculate the corresponding plane equations;

[0101] (2) Calculate the algebraic distance d from all points to the plane, select a threshold, filter out the points in the model, and record the number of points;

[0102] (3) Iteration: Select the best fitting parameters; during the iteration process, determine whether to stop the iteration based on the expected error rate, the number of best interior points, the total number of samples, the iteration number evaluation factor, and other conditions.

[0103] This application uses the RANSAC method for ground segmentation processing of indoor point clouds. Figure 4 It is the original point cloud image of the indoor environment. Figure 5 This is a rendering of the ground segmentation effect using RANSAC.

[0104] Step 2: Feature Extraction Research Based on Region Growing

[0105] Region growing is a pixel or voxel-based classification method commonly used in image processing and computer vision. It extracts regions from an image that share similar characteristics among neighboring pixels or voxels. Region growing is also widely used in point cloud processing; its strategy and principle are the same, but the application object changes from pixels to points, and it is often used for tasks such as segmentation and feature extraction. An example of region growing is shown below. Figure 6 As shown, the basic process of the region growing method is as follows:

[0106] 1) Select seed points: Select one or more seed points in the point cloud as the starting point for growth.

[0107] 2) Define growth conditions: Set criteria for determining whether two points should belong to the same region. Growth conditions can include characteristics such as distance, angle, color, and density.

[0108] 3) Growth process: Starting from the seed point, neighboring points that meet the growth conditions are gradually found and added to the same region. Newly added points also become new seed points, and the search for their neighbors continues. This process is repeated until no new points can be added.

[0109] 4) Select a new seed point: When a region stops growing (no new points can be added), select a new unclassified seed point to start the growth of the new region.

[0110] This process continues until all points are assigned to a specific region. In this way, a series of consecutive regions are extracted, and each region is a set of points with similar characteristics.

[0111] The point cloud image after ground segmentation is separated and extracted along the vertical direction. For the vertical point set, a region growing method is used to separate line segments, achieving the effect of interpolating discontinuous point sets. For example... Figure 7 The figures show (a) the interpolation effect without processing and (b) the interpolation effect after interruption.

[0112] Step 3: Algorithm Research Based on Gaussian Process Regression

[0113] A Gaussian process is a stochastic process defined over a continuous domain, where the joint distribution of any finite number of points is a Gaussian distribution. In a Gaussian process, each point is considered a random variable. A Gaussian process is mainly defined by two parts: the mean function and the covariance function, also called the kernel function.

[0114] m(t)=E[f(t)] (4)

[0115] K(t,t')=E[(f(t)-m(t))(f(t')-m(t'))] (5)

[0116] f(t)~gp(m(t),K(t,t'))(6)

[0117] Gaussian process regression (GPR), based on Gaussian processes, aims to find the corresponding output value for a given input point, i.e., to make a prediction. Specifically, given a set of training data and its observations, the posterior distribution of the output value at a new input point can be calculated using Gaussian processes and Bayesian rules. This posterior distribution provides the prediction: the mean of the prediction is the most likely output value, while the variance of the prediction represents the uncertainty of the prediction.

[0118] Gaussian process regression is used to perform vertical interpolation on the processed point cloud to densify the vertically sparse point cloud. For example... Figure 8 This is a schematic diagram illustrating the effect of Gaussian interpolation on 100 sets of vertical points that have undergone vertical region growing processing.

[0119] Figure 9 This is the original point cloud image of the environment. Figure 10 and Figure 11 These are the final images after a complete process of processing a single frame of point cloud data. Figure 10 The red dots represent prediction points.

[0120] This application provides a computer device according to an embodiment. The computer device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0121] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete this application.

[0122] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor and memory.

[0123] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0124] The memory can be used to store the computer program and / or module, and the processor implements various functions of the computer device by running or executing the computer program and / or module stored in the memory, and by calling the data stored in the memory.

[0125] If the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0126] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for high-density reconstruction of 3D point clouds based on Gaussian process regression, characterized in that, Includes the following steps: The original point cloud dataset was preprocessed using a random sampling consensus algorithm to obtain point cloud data with ground points removed. The feature extraction of line segments is performed on the point cloud data after removing ground points using the region growing method to obtain point cloud data with separated longitudinal line bundles; The point cloud data of the separated longitudinal line bundles is densified to obtain newly constructed point cloud data; specifically, it includes: using the Gaussian process regression method to perform longitudinal interpolation on the processed point cloud to densify the longitudinal point cloud data; the Gaussian process regression method is based on the Gaussian process, and for a given input point, it finds its corresponding output value, that is, makes a prediction. The Gaussian process includes a mean function and a covariance function, as detailed below: in, It is a mean function. For the real function, To find the sign of the mean, Let covariance function be used. Processed as a Gaussian process. For the actual function parameters, The first derivative of the actual function parameter; The prediction specifically involves: given a set of training data and the observed values ​​of the training data, calculating the posterior distribution of the output value at a new input point using Gaussian processes and Bayesian rules; obtaining the prediction based on the posterior distribution, where the mean of the prediction is the most likely output value, and the variance of the prediction represents the uncertainty of the prediction, as detailed below: in, This is the measurement dataset observed by lidar in Cartesian coordinates. For the observed azimuth information, The distance value of the observed point. The number of points in one frame of laser scanning. For the observations in the training set, To reduce noise that affects the observed values, To conform to a positive distribution, The method for Gaussian noise, This represents the number of points captured in one laser frame. The distance of the original point cloud measurement data. The test distance in the point cloud. The azimuth angle is the original measurement data of the point cloud. This refers to the test azimuth angle in the point cloud. To transform the rank of a matrix, It is an identity matrix with appropriate dimensions; A laser odometry system was built using the newly constructed point cloud data to generate a point cloud relative pose graph and a point cloud map.

2. The method for high-density reconstruction of 3D point clouds based on Gaussian process regression according to claim 1, characterized in that, The preprocessing of the raw point cloud data using the random sampling consensus algorithm includes: For the original point cloud dataset, based on the point normal form equation of the plane, the random sampling consensus algorithm is used to extract as few initial data as possible that meet the feasible conditions, and the consensus dataset is used to iteratively expand the dataset that conforms to the model.

3. The method for high-density reconstruction of 3D point clouds based on Gaussian process regression according to claim 2, characterized in that, The random sampling consensus algorithm includes the following steps: Step 1-1: Randomly select three points in the initial point cloud dataset and calculate the corresponding plane equations; Step 1-2: Calculate the algebraic distance d from all points to the plane, select a threshold, filter out points within the model, and record the number of points; Steps 1-3: Iteration, selecting the best fitting parameters; during the iteration process, determine whether to terminate the iteration based on the termination condition; the iteration termination condition is one or a combination of several of the following: expected error rate, optimal number of interior points, total number of samples, or iteration count evaluation factor.

4. The method for high-density reconstruction of 3D point clouds based on Gaussian process regression according to claim 3, characterized in that, The plane equation is the equation corresponding to all points in space that lie in the same plane, as follows: Given the plane normal vector On the surface Then the point-normal form of the plane is as follows: The formula Unfold, and you get: in, This is a constant term.

5. The method for high-density reconstruction of 3D point clouds based on Gaussian process regression according to claim 1, characterized in that, The method of extracting line segment features from point cloud data after removing ground points using the region growing method includes: Step 2-1: Select one or more seed points in the point cloud as the starting point for growth; Step 2-2: Set criteria for determining whether two points should be classified into the same region; Steps 2-3: Starting from the seed point, gradually find and add neighboring points that meet the growth conditions to the same area; the newly added point will also become a new seed point, and the search for its neighboring points will continue; this process is repeated until no new points can be added, then growth stops; the growth conditions include distance, angle, color, and density characteristics; Steps 2-4: When a region stops growing, select a new unclassified seed point to start growing a new region.

6. A high-density reconstruction system for three-dimensional point clouds based on Gaussian process regression for implementing the method of claim 1, characterized in that, include: The preprocessing module is used to preprocess the original point cloud dataset using a random sampling consensus algorithm to obtain point cloud data with ground points removed. The feature extraction module is used to extract line segment features from point cloud data with ground points removed using the region growing method, so as to obtain point cloud data with separated longitudinal line bundles. The densification module is used to densify the point cloud data of the separated longitudinal line bundles to obtain newly constructed point cloud data. The laser odometry construction module is used to build a laser odometry using newly constructed point cloud data, and generate a point cloud relative pose graph and a point cloud map.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-5.

Images