Intelligent wheelchair ground point cloud segmentation method and system

Abstract

The application discloses an intelligent wheelchair ground point cloud segmentation method, first, a radar installed on the wheelchair is used to collect point clouds, the original point cloud data is preprocessed, then a multi-resolution division method based on a radial interval and an angle interval is used to perform grid division on the point clouds to obtain sub-grid units, then the sub-grid units are classified based on height difference and variance information, finally, a Bayesian probability model is used to estimate the point clouds in the PG grid to obtain ground points therein, all ground points are merged to obtain estimated ground point clouds and are output. The application has less required computing resources and can obtain higher-precision ground segmentation results, can well solve problems that a traditional method cannot adapt to complex terrains and sparseness of point clouds and is not good in real-time performance, and can improve the efficiency of drivable area detection of intelligent wheelchairs or other mobile robots.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent wheelchair perception, and particularly relates to an intelligent wheelchair ground point cloud segmentation method and system. BACKGROUND

[0002] In recent years, with the rapid development of intelligent driving technology, automatic driving systems gradually move from theoretical exploration to practical application, and show great potential in the fields of automobiles, unmanned distribution, urban security, etc. In these application scenarios, environment perception technology, as the core module of the intelligent driving system, provides necessary environmental information support for navigation, path planning and obstacle avoidance. As a key environment perception sensor, laser radar has become an indispensable tool in the field of intelligent driving, with its high precision, strong robustness and all-weather working ability. Intelligent wheelchair, as a typical application scenario combining mobile robot technology and medical auxiliary equipment, has higher demand for environment perception technology in autonomous navigation. Unlike intelligent cars mainly driving on structured roads, intelligent wheelchairs need to operate in more complex unstructured environments, such as homes, hospitals, parks, etc. These environments often have irregular terrain, fuzzy road boundaries, dynamic obstacles and other challenges, making the ground segmentation task more difficult. Therefore, accurate ground point cloud segmentation is of great significance to intelligent wheelchairs, which is a key step for detecting safe and drivable areas, realizing autonomous navigation and ensuring user safety.

[0003] The current methods for ground point cloud segmentation can be mainly divided into four categories: 1) the method based on the height map, which realizes fast processing by projecting the three-dimensional point cloud to a two-dimensional grid and extracting the grid features, but this method is prone to over-segmentation in complex terrain and is difficult to adapt to steep terrain or overhanging structures; 2) the method based on plane fitting, which regards the ground as a locally flat area and divides the ground points by fitting a plane model, but this method has poor adaptability to complex and irregular terrain, especially in actual unstructured environments; 3) the method based on ground plane estimation, which estimates the ground height by using statistical models such as Gaussian regression or Markov random field, although the segmentation accuracy is high, but the computational complexity is too high, which is not conducive to real-time application; 4) the method based on deep learning, which relies on advanced neural network structures to perform semantic segmentation on point cloud data, has good performance but has problems such as lack of data sets, poor network interpretability and high demand for computing resources. The patent "Three-stage point cloud ground segmentation method based on ground plane fitting (202311711321.8)" proposes a point cloud ground segmentation algorithm, which divides the point cloud by using a high-resolution infinite far coordinate grid and voxel downsampling, fits the ground plane to preliminarily divide the point cloud, then extracts false negative points and false positive points by ground plane fitting, and finally realizes accurate ground point cloud segmentation through three-stage optimization adjustment, but this method has some shortcomings: the same resolution is used for grid division for both close-range and long-range point clouds, and the same threshold is used for screening, which is easy to miss the sparse ground point cloud in the distance, and the method of multiple ground plane fitting is time-consuming and lacks real-time performance. SUMMARY

[0004] To solve the above problems, the present application discloses an intelligent wheelchair ground point cloud segmentation method and system, which accurately segments the ground point cloud based on a multi-resolution grid division method, a grid classification method and a ground point probability estimation method, requires less computing resources and can obtain higher precision ground segmentation results, can well solve the problems that traditional methods cannot adapt to complex terrain and the sparsity of point clouds, and have poor real-time performance, and can help intelligent wheelchairs or other mobile robots detect drivable areas and realize autonomous navigation.

[0005] To achieve the above purpose, the technical scheme of the present application is as follows:

[0006] An intelligent wheelchair ground point cloud segmentation method, comprising the following steps:

[0007] S1, collect point cloud by a laser radar installed on the wheelchair with a height of h sensor , preprocess the original point cloud data, filter the point cloud data according to the preset distance and height threshold;

[0008] S2, using a multi-resolution partitioning method based on radial intervals and angular intervals to grid the point cloud, obtaining sub-grid units, the partitioning method uses different resolution grids to partition the point cloud in different radial intervals and angular intervals, to adapt to the sparsity of the point cloud, and the attributes of each sub-grid unit are calculated and stored during the gridding process;

[0009] S3, classifying the sub-grid units based on height difference and variance information into three categories: G, NG, and PG. The G category is a grid of all ground points (Ground Cell); the NG category is a grid of all non-ground points (Potential Ground Cell); and the PG category is a grid that may contain ground points (Non-Ground Cell);

[0010] S4, calculating the ground point probability based on the Bayesian probability model, estimating the ground points in the PG grid to obtain the ground points therein, merging all the ground points to obtain the estimated ground point cloud and output; the merging method is to take the union of the ground points of the G category and the ground points obtained by Bayesian estimation.

[0011] Further, step S1 is explained in detail:

[0012] During the laser radar acquisition process, laser reflection may cause noise points to appear below the actual ground, causing under-segmentation problems. Therefore, the original point cloud data is preprocessed using a multi-condition filtering method based on height and distance to filter the point cloud data, which will reject points with z values below a certain threshold within a certain distance, thereby effectively reducing outliers and significantly reducing their interference with the ground segmentation results.

[0013] Further, step S2 is explained in detail:

[0014] The fixed interval polar coordinate partitioning method is prone to sparsity problems when processing point clouds in long distance areas, and the small size of the grid in the near distance area may also cause feature estimation errors, so different resolution grids are used to partition the point cloud in different radial intervals and angular intervals to adapt to the sparsity of the point cloud and improve the segmentation accuracy. Specifically, M z concentric regions are unequally divided in the radial direction according to the distribution characteristics of the point cloud, and M z angular intervals are divided in the angular direction, obtaining M z ×N zone grid cell; the grid is further subdivided into sub-grid cells using different radial and angular resolutions, the radial and angular resolutions of which are determined by dividing the radial boundary difference of the concentric area and the boundary difference of the angular interval by the number of subdivisions, respectively, and the number of subdivisions is flexibly adjusted according to the different positions of the grid; the lowest point, the highest point, the height variance and the number of point clouds of each sub-grid cell are calculated as its attributes and stored during the subdivision process.

[0015] Further, step S3 is explained in detail:

[0016] Ground estimation on a large amount of point cloud data is time-consuming, in order to improve the segmentation efficiency of the algorithm, the sub-grid cells are classified based on height difference and variance information, and the G category grid point cloud is marked as ground point in advance, reducing the points that need to be estimated by Bayesian probability, greatly improving the segmentation speed. Specifically, all sub-grid cells are initialized as grids that may contain ground points (PG); the distance from the lowest point of each sub-grid cell to the sensor is calculated and its height attribute is updated, when the height of the lowest point of the grid is less than the set lower limit of height, the lower limit of height is taken as the height attribute of the grid, when the height of the lowest point of the grid is greater than the set upper limit of height, the height of the sensor is taken as the height attribute of the grid, and when it is between the upper and lower limits of height, the height of the lowest point of the grid is taken as its height attribute.

[0017] The sub-grid cells that meet the following conditions are marked as ground grids (G): the number of point clouds in the grid exceeds the set minimum number of points, the height variance of the grid is greater than zero and less than threshold one, and the height attribute of the grid is less than the preset threshold two; dynamically adjust threshold one based on the distance of the grid; mark the sub-grid cells that meet any of the following conditions as non-ground grids (NG): condition one, the distance attribute of the grid is greater than the preset threshold three, and the height attribute of the grid is equal to the height of the sensor; condition two, the distance attribute of the grid cell is greater than the preset threshold three, and the height difference between the highest point and the lowest point in the grid is greater than the preset threshold four.

[0018] Due to the sparsity of point cloud distribution in the far distance area, the height variance increases, in order to maintain high recognition rate in the near distance area and higher tolerance to point cloud height variance in the far distance area, height variance threshold must be adaptively adjusted: the value obtained by multiplying the preset distance scaling factor by the grid distance is taken as the height variance threshold one of different grids, to realize adaptive adjustment of threshold one based on distance, thereby improving the accuracy and robustness of ground grid recognition in the far distance area.

[0019] Further, step S4 is explained in detail:

[0020] PG sub-grid cell ground point estimation based on Bayesian probability model: Generally, the plane point cloud can be distinguished according to the angle characteristics of the point cloud, but in the outdoor environment, there are usually non-ground structures such as car hoods or roofs, in order to reliably distinguish ground points, the joint probability of the angle probability and the height difference probability of each point is calculated to estimate the possibility of being a ground point; the included angle between the normal vector of the point and the vertical direction of the ground is calculated, if the included angle is less than a preset boundary angle value, the angle probability of the point is set to 1, indicating that it is more likely to belong to the ground point; otherwise, it is set to 0, according to experience and point cloud distribution characteristics, different boundary angle values are set in different regions; the difference between the highest point height and the lowest point height in the grid is calculated as the grid height difference, the height difference is reduced by a threshold value five, to obtain a difference value, the exponential form of the difference value is calculated through an exponential function, and after being added by one, the reciprocal is taken as the height difference probability; the angle probability and the height difference probability are multiplied to obtain the ground point probability, and the points with a ground point probability greater than a preset threshold value six are estimated as ground points.

[0021] Considering the sparsity of the long-distance point cloud, in order to adapt to the height difference requirements at different distances, the height difference threshold value needs to be adaptively adjusted: the threshold value five is set as a dynamic function related to the distance of the point to the sensor, and the value is calculated through an exponential form. Specifically, the threshold value five is equal to an initial value multiplied by the exponential growth of the distance, and the growth rate is controlled by a constant. Among them, the initial value determines the size of the threshold value five at a short distance, and the growth rate controls the growth speed of the threshold value five with the increase of the distance. This definition can dynamically adjust the tolerance range of the height difference according to the distance of the point to the sensor, and adapt to the point cloud characteristics in different distance regions.

[0022] The application provides an intelligent wheelchair ground point cloud segmentation system, which comprises:

[0023] A laser radar scanning module is arranged for collecting three-dimensional point cloud information in the moving process of the wheelchair.

[0024] An original point cloud preprocessing module is arranged for filtering the collected point cloud.

[0025] A multi-resolution division grid module is arranged for adopting differentiated resolution division for the point cloud in different radial intervals and angle intervals, so that the ground segmentation precision is improved.

[0026] A grid attribute storage module is arranged for calculating the lowest point, the highest point, the height variance and the point cloud quantity of the grid when the grid is divided, so as to store the grid attributes.

[0027] A grid classification module is arranged for classifying the subdivided sub-grid units, finding the grid with all ground points, and reducing the number of ground points to be estimated.

[0028] a Bayesian probability ground point estimation module, configured to perform probability estimation on the ground point grid classified in the grid classification module to obtain ground points in the grid;

[0029] an estimated ground point merging module, configured to merge all estimated ground points to obtain a set of estimated ground point clouds.

[0030] The present application has the following advantages:

[0031] (1) The present method uses different resolutions to divide the grid for the point cloud in different radial intervals and angle intervals, so that the sparsity problem of the point cloud can be adapted when processing the point cloud in a long distance area, the ground points in the distance are accurately identified, and the ground segmentation accuracy is improved.

[0032] (2) Ground estimation on a large amount of point cloud data is time-consuming, in order to improve the segmentation efficiency of the algorithm, the present method classifies the sub-grid unit based on height difference and variance information, marks the grid point cloud of the G category as a ground point in advance, to reduce the points that need to be estimated by Bayesian probability, and greatly improves the segmentation speed; the height variance threshold adaptive adjustment strategy is adopted to improve the accuracy and robustness of the grid classification in the long distance area.

[0033] (3) The present method estimates the possibility of each point being a ground point by calculating the joint probability of the angle probability and the height difference probability of each point, greatly reduces the misjudgment of the non-ground structure such as the engine hood or the roof of the car in the PG grid, and makes the ground segmentation result have high reliability. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 a flow chart of the intelligent wheelchair ground point cloud segmentation algorithm described in the present application;

[0035] Figure 2 a multi-resolution point cloud grid division diagram based on radial intervals and angle intervals described in the present application;

[0036] Figure 3 a system diagram of the intelligent wheelchair ground point cloud segmentation system described in the present application. DETAILED DESCRIPTION

[0037] The present application will be further illustrated below in conjunction with the drawings and specific embodiments, and it should be understood that the following specific embodiments are only used to illustrate the present application and not to limit the scope of the present application.

[0038] As shown in the figure, the intelligent wheelchair ground point cloud segmentation method described in the present application comprises the following steps:

[0039] Step 1: Collect point cloud P using the lidar on the wheelchair. Preprocess the raw point cloud data. Specifically, use a multi-condition filtering method based on height and distance to filter the point cloud data. It will reject point clouds with z values ​​below a set threshold within a certain distance, thereby effectively reducing outliers.

[0040] The point cloud is rasterized using a multi-resolution partitioning method based on radial and angular intervals:

[0041] Step 2: In the radial direction, based on the distribution characteristics of the point cloud around the sensor, divide M into regions at unequal intervals. Z There are three concentric radial regions, each with specific minimum and maximum radial boundaries (L). min，m L max，m ), m = 1, 2, 3, ..., M Z L max，m =L min，m+1 That is, the maximum boundary of each radial region is used as the minimum boundary of the next radial region, and the minimum boundary of the first radial region is the closest distance for radar scanning, and the Mth radial region... Z The maximum boundary of each radial region is the farthest distance of the radar scan, L. max，MZ =L max L min，1 =L min .like Figure 2 The given example divides the region into four concentric radial regions: S202, S204, S206, and S208.

[0042] Step 3: In the angular direction, divide N according to the point cloud distribution characteristics and requirements. Z An angle range, such as Figure 1 As shown, each interval has minimum and maximum angle boundaries (θ). min，n， θ max，n ), n = 1, 2, 3, ..., N Z , where θ max，n =θ min，n+1 That is, the maximum boundary of each angular region is used as the minimum boundary of the next angular region. For example... Figure 2 The given example divides the area into four angular regions: S210, S212, S214, and S216.

[0043] Step 4: Further subdivide the grid cells: After the subdivisions in steps 2 and 3, we can obtain M. z ×N z For each grid cell, the corresponding set of grid cells is defined as follows:

[0044]

[0045] Z m,nrepresents the grid cell of the m-th radial interval and the n-th angular interval. Z m,n is defined as: Z m,n = {p k ∈ Z

[0046] P|L min，m ≤ ρ k ≤ L max，m , θ min，n ≤ θ k ≤ θ max，n}

[0047] wherein is the radial distance of the point p k , θ k = arctan2(y k / x k ). Each grid cell Z m,n is further subdivided using different resolutions in radial and angular directions, resulting in respective sub-grid cells, the resolution in radial and angular directions of each grid cell is defined as:

[0048]

[0049] wherein N r,m,n and N θ,m,n are the number of grids in radial and angular directions of the corresponding grid cell Z m,n respectively; each subdivided sub-grid cell is denoted by R i，j,m,n , which contains all points p k that satisfy the following conditions:

[0050] R i,j,m,n

[0051] = {p k ∈ Z m,n | ΔL m,n · (i-1) ≤ ρ k - L min，m ≤ ΔL m,n · i, Δθ m,n · (j-1) ≤ θ k - θ min,n ≤ Δθ m,n · j}

[0052] wherein i = 1, 2, 3, …, N r,m,n , j = 1, 2, 3, …, N θ,m,n , m = 1, 2, 3, …, (M Z - 1), n = 1, 2, 3, …, (N Z - 1); as Figure 2In the given example, the shaded part S218 is divided by combining S206 and S216, and the shaded part S220 is divided by combining S208 and S216. S218 and S220 are respectively subdivided into 2 and 4 radial grids in the radial direction, and respectively subdivided into 6 and 4 angle grids in the angle direction. It can be seen that the resolutions used by the two are different.

[0053] Step 5: Record the sub-grid cell attributes: During the process of gridding, the minimum point height z min , the maximum point height z max , the height variance and the point cloud number N of each sub-grid cell are calculated as the attributes of the sub-grid cell and stored.

[0054] Classify the sub-grid cells based on the height difference and variance information:

[0055] Step 6: Give the classification definition, all sub-grid cells can be classified into Ground Cell (G) which contains all ground points, Potential Ground Cell (PG) which may contain ground points, and Non-Ground Cell (NG) which contains no ground points. Initialize all sub-grid cells and mark them as class PG first.

[0056] Step 7: Calculate the Euclidean distance d from the sensor to the minimum point of each sub-grid cell as the distance attribute of the grid.

[0057] d = dis(min(ρ k ))

[0058] Step 8: Update the height attribute H c of the sub-grid cell according to the minimum point height, to ensure that the height attribute of the sub-grid cell is reasonable. Specifically:

[0059]

[0060] where h sensor is the sensor height, h min and h max are the set minimum and maximum heights of the grid cell.

[0061] Step 9: Introduce an adaptive variance threshold t v : The variance threshold t v is defined as an adaptive function related to the distance d:

[0062] t v = k·d

[0063] where k is a distance scaling factor that scales the distance to the variance threshold and is uniformly applicable to all cells.

[0064] Step 10: Label the sub-grid cells belonging to class G: Specifically, label a sub-grid cell as G if it satisfies the following three conditions: (1) the number of point clouds N in the grid exceeds the set minimum number of points N min ; (2) the height variance of the grid satisfies (3) the height of the grid H C is less than the threshold value h mid . All point clouds in the sub-grid cells of class G are estimated as ground points, and their collection is denoted as

[0065] Step 11: Label the sub-grid cells belonging to class NG: Specifically, label a sub-grid cell as NG if it satisfies any one of the following two conditions: (1) the distance d of the grid cell is greater than the preset maximum distance threshold L max,3 , and the grid height is equal to the sensor height h sensor ; (2) the grid distance d exceeds L max,3 and the height difference Δz = z max -z min is greater than the height difference threshold Δh max . When the grid distance d exceeds L max,3 and the height difference Δz = z max -z min is greater than the height difference threshold Δh max , the grid is also considered as a non-ground grid.

[0066] Based on the Bayesian probability model, estimate the ground points in the sub-grid cells classified as PG:

[0067] Step 12: Give the Bayesian probability model: Let be the likelihood function of the probability model, where Ω represents the parameter set of all models, be the observed point cloud data. Assuming that the point cloud distribution in each sub-grid cell PG is independent of each other, the overall probability of the model can be represented as:

[0068]

[0069] where ω i and represent the parameters and observations of the i-th point in the grid, respectively.

[0070] In each PG grid, estimate the probability of each point belonging to the ground based on its angle and height difference features. Specifically, the probability of the ground point of point p PG,i is jointly estimated by its angle probability α(θ i ) and height difference probability β(Δz i ,d i ):

[0071]

[0072] Step 13: Calculate the angle probability α(θ i ):We first calculate the normal vector n i of each point and define the angle θ i between the normal vector and the vertical direction z=[0,0,1] on the ground. τ The normal vector of the ground point cloud is often perpendicular to the X-Y plane of the sensor frame. The calculation method is as follows:

[0073]

[0074] θ τ is the boundary value, that is, the larger θ τ , the more conservative the standard. According to experience and the distribution characteristics of the point cloud, different boundary values θ i can be set in different areas to ensure the reliability of the evaluation results.

[0075] Step 14: Give an adaptive height difference threshold function: We define an adaptive height difference threshold κ(d i )=γ·exp(δ·d i ), which is used to adapt to the requirements of height difference at different distances, where γ and δ are constants that control the initial value and growth rate of the threshold.

[0076] Step 15: Calculate the height difference probability β(Δz i ,d i ): Calculate the height difference probability combined with the adaptive height difference threshold function in Step 14, as follows:

[0077]

[0078] where the height difference Δz i is defined as the difference between the height z min of the current point and the height z i of the lowest point in the grid, d i is the distance attribute of the i-th grid, and k(d PG,i ) is the adaptive height difference threshold.

[0079] Step 16: After Steps 12 to 15, we can get the estimated ground point cloud in all PG sub-grid units, which is represented as:

[0080]

[0081] where p i is the i-th point in PG, and [·] represents the Iverson bracket, which returns true when the condition is met and divides the corresponding point into . Finally, combine with the Merging, we can get the estimated ground point cloud output by the ground point cloud segmentation algorithm

[0082] It should be noted that the above content only illustrates the technical idea of the present application, and cannot be used to limit the protection scope of the present application. For ordinary skilled persons in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which fall within the protection scope of the claims of the present application.

Claims

1. A method for ground point cloud segmentation for intelligent wheelchairs, characterized in that, The method includes the following steps: S1, via installation on a wheelchair at a height of h sensor The lidar collects point clouds, preprocesses the raw point cloud data, and filters the point cloud data according to preset distance and height thresholds; S2. The point cloud is divided into raster units using a multi-resolution partitioning method based on radial and angular intervals to obtain sub-raster units. The partitioning method uses different resolutions to divide the raster for point clouds in different radial and angular intervals to adapt to the sparsity of the point cloud. During the rasterization process, the attributes of each sub-raster unit are calculated and stored. S3. Classify sub-grid cells based on height difference and variance information into three categories: G, NG, and PG. The G category consists entirely of ground points; the NG category consists entirely of non-ground points; and the PG category consists of grid cells that may contain ground points. S4. Calculate the ground point probability based on the Bayesian probability model, estimate the ground points in the point cloud of the PG grid, merge all ground points to obtain the estimated ground point cloud and output it; the merging method is to merge the ground points of category G with the ground points estimated by Bayesian method. Specifically, it includes: S41: Calculate the angular probability α(θ) i Specifically, first calculate the normal vector n of each point. i And define the angle between the z-axis and the direction z = [0,0,1] perpendicular to the ground as θ. i The normal vectors of the ground point cloud are often perpendicular to the XY plane of the sensor frame; the angle probability is calculated as follows: Where θ τ Different boundary values ​​θ are set for different regions as angular boundary values. τ To ensure the reliability of its evaluation results; S42: Calculate the height difference probability β(Δz) based on height difference and distance attributes. i ,d i The calculation method is as follows: Among them, the height difference Δz i Defined as the height z of the current point i The height z of the lowest point in the grid min The difference, d i Let κ(d) be the distance attribute of the i-th grid cell. i The adaptive height difference threshold is calculated using an exponential form: k(d i )=γ·exp(δ·d i ) Where γ and δ are constants controlling the initial value of the threshold and the growth rate; S43: Based on the angular probability α(θ) of S41 and S42 i ) and height difference probability β(Δz) i ,d i The probability p(Z) of the ground point is calculated by joint estimation. i |ω i ), such as point p PG,i If the probability of a ground point is greater than a preset probability threshold, then that point is estimated to be a ground point, and the corresponding set of ground points is represented as follows: Where, p PG,i Let be the i-th point in PG, and [·] represent Iversonbracket. Return true when the condition is met, and assign the corresponding point to PG. middle; S44: Output all ground point clouds Specifically, For S43 With S34 The union of; 2. The method for ground point cloud segmentation for intelligent wheelchairs according to claim 1, characterized in that: The grid division method described in step S2 specifically includes: S21: In the radial direction, based on the distribution characteristics of the point cloud around the sensor, M is divided at unequal intervals. Z There are three concentric radial regions, each with a specific minimum radial boundary L. min,m and the maximum radial boundary L max,m m = 1, 2, 3, ..., M Z L max,m =L min,m+1 That is, the maximum boundary of each radial region is used as the minimum boundary of the next radial region, and the minimum boundary of the first radial region is the closest distance for radar scanning, and the Mth radial region... Z The maximum boundary of each radial region is the farthest distance of the radar scan, L. max,MZ =L max ,L min,1 =L min ; S22: In the angular direction, based on the point cloud distribution characteristics and requirements, divide N... Z There are angular intervals, each with a minimum angular boundary θ. min,n and the maximum angular boundary θ max,n n = 1, 2, 3, ..., N Z , where θ max,n =θ min,n+1 That is, the maximum boundary of each angular region is used as the minimum boundary of the next angular region; the radial region and the angular interval are combined to obtain M. z ×N z For each grid cell, the corresponding set of grid cells is defined as follows: Z m,n Z represents the grid cell representing the m-th radial region and the n-th angular interval; m,n Defined as: Z m,n ={p k ∈ P|L min,m ≤ρ k ≤L max,m ,i min,n ≤θ k ≤θ max,n } in For point p k radial distance, θ k =arctan2(y k / x k ); S23: For each grid cell Z m,n Further subdivision is performed using different resolutions in the radial and angular directions to obtain individual sub-grid cells. The resolutions in the radial and angular directions of each grid cell are defined as follows: Where, N r,m,n and N θ,m,n These are the corresponding grid cells Z. m,n The number of grid cells in the radial and angular directions; each subdivided subgrid cell is represented by R. i,j,m,n This indicates that it contains all points p that satisfy the following conditions. k : R i,j,m,n ={p k ∈Z m,n |ΔL m,n ·(i-1)≤ρ k -L min,m ≤ΔL m,n ·i,Δθ m,n ·(j-1)≤θ k -θ min,n ≤Δθ m,n ·j} Among them, i=1,2,3,…,N r,m,n , j = 1, 2, 3, …, N θ,m,n , m = 1, 2, 3, ..., (M Z -1), n ​​= 1, 2, 3, ..., (N Z -1); S24: During the rasterization process, calculate the lowest point height z of each sub-raster cell. min The highest point height z max Height variance The number of point clouds, N, is used as an attribute of the sub-raster cell and stored.

3. The method for ground point cloud segmentation for intelligent wheelchairs according to claim 1, characterized in that: The classification method described in step S3 specifically includes: S31: Sub-raster cell initialization; all sub-raster cells are first marked as PG category; S32: Calculate the Euclidean distance d from the lowest point of each sub-grid cell to the sensor as the distance attribute of the grid; d=dis(min(ρ k )) S33: Update the height attribute H of the sub-grid cell based on the lowest point height. C To ensure that the height attribute of the sub-grid cell is reasonable, specifically: Where h sensor h is the sensor height. min and h max The minimum and maximum heights of the set grid cells; S34: Mark the sub-raster cells belonging to category G; specifically, mark the sub-raster cells that meet the following three conditions as G: (1) The number of point clouds N in the raster exceeds the set minimum number of points N. min (2) Grid height variance satisfy (3) Grid height H C Less than the threshold h mid Point clouds in all sub-raster cells of category G are estimated as ground points, and their set is represented as... Where t v For adaptive variance threshold, t v = k·d, where k is the distance scaling factor and d is the raster distance attribute; S35: Mark the sub-grid cells belonging to category NG; specifically, mark the sub-grid cells that meet either of the following two conditions as NG: (1) The distance d of the grid cell is greater than the preset maximum distance threshold L. max,3 And the grid height is equal to the sensor height h. sensor (2) The grid distance d exceeds L max,3 And the height difference Δz=z max -z min Greater than the height difference threshold Δh max At that time, the grid will also be regarded as a non-ground grid.

4. The method for ground point cloud segmentation for intelligent wheelchairs according to claim 1, characterized in that: Including an intelligent wheelchair ground point cloud segmentation system: The lidar scanning module is used to collect three-dimensional point cloud information during the movement of the wheelchair; The raw point cloud preprocessing module is used to filter the acquired point clouds; The multi-resolution raster segmentation module is used to segment point clouds with different resolutions for different radial and angular intervals, thereby improving the accuracy of ground segmentation. The raster attribute storage module is used to calculate the lowest point, highest point, height variance, and number of points in a raster when dividing it into raster attributes, and store them as raster attributes. The grid classification module is used to classify the subdivided sub-grid units. It first finds grids that are all ground points, thus reducing the number of ground points that need to be estimated later. The Bayesian probabilistic ground point estimation module is used to perform probability estimation on the potential ground point grids classified by the grid classification module to obtain the ground points therein. The estimated ground point merging module is used to merge all estimated ground points to obtain an estimated ground point cloud set.

Images