Road edge detection method and device, vehicle control unit and computer readable storage medium

Abstract

The embodiment of the invention provides a road edge detection method and device, a vehicle control unit and a computer readable storage medium, and relates to the technical field of target detection, and the method comprises the steps: obtaining a three-dimensional point cloud space, segmenting the three-dimensional point cloud space based on a cylindrical coordinate system, and obtaining a target volume element set composed of a plurality of target volume elements; determining a volume element category corresponding to each target volume element in the target volume element set; determining a road edge point set from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements; and fitting the road edge points in the road edge point set to obtain a road edge line. The problem of low road edge recognition accuracy caused by direct recognition of road edge line features is avoided, the target volume elements are obtained through cylindrical coordinate system segmentation, three-dimensional points contained in the target volume elements are uniform, the accuracy of target volume element classification is improved, and the accuracy of road edge recognition is improved. And the accuracy of road edge identification based on the volume element category is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to a curb detection method, device, vehicle controller and computer-readable storage medium. Background Art

[0002] With the rapid development of science and technology, more and more vehicles are equipped with autonomous driving or assisted driving functions. In autonomous driving or assisted driving systems, vehicles need to perceive the surrounding environment and extract key information from it. To ensure the normal driving of the vehicle, it is necessary to correctly perceive the roadside.

[0003] In the related technology, there are two main ways to detect the curb. The first is to use laser radar to scan the surrounding environment, extract the curb features in the surrounding environment (such as the height difference of the curb), and directly identify the curb from the surrounding area. The second is to use a visual camera to capture a two-dimensional image of the surrounding environment, recognize the two-dimensional image based on the image features of the curb, and directly identify the curb from the two-dimensional image.

[0004] However, in actual environments, the types and shapes of curbs vary, and there may be many defects on the curbs. It is difficult to accurately identify the curbs using the above solution, resulting in a low curb recognition rate and poor accuracy of the recognition result. Summary of the invention

[0005] In view of this, the present invention aims to propose a curb detection method, device, vehicle controller and computer-readable storage medium to solve the problem that the prior art has a low curb recognition rate and poor recognition result accuracy.

[0006] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0007] In a first aspect, the present invention provides a road edge detection method, the method comprising:

[0008] Acquire a three-dimensional point cloud space, and segment the three-dimensional point cloud space based on a cylindrical coordinate system to obtain a target volume element set consisting of a plurality of target volume elements; wherein a radial distance of the target volume element in the three-dimensional point cloud space is proportional to a volume of the target volume element;

[0009] Determine the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category;

[0010] Determine a set of curb points from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements respectively;

[0011] The roadside points in the roadside point set are fitted to obtain a roadside line.

[0012] Optionally, segmenting the three-dimensional point cloud space based on a cylindrical coordinate system to obtain a target volume element set consisting of a plurality of target volume elements includes:

[0013] Mapping the three-dimensional point cloud space into a cylindrical coordinate system;

[0014] The cylindrical coordinate system is divided to obtain a plurality of volume elements; wherein the plurality of target volume elements have the same radial length, angular arc and height in the cylindrical coordinate system;

[0015] Based on the position of the volume element in the cylindrical coordinate system, multiple target volume elements are determined from the multiple volume elements to obtain a target volume element set consisting of the multiple target volume elements; wherein the number of the multiple target volume elements is less than the number of the multiple volume elements.

[0016] Optionally, the determining a plurality of target volume elements from the plurality of volume elements based on the positions of the volume elements in the cylindrical coordinate system to obtain a target volume element set consisting of the plurality of target volume elements comprises:

[0017] Determine the number of three-dimensional points contained in each volume element in each stacked volume element set; wherein the stacked volume element set is formed by stacking a plurality of volume elements having the same radial distance and angular radian;

[0018] The volume element in the stacked volume element set whose number of three-dimensional points is greater than or equal to a preset number and whose bottom layer is determined as a target volume element, thereby obtaining a target volume element set consisting of target volume elements corresponding to the plurality of stacked volume element sets.

[0019] Optionally, determining a set of curb points from the three-dimensional point cloud based on the volume element categories respectively corresponding to the target volume elements comprises:

[0020] Determining a road surface area and a non-road surface area from the three-dimensional point cloud space based on the volume element categories corresponding to the target volume elements respectively;

[0021] A three-dimensional point in the road surface area adjacent to the non-road surface area is determined as a roadside point.

[0022] Optionally, determining the three-dimensional point in the road surface area adjacent to the non-road surface area as a roadside point includes:

[0023] Searching the road surface area in both sides of the vehicle respectively, and determining the road surface points at the edge positions of the road surface area as curb seed points;

[0024] Cluster growth is performed on the curb seed points to obtain a curb point set including the curb seed points and curb growth points.

[0025] Optionally, performing cluster growth on the roadside seed points to obtain a roadside point set including the roadside seed points and roadside growth points includes:

[0026] determining a road centerline according to the road area;

[0027] Taking the curb seed point as the starting point, search for adjacent road surface points in the extension direction of the road centerline, and determine the road surface points adjacent to the non-road area and whose distance from the non-road area is less than a preset distance as curb growth points, to obtain a curb point set including the curb seed point and the curb growth point.

[0028] Optionally, fitting the roadside points in the roadside point set to obtain a roadside line includes:

[0029] Performing denoising on the roadside points in the roadside point set, removing the noise points in the roadside point set, and obtaining a target roadside point set;

[0030] The roadside points in the target roadside point set are fitted to obtain a roadside line.

[0031] In a second aspect, the present invention provides a road edge detection device, the device comprising:

[0032] A volume element module is used to obtain a three-dimensional point cloud space, segment the three-dimensional point cloud space based on a cylindrical coordinate system, and obtain a target volume element set consisting of a plurality of target volume elements; wherein a radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element;

[0033] A category module, used to determine the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category;

[0034] A roadside point module, used for determining a roadside point set from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements respectively;

[0035] The roadside line module is used to fit the roadside points in the roadside point set to obtain the roadside line.

[0036] Optionally, the volume element module includes:

[0037] A mapping submodule, used for mapping the three-dimensional point cloud space into a cylindrical coordinate system;

[0038] A segmentation submodule, used for segmenting the cylindrical coordinate system to obtain a plurality of volume elements; wherein the plurality of target volume elements have the same radial length, angular arc and height in the cylindrical coordinate system;

[0039] The volume element submodule is used to determine multiple target volume elements from the multiple volume elements based on the positions of the volume elements in the cylindrical coordinate system, and obtain a target volume element set composed of the multiple target volume elements; wherein the number of the multiple target volume elements is less than the number of the multiple volume elements.

[0040] Optionally, the volume element submodule includes:

[0041] A quantity submodule, used to determine the number of three-dimensional points contained in each volume element in each stacked volume element set; wherein the stacked volume element set is formed by stacking a plurality of volume elements having the same radial distance and angular radian;

[0042] The volume element set submodule is used to determine the volume element with the number of three-dimensional points greater than or equal to the preset number and the bottom layer in the stacked volume element set as the target volume element, and obtain a target volume element set composed of target volume elements corresponding to multiple stacked volume element sets respectively.

[0043] Optionally, the roadside point module includes:

[0044] A region submodule, configured to determine a road surface region and a non-road surface region from the three-dimensional point cloud space based on the volume element categories respectively corresponding to the target volume elements;

[0045] The roadside point determination submodule is used to determine the three-dimensional points in the road surface area that are adjacent to the non-road surface area as roadside points.

[0046] Optionally, the roadside point determination submodule includes:

[0047] A search submodule, for searching the road surface area in the directions of both sides of the vehicle, and determining the road surface points at the edge of the road surface area as curb seed points;

[0048] The growth submodule is used to perform cluster growth on the curb seed points to obtain a curb point set including the curb seed points and curb growth points.

[0049] Optionally, the growth submodule includes:

[0050] A road centerline submodule, used to determine a road centerline according to the road area;

[0051] The curb point growth submodule is used to search for adjacent road surface points in the extension direction of the road surface centerline with the curb seed point as the starting point, and determine the road surface points adjacent to the non-road surface area and whose distance from the non-road surface area is less than a preset distance as curb growth points, thereby obtaining a set of curb points including the curb seed point and the curb growth point.

[0052] Optionally, the roadside module includes:

[0053] A denoising submodule, used for performing denoising processing on the roadside points in the roadside point set, removing the noise points in the roadside point set, and obtaining a target roadside point set;

[0054] The fitting submodule is used to fit the roadside points in the target roadside point set to obtain a roadside line.

[0055] In a third aspect, the present invention provides a vehicle controller, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned curb detection method when executing the computer program.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium, which, when the instructions in the computer-readable storage medium are executed by a processor of a vehicle controller, enables the vehicle controller to execute the above-mentioned curb detection method.

[0057] In a fifth aspect, the present invention provides a vehicle, comprising the above-mentioned vehicle controller.

[0058] Compared with the prior art, the curb detection method, device, vehicle controller and computer-readable storage medium described in the present invention have the following advantages:

[0059] In summary, an embodiment of the present invention provides a method for detecting a curb, including: obtaining a three-dimensional point cloud space, segmenting the three-dimensional point cloud space based on a cylindrical coordinate system, and obtaining a target volume element set consisting of multiple target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; determining the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category; determining a curb point set from the three-dimensional point cloud based on the volume element category corresponding to the target volume element; and fitting the curb points in the curb point set to obtain a curb line. In the embodiment of the present invention, it is possible to determine whether the target volume element is a road surface category, and determine the curb line from the target point cloud according to the characteristic of whether the target volume element is a road surface category, thereby avoiding the problem of low curb recognition accuracy caused by directly identifying the curb line feature, and obtaining the target volume element by segmenting the cylindrical coordinate system, so that the three-dimensional points contained in the target volume element can be relatively uniform, thereby improving the accuracy of classifying the target volume element, thereby further improving the accuracy of curb recognition based on the volume element category. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings constituting 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 an improper limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 A flowchart of a method for detecting a road edge provided by an embodiment of the present invention;

[0062] Figure 2 A schematic diagram of a target volume element set provided by an embodiment of the present invention;

[0063] Figure 3 A schematic diagram of a roadside point provided by an embodiment of the present invention;

[0064] Figure 4 A flowchart of another method for detecting a curb provided by an embodiment of the present invention;

[0065] Figure 5 A top view of a volume element provided by an embodiment of the present invention;

[0066] Figure 6 A structural block diagram of a curb detection device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0068] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0069] Reference Figure 1 , Figure 1 A flowchart of the steps of a curb detection method provided by an embodiment of the present invention.

[0070] Step 101, obtain a three-dimensional point cloud space, segment the three-dimensional point cloud space based on a cylindrical coordinate system, and obtain a target volume element set consisting of multiple target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element.

[0071] The point data set on the surface of an object obtained by a measuring instrument is called a point cloud. For example, a 3D coordinate measuring machine can be used to obtain a relatively sparse point cloud, while a 3D laser scanner or a photographic scanner can be used to obtain a relatively dense point cloud.

[0072] Taking the application of three-dimensional point cloud in the field of autonomous driving as an example, in order to detect targets around the vehicle, a laser radar can be installed on the vehicle, and the laser radar can emit dense laser beams to the surrounding environment, and then generate a three-dimensional point cloud of the surrounding environment based on the reflection of the laser beam on the object. It should be noted that in the embodiment of the present application, the method of obtaining a three-dimensional point cloud through a laser radar is not specifically limited. For example, the three-dimensional point cloud of the environment in front of the vehicle can be obtained by the front laser radar of the vehicle, and a panoramic three-dimensional point cloud around the vehicle can be generated by multiple laser radars arranged around the vehicle. The panoramic three-dimensional point cloud around the vehicle can also be obtained by a rotating laser radar arranged on the roof. Similarly, in the embodiment of the present application, the method of obtaining a three-dimensional point cloud is not specifically limited. For example, the three-dimensional point cloud of the vehicle's surrounding environment can be obtained by the above-mentioned laser radar, and the distance and direction between multiple points in the surrounding environment and the vehicle can be calculated by a binocular stereo vision solution to generate a three-dimensional point cloud of the vehicle's surrounding environment.

[0073] Since each point in the three-dimensional point cloud contains orientation information relative to the point cloud shooting party (such as a vehicle), after obtaining the three-dimensional point cloud, each point in the three-dimensional point cloud can be projected into a space with the point cloud shooting party as the origin to obtain a three-dimensional point cloud space.

[0074] In order to facilitate the feature extraction and classification of point cloud features in the three-dimensional point cloud space, the three-dimensional point cloud space can be segmented to obtain multiple smaller volume elements, so that the three-dimensional points in the same volume element can be classified as a whole to improve the classification efficiency. Among them, the volume element (Volume Pixel, Voxel) can be referred to as voxel, which can be expressed by stereo rendering or extracting polygonal isosurfaces with a given threshold contour. Conceptually similar to the smallest unit of two-dimensional space - pixel, pixel is used in the image data of two-dimensional computer images. Voxel can be used to represent the spatial unit in three-dimensional space.

[0075] Generally speaking, since the laser beam emitted by the LiDAR can only propagate in a straight line, the farther away the laser beam is from the laser emission position, the lower the density of the laser beam that can be received by the surface of the object, resulting in a lower point cloud density at a distance from the point cloud shooting party. If the equal volume segmentation method is used to segment the 3D point cloud space into equal volume elements, then in theory, the farther away the point cloud shooting party is, the fewer 3D points will be contained in the volume element, which is likely to produce a large number of empty voxels that do not contain any 3D points. These empty voxels cannot be classified due to the lack of 3D points, affecting the subsequent point cloud classification effect.

[0076] In an embodiment of the present application, after obtaining the three-dimensional point cloud space, the three-dimensional point cloud space can be segmented based on a cylindrical coordinate system to obtain a target volume element set consisting of multiple target volume elements.

[0077] Reference Figure 2 , Figure 2 A schematic diagram of a target volume element set provided by an embodiment of the present application is shown in FIG. Figure 2 As shown in FIG. 1 , the three-dimensional point cloud space can be segmented by a cylindrical coordinate system to obtain multiple target volume elements. Among these target volume elements, the target volume element that is farther away from the point cloud shooting party occupies a larger volume, which can offset the effect of the three-dimensional points becoming sparse due to the distance to a certain extent. Therefore, the above segmentation method can make the radial distance of the target volume element in the three-dimensional point cloud space proportional to the volume of the target volume element. In other words, the number of three-dimensional points in the obtained volume element is relatively consistent, avoiding the appearance of a large number of empty voxels, which helps to improve the subsequent classification effect of the point cloud. Among them, the radial distance represents the distance between the target volume element and the axis of the cylindrical coordinate system.

[0078] Step 102: Determine the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category.

[0079] After obtaining the target volume element set, the target volume element can be subjected to category detection to determine the volume element category corresponding to each volume element, wherein the volume element category of the volume element can characterize the object category corresponding to the three-dimensional point in the volume element. For example, if the three-dimensional point contained in a volume element is formed by a laser beam reflected by a vehicle, the volume element category corresponding to the volume element can be a vehicle; if the three-dimensional point contained in a volume element is formed by a laser beam reflected by the ground, the volume element category corresponding to the volume element can be a road surface. The volume element category can include a road surface category and a non-road surface category, and the road surface category represents the category of the road surface on which the vehicle is currently traveling.

[0080] In the embodiment of the present application, a point cloud classification model obtained by training a convolutional neural network can be used to determine the volume element category corresponding to each target volume element. Specifically, after obtaining the target volume element, the target volume element can be input into the point cloud classification model to obtain the volume element category corresponding to the target volume element output by the point cloud classification model; or the entire three-dimensional point cloud (including the data of each three-dimensional point in the three-dimensional point cloud) can be directly input into the point cloud classification model, and the point cloud classification model can be used to segment the point cloud to obtain the target volume element, and determine the volume element category of the target volume element.

[0081] In an embodiment of the present application, a point cloud classification model obtained through training of a convolutional neural network may be used to perform semantic segmentation on a target volume element set, and to determine road volume elements and non-road volume elements from the target volume element set.

[0082] Step 103: determining a set of curb points from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements.

[0083] After obtaining the volume element category corresponding to each target volume element, a set of curb points can be determined from the three-dimensional point cloud based on the volume element category, wherein the curb point represents a three-dimensional point formed at the curb position of the road.

[0084] Since the curb is usually located at the junction of two different areas, such as the junction of the road surface and the green belt, the junction of the road surface and the sidewalk, the junction of the road surface and the guardrail, etc. Therefore, a volume element category pair can be set in advance, and according to the volume element category pair, an adjacent target volume element matching the volume element category pair is searched from multiple target volume elements, and the adjacent target volume elements are determined as the target volume element pair, so that the three-dimensional point at the intersection of two target volume elements in the adjacent target volume element pair is determined as the curb point.

[0085] For example, refer to Figure 3 , Figure 3 A schematic diagram of a roadside point provided in an embodiment of the present application is shown. Figure 3 As shown, the pavement category and the green belt category can constitute a volume element category pair. If the volume element category of a target volume element 31 is a pavement category, and the volume element category of a target volume element 32 is a green belt category, and the target volume element 31 is adjacent to the target volume element 32, then the target volume element 31 and the target volume element 32 can be determined as a target volume element pair, and the three-dimensional point 33 whose distance from the intersection line of the target volume element 31 and the target volume element 32 is less than or equal to the preset distance can be determined as a curb point.

[0086] After all the target volume element pairs in the target volume element set, a roadside point set is obtained based on all the target volume element pairs.

[0087] Step 104, fitting the roadside points in the roadside point set to obtain a roadside line.

[0088] On real roads, curbs may be blocked by obstacles, such as vehicles parked on both sides of the road, resulting in no curb information in the three-dimensional point cloud taken on the road. As a result, the curb points determined from the three-dimensional point cloud are not continuously distributed. Therefore, after determining the curb point set, the curb points in the curb point set can also be fitted to obtain a continuous curb line.

[0089] In summary, an embodiment of the present invention provides a method for detecting a curb, including: obtaining a three-dimensional point cloud space, segmenting the three-dimensional point cloud space based on a cylindrical coordinate system, and obtaining a target volume element set consisting of multiple target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; determining the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category; determining a curb point set from the three-dimensional point cloud based on the volume element category corresponding to the target volume element; and fitting the curb points in the curb point set to obtain a curb line. In the embodiment of the present invention, it is possible to determine whether the target volume element is a road surface category, and determine the curb line from the target point cloud according to the characteristic of whether the target volume element is a road surface category, thereby avoiding the problem of low curb recognition accuracy caused by directly identifying the curb line feature, and obtaining the target volume element by segmenting the cylindrical coordinate system, so that the three-dimensional points contained in the target volume element can be relatively uniform, thereby improving the accuracy of classifying the target volume element, thereby further improving the accuracy of curb recognition based on the volume element category.

[0090] Reference Figure 4 , is a step flow chart of another curb detection method provided by an embodiment of the present invention.

[0091] Step 201, mapping the three-dimensional point cloud space into a cylindrical coordinate system.

[0092] In the embodiment of the present application, the point cloud data initially collected by the laser radar and other equipment is only the point cloud data composed of the three-dimensional point data. After obtaining the point cloud data, the three-dimensional point cloud information can be unified into the Cartesian coordinate system based on the external parameters of the laser radar and other shooting equipment to obtain the three-dimensional point cloud space. The Cartesian coordinate system is a general term for the rectangular coordinate system and the oblique coordinate system.

[0093] After obtaining the three-dimensional point cloud space, the three-dimensional point cloud in the three-dimensional point cloud space can be mapped as a whole to the cylindrical coordinate system. The cylindrical coordinate system is a three-dimensional coordinate system, which is an extension of the two-dimensional polar coordinate system to the z-axis. The added third coordinate is specifically used to indicate the height of point P from the xy-plane. According to the convention established by the International Organization for Standardization, the radial distance, azimuth, and altitude are marked as ρ, θ, and z, respectively. In other words, in the cylindrical coordinate system, (ρ, θ, z) can be used to determine the coordinates of a point, which is similar to the use of (x, y, z) to determine the coordinates of a point in the three-dimensional rectangular coordinate system.

[0094] Step 202, segmenting the cylindrical coordinate system to obtain a plurality of volume elements; wherein the plurality of target volume elements have the same radial length, angular radian and height in the cylindrical coordinate system.

[0095] Due to the characteristics of the cylindrical coordinate system, the three dimensions of the radial distance ρ, azimuth angle θ and height z of the cylindrical coordinate system can be evenly divided to obtain multiple volume elements with the same radial length, angular radian and height.

[0096] like Figure 5 As shown, Figure 5 A top view of a volume element provided in an embodiment of the present application is shown, wherein the radial length of volume element 51 is L, and the angular radian is α, and the radial length of volume element 52 is L, and the angular radian is α. Since the volume element is obtained by equally dividing the radial distance, azimuth and altitude of the cylindrical coordinate system, volume elements 51 and 52 have the same radial length, angular radian and altitude. However, due to the different distances between volume elements 51 and 52 from the center of the cylindrical coordinate system, there is a difference in volume between the two, and the volume of volume element 52 is greater than that of volume element 51.

[0097] Step 203, based on the position of the volume element in the cylindrical coordinate system, determine multiple target volume elements from the multiple volume elements to obtain a target volume element set consisting of the multiple target volume elements; wherein the number of the multiple target volume elements is less than the number of the multiple volume elements.

[0098] After multiple volume elements are obtained based on the cylindrical coordinate system, target recognition can be performed based on the multiple volume elements as units, and the object category corresponding to each volume element can be determined, thereby completing the recognition operation of the entire three-dimensional point cloud. For example, obstacles, road surfaces, traffic lights, pedestrians, etc. can be identified from the three-dimensional point cloud.

[0099] In this embodiment of the present application, since the curb needs to be identified and the curb is usually located near the road surface, that is, at a lower position of the three-dimensional point cloud, the multiple volume elements obtained in the above steps can also be screened to obtain the target volume elements near the ground to reduce the number of volume elements that need to be processed.

[0100] Sub-step 2031, determining the number of three-dimensional points contained in each volume element in each stacked volume element set; wherein the stacked volume element set is formed by stacking a plurality of volume elements having the same radial distance and angular radian.

[0101] In the embodiment of the present application, in order to reduce the number of volume elements to be processed, the volume elements may be filtered. Specifically, all volume elements may be divided into a plurality of stacked volume element sets, each stacked volume element set being formed by stacking a plurality of volume elements having the same radial distance and angular radian, that is, each stacked volume element set represents a stack of volume elements in a cylindrical coordinate system.

[0102] Sub-step 2032, determining the volume element in the stacked volume element set whose number of three-dimensional points is greater than or equal to a preset number and whose bottom layer is a target volume element, and obtaining a target volume element set consisting of target volume elements corresponding to multiple stacked volume element sets respectively.

[0103] Since the bottom-most volume element in each stacked volume element set usually represents a ground position, for each stacked volume element set, the bottom-most volume element can be determined as the target volume element.

[0104] In addition, since there may be obstacles on the ground, there may not be valid three-dimensional points in all volume elements close to the ground. For example, if a vehicle is parked on the side of the road, the ground under the vehicle cannot reflect the laser beam due to the obstruction of the vehicle, and the volume element where the ground under the vehicle is located does not contain the corresponding three-dimensional point.

[0105] Therefore, in an embodiment of the present application, in order to further improve the efficiency and speed of processing the target volume elements, the volume elements are also screened according to the number of three-dimensional points contained in the volume elements, and the volume elements in each set of stacked volume elements whose number of three-dimensional points is greater than or equal to a preset number and which are located at the bottom layer of the stacked volume elements are determined as target volume elements.

[0106] Step 204: Determine the volume element category corresponding to each target volume element in the target volume element set.

[0107] This step can be referred to as step 102, and will not be described in detail in this embodiment of the present application.

[0108] Step 205 : determining a road surface area and a non-road surface area from the three-dimensional point cloud space based on the volume element categories corresponding to the target volume elements.

[0109] In the embodiment of the present application, the volume element categories may include at least a road surface category and a non-road surface category. Therefore, there are at least two categories of target volume elements, one is a road surface category target volume element, and the other is a non-road surface category target volume element.

[0110] The 3D points in all target volume elements of the road category can be fused to determine the road surface area from the 3D point cloud space; the 3D points in all target volume elements of the non-road category can be fused to determine the non-road surface area from the 3D point cloud space.

[0111] It should be noted that in an embodiment of the present application, only target volume elements of the pavement category can be identified from the target volume element set, and other target volume elements in the target volume element set except the target volume elements of the pavement category can be determined as target volume elements of the non-pavement category, so as to further improve the efficiency and speed of classifying the target volume element set.

[0112] Step 206: Determine the three-dimensional points in the road surface area that are adjacent to the non-road surface area as roadside points.

[0113] After the road surface area and the non-road surface area are determined, since the curb on the real road is usually used to distinguish different areas, for example, the curb can be used to distinguish the road from the green belt, and can also be used to distinguish the road from the sidewalk, etc. Therefore, the curb point can be determined from the junction of the road surface area and the non-road surface area.

[0114] In one embodiment, the 3D points in the road area adjacent to the non-road area can be determined as roadside points. In another embodiment, the 3D points in the non-road area adjacent to the road area can also be determined as roadside points to obtain a roadside point set.

[0115] Optionally, step 206 may include:

[0116] Sub-step 2061 , searching the road surface area in both sides of the vehicle respectively, and determining the road surface points at the edge positions of the road surface area as curb seed points.

[0117] In the embodiment of the present application, in order to improve the efficiency and speed of determining the roadside points and avoid processing all three-dimensional points in the road surface area and the non-road surface area, a cluster growth method can also be used to obtain a roadside point set composed of multiple roadside points.

[0118] Specifically, the road surface area is searched in both sides of the vehicle from which the three-dimensional point cloud is obtained, and a road surface point at the edge of the road surface area is determined as a roadside seed point.

[0119] Furthermore, since there may be obstacles at the edge of the road area, the edge position of the road area is not necessarily the actual roadside position. Therefore, whether the edge position of the road area is adjacent to the non-road area can also be used as a screening condition for determining the roadside seed point. Specifically, the road area can be searched in both sides of the vehicle that obtains the three-dimensional point cloud, and a road point at the edge position of the road area that is adjacent to the non-road area is determined as a roadside seed point.

[0120] It should be noted that on a normal road, there are two curbs located on the left and right sides of the road. Therefore, in an embodiment of the present application, the left and right sides of the vehicle are searched separately to obtain two curb seed points on the left and right sides of the vehicle, and then each curb seed point is clustered and grown separately to obtain two curb point sets representing the two curbs on both sides of the road.

[0121] Sub-step 2062, clustering and growing the curb seed points to obtain a curb point set including the curb seed points and curb growth points.

[0122] In an embodiment of the present application, after obtaining a curb seed point, the curb seed point can be added to a curb point set, and a cluster growth algorithm can be used to cluster and grow the curb seed point, find a curb point near the curb seed point, and add the found curb point to the curb point set, then find the next curb point based on the found curb point, and add the next curb point to the curb point set, and so on, until no new curb point can be matched. Among them, the cluster growth algorithm can include three types of clustering methods, namely DBSCAN clustering, regional growth clustering, and K-means clustering, and technicians can flexibly select the desired cluster growth algorithm according to actual needs, and the embodiment of the present application does not specifically limit this.

[0123] Optionally, sub-step 2062 may include:

[0124] Sub-step A1, determining a road centerline according to the road area.

[0125] In the embodiment of the present application, in order to improve the efficiency of cluster growth, the direction of cluster growth can also be determined to avoid the loss of computing power caused by searching for roadside points in multiple directions during the growth process. At the same time, the accuracy of the roadside points determined by cluster growth can also be improved.

[0126] Since the curb is usually distributed on both sides of the road along the extension direction of the road, the direction of the curb is the same as the direction of the road. Therefore, in the embodiment of the present application, the growth direction of the clustering process can be determined based on the direction of the road. Specifically, after the road surface area is obtained, the road surface centerline can be determined based on the road surface area.

[0127] Sub-step A2, taking the curb seed point as the starting point, searches for adjacent road surface points in the extension direction of the road centerline, and determines the road surface points adjacent to the non-road area and whose distance from the non-road area is less than a preset distance as curb growth points, thereby obtaining a curb point set including the curb seed point and the curb growth point.

[0128] When the curb seed points are selected from the road surface points, cluster growth is performed based on the curb seed points in the extension direction of the road surface centerline (ie, in the direction parallel to the road surface centerline).

[0129] For example, at the beginning of cluster growth, in the extension direction of the road centerline of the road seed point, the road surface points are searched from near to far, and the first road surface point that is adjacent to the non-road surface area and whose distance to the non-road surface area is less than or equal to the preset distance is determined as a road surface growth point and added to the set. After the first road surface growth point is added to the set, in the extension direction of the road centerline of the road surface growth point that has been added to the set, the road surface points are searched from near to far, and the first road surface point that is adjacent to the non-road surface area and whose distance to the non-road surface area is less than or equal to the preset distance is determined as the next road surface growth point and added to the set.

[0130] It should be noted that the extension direction of the above-mentioned road centerline may include one direction, such as the direction in which the road centerline extends forward, or may include two directions at the same time, such as the direction in which the road centerline extends forward and the direction in which the road centerline extends backward. At the same time, since in the embodiment of the present application, the curb seed points may include the curb seed points corresponding to the left curb of the road and the curb seed points corresponding to the right curb of the road, the curb seed points on both sides of the road may be grown using the above-mentioned growth process to obtain two curb point sets corresponding to the curbs on both sides of the road.

[0131] Step 207, performing denoising processing on the roadside points in the roadside point set, removing the noise points in the roadside point set, and obtaining a target roadside point set.

[0132] Since there may be noise points in the roadside point set that deviate greatly from the actual roadside location, after obtaining the roadside point set, the roadside point set can also be denoised to remove the noise points in the roadside point set that deviate greatly from the actual roadside, so as to obtain a target roadside point set with higher accuracy.

[0133] Specifically, a filtering method such as a passthrough filter, a Kalman filter or a matched filter can be used to filter the roadside point set, remove outliers in the roadside point set, and obtain a target roadside point set that more accurately reflects the roadside. It should be noted that other methods can also be used to denoise the roadside point set, and technicians can use appropriate denoising methods according to actual needs, and the embodiments of the present application do not specifically limit this.

[0134] Step 208, fitting the roadside points in the target roadside point set to obtain a roadside line.

[0135] After obtaining the target roadside point set, the roadside points in the target roadshow point set can be fitted to obtain the roadside line. It should be noted that since there can be two roadside point sets (corresponding to both sides of the road), there can also be two target roadshow point sets, and then two roadside lines corresponding to both sides of the road can be determined.

[0136] Furthermore, after obtaining two curb lines, the left curb line and the right curb line can be determined based on the positional relationship between the curb lines and the center line of the road. For example, the curb line located on the right side of the center line of the road can be determined as the right curb line, and the curb line on the left side of the center line of the road can be determined as the left curb line.

[0137] In summary, an embodiment of the present invention provides a method for detecting a curb, including: obtaining a three-dimensional point cloud space, segmenting the three-dimensional point cloud space based on a cylindrical coordinate system, and obtaining a target volume element set consisting of multiple target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; determining the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category; determining a curb point set from the three-dimensional point cloud based on the volume element category corresponding to the target volume element; and fitting the curb points in the curb point set to obtain a curb line. In the embodiment of the present invention, it is possible to determine whether the target volume element is a road surface category, and determine the curb line from the target point cloud according to the characteristic of whether the target volume element is a road surface category, thereby avoiding the problem of low curb recognition accuracy caused by directly identifying the curb line feature, and obtaining the target volume element by segmenting the cylindrical coordinate system, so that the three-dimensional points contained in the target volume element can be relatively uniform, thereby improving the accuracy of classifying the target volume element, thereby further improving the accuracy of curb recognition based on the volume element category.

[0138] Based on the above embodiments, an embodiment of the present invention further provides a curb detection device.

[0139] refer to Figure 6 , Figure 6 The following is a structural block diagram of a roadside detection device provided by an embodiment of the present invention:

[0140] The volume element module 601 is used to obtain a three-dimensional point cloud space, segment the three-dimensional point cloud space based on a cylindrical coordinate system, and obtain a target volume element set consisting of a plurality of target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element;

[0141] The category module 602 is used to determine the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category.

[0142] The roadside point module 603 is used to determine a roadside point set from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements.

[0143] The roadside line module 604 is used to fit the roadside points in the roadside point set to obtain a roadside line.

[0144] Optionally, the volume element module includes:

[0145] The mapping submodule is used to map the three-dimensional point cloud space into a cylindrical coordinate system.

[0146] The segmentation submodule is used to segment the cylindrical coordinate system to obtain a plurality of volume elements; wherein the radial length, angular radian and height of the plurality of target volume elements in the cylindrical coordinate system are the same.

[0147] The volume element submodule is used to determine multiple target volume elements from the multiple volume elements based on the positions of the volume elements in the cylindrical coordinate system, and obtain a target volume element set composed of the multiple target volume elements; wherein the number of the multiple target volume elements is less than the number of the multiple volume elements.

[0148] Optionally, the volume element submodule includes:

[0149] A quantity submodule, used to determine the number of three-dimensional points contained in each volume element in each stacked volume element set; wherein the stacked volume element set is formed by stacking a plurality of volume elements having the same radial distance and angular radian

[0150] The volume element set submodule is used to determine the volume element with the number of three-dimensional points greater than or equal to the preset number and the bottom layer in the stacked volume element set as the target volume element, and obtain a target volume element set composed of target volume elements corresponding to multiple stacked volume element sets respectively.

[0151] Optionally, the roadside point module includes:

[0152] The area submodule is used to determine the road surface area and the non-road surface area from the three-dimensional point cloud space based on the volume element categories corresponding to the target volume elements.

[0153] The roadside point determination submodule is used to determine the three-dimensional points in the road surface area that are adjacent to the non-road surface area as roadside points.

[0154] Optionally, the roadside point determination submodule includes:

[0155] The search submodule is used to search the road surface area in both sides of the vehicle and determine the road surface points at the edge of the road surface area as curb seed points.

[0156] The growth submodule is used to perform cluster growth on the curb seed points to obtain a curb point set including the curb seed points and curb growth points.

[0157] Optionally, the growth submodule includes:

[0158] The road surface centerline submodule is used to determine the road surface centerline according to the road surface area.

[0159] The curb point growth submodule is used to search for adjacent road surface points in the extension direction of the road surface centerline with the curb seed point as the starting point, and determine the road surface points adjacent to the non-road surface area and whose distance from the non-road surface area is less than a preset distance as curb growth points, thereby obtaining a set of curb points including the curb seed point and the curb growth point.

[0160] Optionally, the roadside module includes:

[0161] The denoising submodule is used to perform denoising on the roadside points in the roadside point set, remove the noise points in the roadside point set, and obtain a target roadside point set.

[0162] The fitting submodule is used to fit the roadside points in the target roadside point set to obtain a roadside line.

[0163] In summary, an embodiment of the present invention provides a curb detection device, including: a volume element module, used to acquire a three-dimensional point cloud space, segment the three-dimensional point cloud space based on a cylindrical coordinate system, and obtain a target volume element set composed of multiple target volume elements; wherein the radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; a category module, used to determine the volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category. A curb point module, used to determine a curb point set from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements. A curb line module, used to fit the curb points in the curb point set to obtain a curb line. In an embodiment of the present invention, it is possible to determine whether a target volume element is of a road surface category, and based on the characteristics of whether the target volume element is of a road surface category, determine the roadside line from the target point cloud, thereby avoiding the problem of low roadside recognition accuracy caused by directly identifying roadside line features. Furthermore, the target volume element is obtained by segmenting the cylindrical coordinate system, so that the three-dimensional points contained in the target volume element can be relatively uniform, thereby improving the accuracy of classifying the target volume element, thereby further improving the accuracy of roadside recognition based on volume element category.

[0164] An embodiment of the present invention further provides a vehicle controller, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the above-mentioned curb detection method.

[0165] The present invention provides a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by a processor of a vehicle controller, the vehicle controller can execute the above-mentioned curb detection method.

[0166] An embodiment of the present invention further provides a vehicle, comprising the above-mentioned vehicle controller.

[0167] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned device embodiments and will not be repeated here.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0169] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A roadside detection method, characterized in that: The method comprises: Acquire a three-dimensional point cloud space, and segment the three-dimensional point cloud space based on a cylindrical coordinate system to obtain a target volume element set consisting of a plurality of target volume elements; wherein a radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; Determining a volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category; Determining a set of curb points from the three-dimensional point cloud based on volume element categories corresponding to the target volume elements; The roadside points in the roadside point set are fitted to obtain a roadside line.

2. The method according to claim 1, characterized in that The three-dimensional point cloud space is segmented based on a cylindrical coordinate system to obtain a target volume element set consisting of a plurality of target volume elements, including: Mapping the three-dimensional point cloud space into a cylindrical coordinate system; Segmenting the cylindrical coordinate system to obtain a plurality of volume elements; wherein the plurality of target volume elements have the same radial length, angular arc, and height in the cylindrical coordinate system; Based on the position of the volume element in the cylindrical coordinate system, multiple target volume elements are determined from the multiple volume elements to obtain a target volume element set consisting of the multiple target volume elements; wherein the number of the multiple target volume elements is less than the number of the multiple volume elements.

3. The method according to claim 2, characterized in that The step of determining a plurality of target volume elements from the plurality of volume elements based on positions of the volume elements in the cylindrical coordinate system, and obtaining a target volume element set consisting of the plurality of target volume elements, comprises: Determining the number of three-dimensional points contained in each volume element in each stacked volume element set; wherein the stacked volume element set is formed by stacking a plurality of volume elements having the same radial distance and angular radian; The volume element at the bottom layer whose number of three-dimensional points in the stacked volume element set is greater than or equal to a preset number is determined as a target volume element, and a target volume element set consisting of target volume elements corresponding to multiple stacked volume element sets is obtained.

4. The method according to claim 1, wherein The determining of a set of curb points from the three-dimensional point cloud based on the volume element categories corresponding to the target volume elements includes: determining a road surface area and a non-road surface area from the three-dimensional point cloud space based on volume element categories corresponding to the target volume elements; A three-dimensional point in the road surface area that is adjacent to the non-road surface area is determined as a roadside point.

5. The method according to claim 4, characterized in that The step of determining a three-dimensional point in the road surface area adjacent to the non-road surface area as a roadside point comprises: Searching the road surface area in both sides of the vehicle, and determining road surface points at edge positions of the road surface area as curb seed points; Cluster growth is performed on the roadside seed points to obtain a roadside point set including the roadside seed points and roadside growth points.

6. The method according to claim 5, characterized in that The clustering and growing of the roadside seed points to obtain a roadside point set including the roadside seed points and roadside growth points includes: determining a road centerline based on the road area; Taking the curb seed point as the starting point, search for adjacent road surface points in the extension direction of the road surface centerline, and determine the road surface points adjacent to the non-road surface area and whose distance from the non-road surface area is less than a preset distance as curb growth points, to obtain a curb point set including the curb seed point and the curb growth point.

7. The method according to claim 1, characterized in that The step of fitting the roadside points in the roadside point set to obtain a roadside line includes: Performing denoising on the roadside points in the roadside point set to remove noise points in the roadside point set to obtain a target roadside point set; The roadside points in the target roadside point set are fitted to obtain a roadside line.

8. A roadside detection device, characterized in that: The device comprises: a volume element module, configured to acquire a three-dimensional point cloud space, segment the three-dimensional point cloud space based on a cylindrical coordinate system, and obtain a target volume element set consisting of a plurality of target volume elements; wherein a radial distance of the target volume element in the three-dimensional point cloud space is proportional to the volume of the target volume element; a category module, configured to determine a volume element category corresponding to each target volume element in the target volume element set; wherein the volume element category includes a road surface category and a non-road surface category; A roadside point module, configured to determine a roadside point set from the three-dimensional point cloud based on volume element categories corresponding to the target volume elements; The roadside line module is used to fit the roadside points in the roadside point set to obtain a roadside line.

9. A vehicle controller, characterized in that: The vehicle controller includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the roadside detection method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that When the instructions in the computer-readable storage medium are executed by a processor of a vehicle controller, the vehicle controller is enabled to execute the roadside detection method according to any one of claims 1 to 7.

Images