Ground segmentation method, system, device and medium based on point cloud geometric structure

Abstract

The present invention belongs to the technical field of point cloud data processing. In order to solve the problems of over-segmentation and under-segmentation in the existing methods, the long model training time and the poor model generalization ability, a ground segmentation method, system, equipment and medium based on the point cloud geometric structure are proposed. Based on the distance from each point cloud data to the center of the laser radar, and the vertical angle and azimuth of each point cloud data, a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map of the point cloud data are constructed. Coarse segmentation is performed based on the angle between the connecting line of adjacent line bundle point clouds and the horizontal plane. The ground prior matrix is ​​subjected to elevation filtering and normal vector filtering in turn using the point cloud height information and the obtained normal vector to obtain the ground matrix. The ground matrix is ​​finely segmented based on the iterative plane fitting method. The method has the advantages of high segmentation accuracy, strong robustness and easy implementation.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud data processing, and in particular relates to a ground segmentation method, system, equipment and medium based on point cloud geometric structure. Background Art

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

[0003] The rapid progress of autonomous driving technology is driving major changes in the transportation sector, making traditional transportation develop in an intelligent direction, and providing new possibilities for improving traffic efficiency, road safety and travel convenience. In order to cope with the complex and changing real-time road environment, intelligent driving vehicles must have highly accurate environmental perception and autonomous decision-making capabilities. The former relies on a reliable sensor system and is the core component of autonomous driving technology. Sensors sense and collect rich environmental information around them, use advanced algorithms to analyze and process data, and provide necessary information support and reference for the vehicle's decision-making system. Commonly used sensors in autonomous driving include cameras, GPS, ultrasonic radar, millimeter-wave radar, lidar, inertial navigation, etc. Each sensor has its unique role and advantages in achieving the goal of autonomous driving. By integrating data from multiple sensors, the perception module of the autonomous driving system can build a more comprehensive and accurate environmental perception model, thereby improving the vehicle's autonomous decision-making ability and achieving a safer and more efficient autonomous driving experience.

[0004] Among them, laser radar, as a high-precision perception device, is divided into single-line laser radar and multi-line laser radar according to the number of lines. Compared with other sensors, laser radar can still maintain high accuracy in bad weather and complex lighting conditions. Therefore, multi-line laser radar has been widely used in the three-dimensional perception system of autonomous driving at this stage and plays an indispensable role. Laser radar emits a laser beam within a certain field of view angle range, and uses a receiving component to receive the laser beam reflected back within the range. The time difference between the two is used to directly calculate the distance between the vehicle and the target being measured. At the same time, other relevant information of the reflection point (speed, height, reflection intensity, etc.) can be further obtained, thereby generating high-resolution laser point cloud data containing rich environmental details. These point cloud data can not only help the perception system accurately identify obstacles and other traffic participants, but also provide rich spatial information for tasks such as ground segmentation and environmental map construction.

[0005] Laser point cloud ground segmentation is one of the key technologies in the current fields of autonomous driving, robot navigation, 3D modeling, etc. As an important step in point cloud data preprocessing, it is mainly to separate ground points and non-ground points from the 3D point cloud data obtained by the LiDAR, and then assist in subsequent tasks such as point cloud clustering, obstacle detection, point cloud registration, and high-precision map establishment. Accurate ground segmentation of point cloud data can provide constraints in the vertical direction, thereby constructing a more accurate 3D global map and achieving more accurate positioning. It can also provide accurate prior information for the detection of traversable areas of the autonomous driving system, which helps the vehicle make decisions and plan the driving path, thereby improving driving safety.

[0006] At present, in the field of point cloud ground segmentation, researchers have proposed a variety of methods in order to obtain better segmentation results. Among them, the most widely used is the traditional method based on geometric features. It mainly estimates the model parameters of the ground and distinguishes between ground and non-ground points by applying principal component analysis, random sampling consistency algorithm, Hough transform and other methods to the geometric features of the points in the point cloud. However, this type of method is easily disturbed by outliers and noise, has limited performance when dealing with complex terrain and scenes, has problems of over-segmentation and under-segmentation, and has poor robustness. In recent years, with the continuous development of deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN) and graph neural networks (GNN), researchers have gradually found that these methods also show superior performance in point cloud ground segmentation tasks. This type of method uses deep neural networks to process large-scale data and learn complex patterns from them for automatic feature extraction, which improves the accuracy and efficiency of ground segmentation tasks and performs well in dealing with complex terrain and diverse scenes. However, deep learning-based methods also have many shortcomings. First, deep learning models usually require a lot of computing resources and storage space, especially in the training phase, which is particularly obvious when processing large-scale point cloud data. In addition, deep learning methods usually require large-scale labeled data for training, and the acquisition cost of these labeled data is high, the model training time is long, and the generalization ability of the model is average.

[0007] In summary, there are at least the following problems in point cloud ground segmentation:

[0008] Traditional methods based on geometric features are easily affected by outliers and noise, have limited performance when dealing with complex terrain and scenes, and suffer from problems of over-segmentation and under-segmentation. Solutions based on neural networks require a lot of computing resources and storage space, take a long time to train the model, and have poor model generalization capabilities. Summary of the invention

[0009] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a ground segmentation method, system, equipment and medium based on point cloud geometric structure, which extracts the point cloud geometric structure and then obtains accurate and clear ground point cloud based on coarse segmentation and fine segmentation. It has the advantages of high segmentation accuracy, strong robustness and easy implementation.

[0010] In order to achieve the above object, the present invention adopts the following technical solution:

[0011] In a first aspect, the present invention provides a ground segmentation method based on a point cloud geometric structure, comprising:

[0012] Based on the distance from each point cloud data to the center of the laser radar, as well as the vertical angle and azimuth angle of each point cloud data, a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map are constructed respectively;

[0013] Determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the point to be estimated by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map;

[0014] Based on the angle between the point cloud connection line of the adjacent laser radar beams and the horizontal plane, the ground rough segmentation is performed on the spherical projection depth feature map of the point cloud to obtain a ground prior matrix marking the ground points and non-ground points;

[0015] Using the point cloud height information and the obtained normal vector, the ground prior matrix is ​​processed with elevation filtering and normal vector filtering in turn to obtain a ground matrix;

[0016] The ground matrix is ​​subjected to fine segmentation of the ground point cloud based on an iterative plane fitting method to obtain a ground point set.

[0017] In a second aspect, the present invention provides a ground segmentation system based on point cloud geometry structure, comprising:

[0018] A construction module is configured to: construct a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map based on the distance from each point cloud data to the center of the laser radar, and the vertical angle and azimuth angle of each point cloud data;

[0019] A normal vector solving module is configured to: determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the point to be estimated by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map;

[0020] A coarse segmentation module is configured to: perform ground coarse segmentation on the point cloud spherical projection depth feature map based on the angle between the point cloud connection line of the adjacent beams of the laser radar and the horizontal plane, and obtain a ground priori matrix marking ground points and non-ground points;

[0021] A filtering module is configured to: use the point cloud height information and the obtained normal vector to sequentially perform elevation filtering and normal vector filtering on the ground prior matrix to obtain a ground matrix;

[0022] The fine segmentation module is configured to: perform fine segmentation of the ground point cloud on the ground matrix based on an iterative plane fitting method to obtain a ground point set.

[0023] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0024] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0025] One or more of the above technical solutions have the following beneficial effects:

[0026] In the present invention, a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map of point cloud data are constructed, coarse segmentation is performed based on the angle between the connecting line of adjacent harness point clouds and the horizontal plane, and then fine segmentation is performed based on the iterative plane fitting method to obtain an accurate and clear ground point cloud. This method has the advantages of high segmentation accuracy, strong robustness, and easy implementation, and is of great significance for improving the perception accuracy of the autonomous driving system and making safe and efficient decision-making plans for vehicles.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 is a flowchart of a two-stage ground segmentation method based on point cloud geometry in Embodiment 1 of the present invention;

[0030] Figure 2 is a schematic diagram of a point cloud observation model in three-dimensional space used in the first embodiment of the present invention;

[0031] Figure 3 It is a schematic diagram of constructing a multi-line laser radar spherical projection feature map in the first embodiment of the present invention;

[0032] Figure 4 is a schematic diagram of ground rough segmentation in Embodiment 1 of the present invention;

[0033] Figure 5 It is a schematic diagram of point cloud elevation filtering in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0034] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0035] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0036] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0037] Embodiment 1

[0038] This embodiment discloses a ground segmentation method based on point cloud geometric structure, including:

[0039] Based on the distance from each point cloud data to the center of the laser radar, as well as the vertical angle and azimuth angle of each point cloud data, a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map are constructed respectively;

[0040] Determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the local plane by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map;

[0041] Based on the angle between the connecting line of the adjacent line bundle point clouds and the horizontal plane, the ground rough segmentation is performed on the point cloud spherical projection depth feature map to obtain a ground prior matrix marking the ground points;

[0042] Using the point cloud height information and the obtained normal vector, the ground prior matrix is ​​processed with elevation filtering and normal vector filtering in turn to obtain a ground matrix;

[0043] The ground matrix is ​​subjected to fine segmentation of the ground point cloud based on an iterative plane fitting method to obtain a ground point set.

[0044] This embodiment constructs a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map of point cloud data, performs coarse segmentation based on the angle between the connecting lines of adjacent harness point clouds and the horizontal plane, and then performs fine segmentation based on an iterative plane fitting method to obtain an accurate and clear ground point cloud. It has the advantages of high segmentation accuracy, strong robustness, and easy implementation, and is of great significance to improving the perception accuracy of the autonomous driving system and making safe and efficient decisions and planning for vehicles.

[0045] Combine the following Figure 1The two-stage ground segmentation method based on point cloud geometry proposed in this embodiment is described in detail, including:

[0046] Step 1: Perform point cloud projection preprocessing.

[0047] Step 11: For a line number of , the horizontal angular resolution is , the vertical angular resolution is , the vertical field of view is The number of point clouds obtained by calculating the rotation of each beam is , .in, represents the upper half of the xOy plane, that is, the field of view within the range of z>0, is the field of view within the range of z<0. Generally speaking, the two are symmetrical and have the same angle range, that is, .

[0048] Step 12: Project all point clouds onto the sphere, and then expand them along the positive direction of the x-axis with the intersection of the xOz plane and the sphere to obtain a point cloud of size The spherical projection matrix .

[0049] Step 13: Set a laser point in the point cloud obtained by the laser radar scan , then it can be expressed in three-dimensional space as ,like Figure 2 shown.

[0050] Step 14: Calculate the Azimuth , that is, the angle between each point cloud and the xOz plane, calculated as follows:

[0051] (1)

[0052] Step 15: Calculate the vertical angle , that is, the angle between each point cloud and the xOy plane, calculated as follows:

[0053] (2)

[0054] Step 15: Calculate the row number. Some manufacturers' lidars will directly provide the row number of each point, or calculate it using the following formula:

[0055] (3)

[0056] in, is the vertical angle of the point, and [·] is the rounding symbol.

[0057] Step 16: Calculate the column number using the following formula:

[0058] (4)

[0059] in, is the azimuth of the point, and [·] is the rounding symbol.

[0060] Step 17: Filter invalid points. Points with NaN values ​​in the coordinates or points with abnormal reflection intensity received by the radar are invalid points. The corresponding positions in are marked as invalid values, that is:

[0061]

[0062] Step 18: Filter the reflected noise from the vehicle body. A small part of the laser beam will hit the vehicle body itself, generating noise and affecting the subsequent ground segmentation, so it needs to be filtered out.

[0063] Assume the horizontal distance between the center of the laser radar and the front of the vehicle is , the horizontal distance from the rear of the vehicle is , for the horizontal axis to satisfy the condition Point cloud, project it on the spherical projection matrix The corresponding position in is marked as an invalid value.

[0064] Step 2: Construct a spherical projection depth feature map of the multi-line lidar.

[0065] Step 21: Calculate the distance from each point cloud to the center of the LiDAR , the calculation formula is as follows:

[0066] (5)

[0067] Step 22: Figure 3 As shown, the calculated depth data is filled into the corresponding position in the spherical projection matrix until all elements of the matrix are filled to obtain the point cloud spherical projection depth feature map ,Right now:

[0068] (6)

[0069] in is the row number of the point, is the column number of the point.

[0070] Step 3: Construct the spherical projection angle feature map of the multi-line lidar.

[0071] Step 31: Define the angle structure information vector according to the vertical angle and azimuth angle data calculated in step 1 .

[0072] Step 32: Convert the above calculated As an element to fill the corresponding position in the spherical projection matrix until all elements of the matrix are filled, the point cloud spherical projection angle feature map is obtained. ,Right now:

[0073]

[0074] in is the row number of the point, is the column number of the point.

[0075] Step 4: Fast approximate estimation of the point cloud surface normal vector based on the local plane.

[0076] Step 41: Note is the surface normal vector of the point cloud, and the neighborhood of the point to be estimated is , is the centroid of the neighborhood points, and the calculation formula is as follows:

[0077] (7)

[0078] Step 42: Select the neighborhood of the point to be estimated. Considering that the vertical resolution of the mechanical laser radar is greater than the horizontal resolution, in order to better reflect the local surface information of the point cloud, select the points in a rectangular area of ​​5×3 centered on the point to be estimated as the neighborhood for subsequent calculations.

[0079] Step 43: Constructing the residual ,in, is the distance from the origin to the fitting plane, and the calculation formula is: .

[0080] Step 4.4, from the above steps, we can see that the observation model of the lidar can be expressed as:

[0081] (8)

[0082] in , substituting into the above residual formula, the residual is transformed into ,

[0083] Step 4.5: Since the point to be estimated and its neighboring points are concentrated in a small area, the distance to the laser radar is approximately the same, so the residual is further simplified to .

[0084] Step 46: For each scan point in the point cloud and neighborhood points, calculate the local plane in the least squares sense , this plane can be expressed as:

[0085] (9)

[0086] Step 47: The closed-form solution formula is derived as follows:

[0087] (10)

[0088] in, , .

[0089] Step 48: Project the depth feature map from the sphere Angle feature map with spherical projection Directly and quickly query the relevant data and substitute it into the above formula to calculate .

[0090] Step 49: Setting the viewpoint , for each scan point in the point cloud , the corresponding normal vector is , orient all normal vectors, that is, select their direction toward the viewpoint, satisfying the following formula:

[0091] (11)

[0092] This completes the point cloud normal vector estimation and redirection.

[0093] Step 5: Perform rough ground segmentation based on the spherical projection feature map.

[0094] Step 51: First, make the following assumptions: the sensor is mounted horizontally on a mobile base; the ground curvature is very low; the laser beam of the laser radar at a lower position in this scan must be able to scan the ground.

[0095] Step 52: Based on the assumption, project the depth feature map onto the sphere In the The lower half of is extracted separately to construct the ground prior matrix .

[0096] Step 53: Get the ordinate of the point cloud in the lowest line bundle, that is, the line bundle with row number 1 , which can be obtained directly from the radar or through Calculated, among which, r It refers to the distance from the laser point to the center of the radar.

[0097] Step 54: Perform preliminary screening of ground points. If the vertical coordinate of the point cloud with column number j in the row satisfies , then in the ground prior matrix Mark the point as a ground point, that is:

[0098]

[0099] in, is the sensor installation height, is the elevation filter threshold.

[0100] Step 55: Figure 4 As shown, calculate the angle between the line connecting the two adjacent line point clouds with row numbers i and i-1 in the jth column and the horizontal plane , the calculation formula is as follows:

[0101] (12)

[0102] Step 56: Set the ground tilt angle threshold , if satisfied and , then in the ground prior matrix Mark the point as a ground point and record the ground point , otherwise it is a non-ground point, recorded as a non-ground point , until all elements in the matrix are filled.

[0103] Step 6: Update the ground prior matrix based on the point cloud normal vector information obtained in step 4 .

[0104] Step 61: First, perform point cloud elevation filtering. For the ground prior matrix after rough segmentation that satisfies The point cloud retains its ordinate in Points within the range. Figure 5 As shown, the yellow points are the points retained by the elevation filter, and the red points are the points excluded by the elevation filter.

[0105] Step 62: Perform point cloud normal vector filtering and make the following assumption: if the segmented ground point cloud is close to the actual ground, then the direction of its normal vector should be as vertical as possible.

[0106] Step 63: Calculate the angle between the normal vector of each point and the z-axis direction vector , the calculation formula is as follows:

[0107] (13)

[0108] in, is the normal vector, is the direction vector of the z-axis, Represents the magnitude of a vector.

[0109] Step 64: Set the ground vertical angle threshold , if a point satisfies , then add the point to the ground point set, and the ground prior matrix The corresponding element is updated to 1, otherwise it is updated to 0, and the number of elements with a mark value of 1 is recorded .

[0110] Step 65: After all elements in the ground prior matrix are updated, the ground marker updated matrix is ​​recorded as .

[0111] Step 7: Perform fine segmentation of ground point cloud based on iterative plane fitting.

[0112] Step 71: First define the parameters , for the ground point cloud processed by the above steps The value is calculated and The values ​​are sorted in ascending order.

[0113] Step 72: Select the point cloud The smallest value points as initial ground seed points, constructing an initial ground seed point set .in, is the proportionality coefficient and , is the ground matrix updated in step 6 The number of elements, .

[0114] Step 73: Perform plane fitting on the initial ground seed point set.

[0115] Construct the objective function according to step 43 , and further derive the covariance matrix .

[0116] Step 74: Perform SVD decomposition on the covariance matrix M.

[0117] Right now: .

[0118] in, It is a singular value matrix with singular values ​​on the diagonal, and both U and V are unitary matrices.

[0119] Step 75: Take the singular vector corresponding to the minimum singular value, which is the last column of the U matrix , which is the normal vector of the plane fitted by the ground seed point .

[0120] Step 76: Right Normalize it, the formula is as follows:

[0121] (14)

[0122] Step 77: Calculate each ground seed point The distance to the above fitting plane , the calculation formula is as follows:

[0123] (15)

[0124] Step 78: Setting the distance threshold , if the distance from the point to the plane does not satisfy , then remove the point from the ground seed point set, update the ground point set, and obtain a new ground seed point set .

[0125] Step 79: Use the Jaccard index to calculate the overlap of the point set. The calculation formula is as follows:

[0126] (16)

[0127] Step 710: Calculate the degree of change of the point set , the calculation formula is as follows:

[0128] (17)

[0129] Among them, i represents the current iteration number, Represents the current ground seed point set.

[0130] Step 711: Set the number of iterations I and the point set change threshold , determine whether the conditions are met Or the number of iterations is reached. If none of the conditions are met, steps 72 to 711 are repeated. If any of the conditions is met, the algorithm is considered to have converged and the ground segmentation process is terminated.

[0131] Embodiment 2

[0132] The purpose of this embodiment is to provide a ground segmentation system based on point cloud geometry structure, including:

[0133] A construction module is configured to: construct a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map based on the distance from each point cloud data to the center of the laser radar, and the vertical angle and azimuth angle of each point cloud data;

[0134] A normal vector solving module is configured to: determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the point to be estimated by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map;

[0135] A coarse segmentation module is configured to: perform ground coarse segmentation on the point cloud spherical projection depth feature map based on the angle between the point cloud connection line of the adjacent beams of the laser radar and the horizontal plane, and obtain a ground priori matrix marking ground points and non-ground points;

[0136] A filtering module is configured to: use the point cloud height information and the obtained normal vector to sequentially perform elevation filtering and normal vector filtering on the ground prior matrix to obtain a ground matrix;

[0137] The fine segmentation module is configured to: perform fine segmentation of the ground point cloud on the ground matrix based on an iterative plane fitting method to obtain a ground point set.

[0138] In further embodiments, there is also provided:

[0139] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0140] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0141] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0142] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0143] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0144] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0145] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0146] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0147] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0148] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0149] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A ground segmentation method based on point cloud geometry, characterized in that: include: Based on the distance from each point cloud data to the center of the laser radar, as well as the vertical angle and azimuth angle of each point cloud data, a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map are constructed respectively; Determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the point to be estimated by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map; Based on the angle between the point cloud connection line of the adjacent laser radar beams and the horizontal plane, the ground rough segmentation is performed on the spherical projection depth feature map of the point cloud to obtain a ground prior matrix marking the ground points and non-ground points; Using the point cloud height information and the obtained normal vector, the ground prior matrix is ​​processed with elevation filtering and normal vector filtering in turn to obtain a ground matrix; For the ground matrix, fine segmentation of the ground point cloud is performed based on an iterative plane fitting method to obtain a ground point set; The step of determining the point to be estimated and its neighborhood, determining the local plane corresponding to the point to be estimated, and combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map to obtain the normal vector of the point to be estimated is specifically as follows: Taking the point to be estimated as the center, determine the neighborhood of the point to be estimated; Based on the relationship between the LiDAR observation model and the distance from the point cloud to the center of the LiDAR, the local plane is calculated in the least squares sense. According to the calculated local plane, a representation of a normal vector of the local plane is derived, and the normal vector of the local plane is obtained by searching the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map; It also includes redirecting the normal vector of the local plane obtained, specifically: setting the viewpoint, for the normal vector corresponding to each point cloud data, selecting the direction of the normal vector toward the viewpoint so that it satisfies: ,in, For the i The normal vector corresponding to the point cloud is For the i Point cloud, For viewpoint; Based on the angle between the connecting line of the adjacent beam point clouds and the horizontal plane, the ground rough segmentation is performed on the point cloud spherical projection depth feature map to obtain the ground prior matrix marking the ground points and non-ground points, specifically: Constructing a ground prior matrix according to the point cloud spherical projection depth feature map; Obtain the ordinate of the point cloud in the lowest beam of the laser radar, and determine whether the point cloud in the lowest beam is a ground point based on the sensor installation height and the elevation filter threshold, and update the ground prior matrix; Based on the angle between the line connecting the point clouds of adjacent beams in the lowest beam and the horizontal plane and the value of the preset ground tilt angle threshold, the updated ground prior matrix is ​​updated again to obtain a ground prior matrix marking ground points and non-ground points; The ground matrix performs fine segmentation of the ground point cloud based on an iterative plane fitting method to obtain a ground point set, specifically: Construct an initial ground seed point set, and perform plane fitting on the constructed initial ground seed point set to obtain a covariance matrix; Performing singular value decomposition on the covariance matrix, and selecting a singular vector corresponding to the minimum singular value for normalization; According to the normalized result, the distance from each ground seed point to the fitting plane is calculated; The ground seed point set is updated according to the relationship between the calculated distance and the preset distance threshold; The ground point cloud is sorted according to the ordinate of the point cloud in the lowest line bundle; the number of the selected initial ground seed points is determined according to the number of ground points in the ground matrix, and the initial ground seed point set is constructed.

2. The ground segmentation method based on point cloud geometry structure according to claim 1, characterized in that: The ground prior matrix is ​​processed by elevation filtering and normal vector filtering in turn using the point cloud height information and the obtained normal vector to obtain a ground matrix, which is specifically: Perform point cloud elevation filtering on the ground prior matrix of the marked ground points after rough segmentation; The angle between the normal vector of each point cloud and the z-axis direction vector is calculated. According to the calculated angle and the preset ground vertical angle threshold, the ground prior matrix of the marked ground points after the row point cloud elevation filtering is updated to obtain the ground matrix.

3. A ground segmentation system based on point cloud geometry, characterized in that: include: A construction module is configured to: construct a point cloud spherical projection depth feature map and a point cloud spherical projection angle feature map based on the distance from each point cloud data to the center of the laser radar, and the vertical angle and azimuth angle of each point cloud data; A normal vector solving module is configured to: determine the point to be estimated and its neighborhood, determine the local plane corresponding to the point to be estimated, and obtain the normal vector of the point to be estimated by combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map; A coarse segmentation module is configured to: perform ground coarse segmentation on the point cloud spherical projection depth feature map based on the angle between the point cloud connection line of the adjacent laser radar beams and the horizontal plane, and obtain a ground priori matrix marking ground points and non-ground points; A filtering module is configured to: use the point cloud height information and the obtained normal vector to sequentially perform elevation filtering and normal vector filtering on the ground prior matrix to obtain a ground matrix; A fine segmentation module is configured to: perform fine segmentation of the ground point cloud on the ground matrix based on an iterative plane fitting method to obtain a ground point set; The step of determining the point to be estimated and its neighborhood, determining the local plane corresponding to the point to be estimated, and combining the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map to obtain the normal vector of the point to be estimated is specifically as follows: Taking the point to be estimated as the center, determine the neighborhood of the point to be estimated; Based on the relationship between the LiDAR observation model and the distance from the point cloud to the center of the LiDAR, the local plane is calculated in the least squares sense. According to the calculated local plane, a representation of a normal vector of the local plane is derived, and the normal vector of the local plane is obtained by searching the point cloud spherical projection depth feature map and the point cloud spherical projection angle feature map; It also includes redirecting the normal vector of the local plane obtained, specifically: setting the viewpoint, for the normal vector corresponding to each point cloud data, selecting the direction of the normal vector toward the viewpoint so that it satisfies: ,in, For the i The normal vector corresponding to the point cloud is For the i Point cloud, For viewpoint; Based on the angle between the connecting line of the adjacent beam point clouds and the horizontal plane, the ground rough segmentation is performed on the point cloud spherical projection depth feature map to obtain the ground prior matrix marking the ground points and non-ground points, specifically: Constructing a ground prior matrix according to the point cloud spherical projection depth feature map; Obtain the ordinate of the point cloud in the lowest beam of the laser radar, and determine whether the point cloud in the lowest beam is a ground point based on the sensor installation height and the elevation filter threshold, and update the ground prior matrix; Based on the angle between the line connecting the point clouds of adjacent beams in the lowest beam and the horizontal plane and the value of the preset ground tilt angle threshold, the updated ground prior matrix is ​​updated again to obtain a ground prior matrix marking ground points and non-ground points; The ground matrix performs fine segmentation of the ground point cloud based on an iterative plane fitting method to obtain a ground point set, specifically: Construct an initial ground seed point set, and perform plane fitting on the constructed initial ground seed point set to obtain a covariance matrix; Performing singular value decomposition on the covariance matrix, and selecting a singular vector corresponding to the minimum singular value for normalization; According to the normalized result, the distance from each ground seed point to the fitting plane is calculated; The ground seed point set is updated according to the relationship between the calculated distance and the preset distance threshold; The ground point cloud is sorted according to the ordinate of the point cloud in the lowest line bundle; the number of the selected initial ground seed points is determined according to the number of ground points in the ground matrix, and the initial ground seed point set is constructed.

4. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 2 is completed.

5. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1-2.

Images