A curb detection method and system based on dual laser radars

Abstract

The application discloses a curb detection method and system based on double laser radars, comprising the following steps: acquiring point cloud data of the double laser radars within respective detection angles, calibrating and fusing the point cloud data to obtain fused point cloud; performing grid division on the fused point cloud, sampling Z-axis minimum point cloud in each grid, projecting the Z-axis minimum point cloud onto the included angle middle line of the corresponding grid to obtain a point cloud group; calculating the curvature angle of each point cloud in the point cloud group; the curvature angle is the included angle between the current point cloud and the vectors formed by the current point cloud and two point clouds on the same scanning line; judging whether the curvature angle of the point cloud satisfies the curvature condition of a curb inflection point or not, and detecting the bidirectional straight line feature of the point cloud satisfying the curvature condition of the curb inflection point. The layout scheme of the double laser radars reduces the near field blind area and compensates for the visual angle blocked by the vehicle body, calculates the curvature feature and the straight line feature of the fused point cloud after sampling, guarantees the regularity of the point cloud, and prevents false detection in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of curb detection technology, and in particular to a curb detection method and system based on dual lidar. Background Technology

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

[0003] The lidar installed on existing autonomous vehicles is mostly single, but there are blind spots in the field of vision at close range and the view is blocked by the vehicle body, resulting in low point cloud density in the key detection area directly in front of the vehicle.

[0004] Under structured roads, the road surface is smooth and continuous with a gentle change in height, while the curb is a vertical plane with a steep change in height. The curb is generally higher than the road surface. Therefore, the spatial distribution characteristics of point clouds on the road surface and the curb are different. The distribution of point clouds on the road surface and the curb is as follows: the point cloud rings near the road surface have small radii and large curvatures, and the point clouds are dense; the point clouds far away have large radii and small curvatures, and the point clouds are sparse. In other words, the spacing of radar laser scanning lines increases with the distance. If the same distance parameter is used for differentiation, it will lead to the inability to distinguish nearby objects well or the over-segmentation of distant objects.

[0005] Furthermore, point clouds appear as concentric circles spreading outwards on the road surface and as straight lines on the curb. There are abrupt changes in curvature between the two, and there are two curvature abrupt changes at the curb points on each scan line. When the curvature angle of a point cloud is detected to meet the curvature condition of a curb inflection point, it cannot be determined which side the inflection point is on. If detection is only performed in a single direction, if the first curb inflection point is not detected and only the second inflection point is detected, and the subsequent point cloud detection of the second inflection point does not conform to the straight line feature, then a missed detection will occur. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a roadside detection method and system based on dual lidar. The dual lidar layout reduces near-field blind spots while compensating for viewing angles obstructed by vehicle bodies. After sampling the fused point cloud, curvature and straight-line features are calculated to ensure the regularity of the point cloud and prevent false detections in complex environments.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a curb detection method based on dual lidar, comprising:

[0009] The point cloud data of the two lidars within their respective detection angles are acquired, and the point cloud data of the two lidars are calibrated and fused to obtain the fused point cloud;

[0010] The fused point cloud is divided into grids, the minimum point cloud along the Z-axis in each grid is sampled, and the minimum point cloud along the Z-axis is projected onto the midline of the included angle of the corresponding grid to obtain a point cloud group.

[0011] Calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line;

[0012] Determine whether the curvature angle of the point cloud satisfies the curvature condition of the road edge inflection point. For point clouds that satisfy the curvature condition of the road edge inflection point, perform bidirectional straight line feature detection to obtain the road edge detection result.

[0013] As an alternative implementation, the detection angles of the two lidars are 105° to 270° and 90° to 255°, respectively, and the termination threshold of one lidar must intersect at least with the point cloud farthest from the other lidar on the starting threshold of the other lidar.

[0014] As an alternative implementation method, the calibration fusion process includes: calibrating the overlapping parts of the point cloud data of the two lidars, converting the point cloud data of one lidar to the coordinate system of the other lidar, and then stitching the point clouds together to obtain the fused point cloud.

[0015] As an alternative implementation, the curvature angle is calculated by segmenting the n-values, specifically by determining the gap l between scan lines with negative vertical angles. i Then, select segments based on the smoothness of the gap increment;

[0016] l i =H×tan(90°-α) i+1 )-tan(90°-α i )

[0017] Where, α i H represents the angle between the i-th scan line and the installation horizontal line when the vertical angles of the lidar are arranged from smallest to largest, and H represents the installation height.

[0018] As an alternative implementation method, the detection conditions for bidirectional straight line features are: the curb point cloud that satisfies the curb inflection point curvature condition is an inclined straight line on the XOZ plane, or the curb point cloud is a straight line in the XOY plane.

[0019] As an alternative implementation, when the curb point cloud is a sloping straight line on the XOZ plane, it satisfies the following:

[0020]

[0021] |z i+2 -z i |≥|zi+2 -z i+1 |

[0022] Where, x i x i+1 z represents the X-axis coordinates of the i-th and i+1-th road edge point clouds; i z i+1 z i+2 Let δ be the Z-axis coordinates of the i-th, i+1-th, and i+2-th road edge point clouds; xz The threshold value for the slope of the curb point cloud in the XOZ plane.

[0023] As an alternative implementation, when the curb point cloud is a straight line in the XOY plane, it satisfies the following:

[0024]

[0025] |y i+1 -y i |<δ y

[0026] Where, x i x i+1 Let y be the X-axis coordinates of the i-th and i+1-th road edge point clouds; i y i+1 Let δ be the Y-axis coordinates of the i-th and i+1-th road edge point clouds; xy δ is the slope threshold of the roadside point cloud in the XOY plane; y The distance between adjacent roadside point clouds along the Y-axis.

[0027] Secondly, the present invention provides a curb detection system based on dual lidar, comprising:

[0028] The point cloud fusion module is configured to acquire point cloud data from the two lidars within their respective detection angles, and then calibrate and fuse the point cloud data from the two lidars to obtain a fused point cloud.

[0029] The point cloud projection module is configured to divide the fused point cloud into a grid, sample the minimum point cloud along the Z-axis in each grid, and project the minimum point cloud along the Z-axis onto the midline of the included angle of the corresponding grid to obtain a point cloud group.

[0030] The curvature feature judgment module is configured to calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line;

[0031] The straight line feature detection module is configured to determine whether the curvature angle of the point cloud meets the curvature condition of the road edge inflection point, and to perform bidirectional straight line feature detection on the point cloud that meets the curvature condition of the road edge inflection point, thereby obtaining the road edge detection result.

[0032] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0033] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] To address the problem of excessive blind spots caused by mounting a single 3D LiDAR too high on an autonomous vehicle, this invention proposes a complementary installation layout for dual LiDARs on an autonomous vehicle. This reduces near-field blind spots while compensating for the viewpoint obstructed by the vehicle body, increases the point cloud density in the key detection area directly in front of the vehicle, and calibrates the point cloud of the overlapping part of the dual LiDARs, improving calibration speed.

[0036] For fused point clouds, this invention proposes a multi-feature fusion algorithm based on sampling and segmentation. The algorithm samples the lowest value of the fused point cloud at different angles. In order to ensure the regularity of the point cloud, the sampled lowest value is projected onto the center of the region. The curvature angle and straight line features of the sampled point cloud are calculated. Bidirectional straight line feature detection is performed on the point cloud that meets the curvature to prevent false detection in complex environments.

[0037] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 This is a flowchart of the road edge detection method based on dual lidar provided in Embodiment 1 of the present invention;

[0040] Figure 2 This is a schematic diagram of the overlapping area of ​​the two lidars provided in Embodiment 1 of the present invention;

[0041] Figure 3 This is the minimum point cloud projection image of the Z-axis on the same scan line in the XOY plane provided in Embodiment 1 of the present invention;

[0042] Figure 4 This is a schematic diagram of the point cloud distribution along the roadside provided in Embodiment 1 of the present invention;

[0043] Figure 5This is a diagram showing the gap between adjacent wire harnesses provided in Embodiment 1 of the present invention;

[0044] Figure 6 This is a schematic diagram of bidirectional straight line feature detection provided in Embodiment 1 of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0048] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0049] Example 1

[0050] This embodiment provides a road edge detection method based on dual lidar, such as... Figure 1 As shown, it includes:

[0051] The point cloud data of the two lidars within their respective detection angles are acquired, and the point cloud data of the two lidars are calibrated and fused to obtain the fused point cloud;

[0052] The fused point cloud is divided into grids, the minimum point cloud along the Z-axis in each grid is sampled, and the minimum point cloud along the Z-axis is projected onto the midline of the included angle of the corresponding grid to obtain a point cloud group.

[0053] Calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line;

[0054] Determine whether the curvature angle of the point cloud satisfies the curvature condition of the road edge inflection point. For point clouds that satisfy the curvature condition of the road edge inflection point, perform bidirectional straight line feature detection to obtain the road edge detection result.

[0055] In this embodiment, to address the problem of excessive blind spots caused by installing a single 3D LiDAR too high on an unmanned vehicle, a dual LiDAR installation layout scheme with complementary viewing angles on the unmanned vehicle is proposed. This reduces near-field blind spots while compensating for the viewing angles blocked by the vehicle body, and increases the point cloud density of key detection areas directly in front of the vehicle.

[0056] Then, for the point cloud data of the two lidars within their respective detection angles, the normal distribution transformation (NDT) algorithm is used to calibrate the extrinsic parameters of the two lidars. This algorithm does not depend on the initial transformation matrix between the two lidars, does not require a large number of iterative calculations, and has a fast calibration speed and good effect.

[0057] like Figure 2 As shown, in actual use, the point clouds of the dual lidars only overlap within a certain range in front of the vehicle. Therefore, only the point cloud of the overlapping part needs to be calibrated. The large area of ​​non-overlapping point clouds has no registration points, and calibrating this part will affect the calibration speed. Therefore, the point cloud of the dual lidars needs to be filtered out by region before calibration.

[0058] Specifically:

[0059] The point cloud's angular distribution in the XOY plane is such that, with the LiDAR's Y-axis as 0 degrees, it increases to 360° counterclockwise around the origin of the coordinate system. Due to the vehicle model and installation method, both LiDARs have a 105° blind zone. Therefore, LiDAR 1 needs to detect the starting angle of 105°, where O1B is located, and LiDAR 2 needs to detect the ending angle of 255°, where O2A is located. O2O is the termination threshold for LiDAR 2's detection; all point clouds on it can be detected by LiDAR 1. This ensures the complete overlap area between the two LiDARs. Therefore, the termination threshold line O1A of LiDAR 1 must intersect at least with the point cloud on O2B that is farthest from O2.

[0060] In the triangle formed by O2O1A, ∠O1O2A = 15°, and the length of O1A is much greater than that of O1O2. According to the properties of a triangle, ∠O1AO2 is much less than 15°. To simplify the calculation, ∠O1AO2 is ignored in this embodiment. Therefore, ∠AO1O2 = 165°, meaning the detection angle of lidar #1 is between 105° and 270°. Similarly, the detection angle of lidar #2 is between 90° and 255°. Each lidar must be segmented before calibration, and areas outside the detection angle must be filtered out, acquiring only the point cloud data within the respective detection angles of the two lidars.

[0061] The basic principle of the NDT algorithm is to divide the target point cloud data into multiple three-dimensional grids, calculate the mean x′ and covariance matrix σ of the point cloud in each grid to represent the distribution of the point cloud in the grid, and calculate the probability density P of each point cloud that needs to be matched in the grid.i for:

[0062]

[0063] The probability density represents the negative value of the distance between the registered point cloud and the target point cloud. Therefore, the smaller the distance between the point clouds within the grid, the better the registration effect.

[0064] The cost function P is obtained by summing the negatives of the probability densities of each grid cell:

[0065]

[0066] Where i is the grid number, x′ i σ i Let P be the mean and covariance matrix of the i-th grid. P is used to quantify the point cloud calibration effect; the smaller P is, the better the point cloud calibration effect. The NDT algorithm iteratively searches for the optimal homogeneous coordinate transformation matrix T that minimizes P. T contains the rotation matrix R and the translation matrix t.

[0067] Where R represents the angular difference between the axes of the two coordinate systems, with α representing the rotation around the Z-axis, β representing the rotation around the Y-axis, and γ representing the rotation around the X-axis. The rotation order of the coordinate axes is ZYX, so R can be obtained as:

[0068]

[0069] t represents the deviation between the positions of the two coordinate systems, defined as: t = [x t y t z t ] T ;

[0070] During calibration, with LiDAR 1 as the target, the point cloud data of LiDAR 2 is transformed into the coordinate system of LiDAR 1. This is done by left-multiplying each point cloud of LiDAR 2 by R and adding t to obtain the Cartesian coordinates of the point cloud in the coordinate system of LiDAR 2 in the coordinate system of LiDAR 1. Then, the transformed point cloud data and the point cloud data in the coordinate system of LiDAR 1 are stitched together to obtain the fused point cloud.

[0071] In this embodiment, a parallel dual-LiDAR road edge detection algorithm based on sampling and segmented multi-feature fusion is proposed. First, the fused point cloud is divided into multiple grids, and the minimum Z-axis point cloud within each grid is sampled. To ensure equal spacing between sampled point clouds, the minimum Z-axis point cloud is projected onto the midline of the included angle of the corresponding grid. Second, spatial features, including straight-line and curvature features, are calculated segmentally from the sampled point cloud according to the point cloud line bundle distribution pattern. Different thresholds are used for calculating spatial features based on line bundles with different sparsities. To prevent missed road edge detection, bidirectional straight-line detection is implemented for point clouds that satisfy curvature features. The algorithm is parallelized to improve detection speed.

[0072] Specifically:

[0073] Because the fused point cloud data is large and processing is time-consuming, it needs to be sampled. In this embodiment, sampling is performed according to line bundles and horizontal angles. Each line bundle is divided into grids every 1 degree. The horizontal resolution of the LiDAR is 0.2°, so there are 5 point clouds in each grid. The point cloud with the smallest Z-axis among the 5 point clouds in each grid can reflect the lowest height in this area. The road surface and the curb are the lowest points in their respective areas, so only the sampled point cloud with the smallest Z-axis needs to be processed.

[0074] However, because the location of the lowest point is different within each grid, uneven distribution of the point cloud after sampling can easily occur. For example... Figure 3 As shown, O is the origin of the lidar, and the solid lines radiating outward represent the grid dividing the region. The angle between adjacent solid lines is 1°, and the dashed lines are the midlines of the angles within the grid. Point clouds p2, p3, and p6 represent the lowest point clouds within the corresponding grid regions. It can be seen that the lowest point within adjacent grids may be the two closest point clouds, such as p2 and p3, or the farthest point clouds, such as p3 and p6. The sampled point clouds exhibit irregularity, which is not conducive to subsequent use of the spatial characteristics of the point clouds for curb point cloud detection. Therefore, in this embodiment, the minimum Z-axis point cloud sampled in each grid is projected onto the midline of the angle between the corresponding grids to obtain p1, p4, and p5. The angles of the projected point clouds are equal and the distribution is uniform, preserving the spatial characteristics between adjacent point clouds, which is beneficial for subsequent point cloud processing. Finally, a two-dimensional point cloud group of size 32×360 is formed, where 32 is the total number of line beams of the two lidars and 360 is the total number of horizontal angles divided, reflecting the lowest plane situation of each point cloud detection area, and therefore can be used to detect curb point clouds.

[0075] In practical engineering tasks, real-time performance is an important indicator. In order to improve the processing speed of point clouds, this embodiment adopts a parallel traversal method to sample the fused point cloud. Specifically, the fused point cloud is put into 6 threads, and the fused point cloud is traversed in each thread. The point cloud is assigned to the corresponding grid, and the minimum Z-axis point cloud in the grid is found and stored in a two-dimensional array. The extracted minimum Z-axis point cloud group contains road surface and road edge information. Thus, road boundary points can be extracted based on the spatial characteristics of the point cloud distribution.

[0076] Under structured roads, the road surface is smooth and continuous with a gentle change in height, while the curb is a vertical plane with a steep change in height. The curb is generally 10-25 cm higher than the road surface. Therefore, the spatial distribution characteristics of point clouds on the road surface and the curb are different. Point clouds on the road surface appear as circles that spread outward, while on the curb they appear as straight lines. There is a sudden change in curvature between the two.

[0077] Point cloud distribution models on road surfaces and curbs, such as Figure 4 As shown, the point cloud rings in the near distance have small radii and large curvatures, resulting in dense point clouds; while the point cloud rings in the far distance have large radii and small curvatures, resulting in sparse point clouds. Therefore, it is difficult to unify the parameters of the near and far ends, and it is necessary to process the point cloud rings with different radii separately in a segmented manner. Figure 4 The dashed box in the lower left corner is a top view of the point cloud on the same scan line. It can be seen that the point cloud on the same scan line has two curvature abrupt changes in the XOY plane when it starts scanning the curb and leaves the curb. The abrupt changes produce two curb inflection points. The point cloud between the two curb inflection points is a straight line. Therefore, the curb points are extracted by combining the curvature features and the straight line features.

[0078] First, the curvature angle in the XOY plane is calculated; the curvature angle of the point cloud is the angle between the current point cloud and the vectors formed by the n point clouds on both sides of the same scan line. Figure 4 Taking θ1 as an example, the point corresponding to θ1 and the point clouds on the left and right sides form the vector shown by the arrow. The angle θ1 between the two vectors is the curvature angle of that point, and θ1 is:

[0079]

[0080]

[0081]

[0082] Where, x i y i Let x be the XY coordinate of the i-th point cloud on the same scan line. i+j y i+j Let θ be the XY coordinates of the (i+j)th point cloud, v1 and v2 be the vectors formed by the current point cloud and the point clouds on both sides, respectively. iLet be the curvature angle of the i-th point cloud, and n be the number of point clouds used to calculate the vector. Different line bundles produce different numbers of point clouds on the roadside.

[0083] When the lidar origin is close to the curb, it easily forms a near-straight point cloud region on the sidewalk, as shown in the point cloud within solid box A in Figure 4. Sidewalks are generally made of bricks and tiles, with small protrusions, which generate many point clouds with small curvature angles. In this case, the straight-line and curvature features are similar to those of the curb points. If n is too small, these small angles will be mistakenly detected as curb inflection points, resulting in a large number of false detections because the point cloud spatial features are the same as those of the curb points. However, these small angles are generally formed by a very small number of point clouds that are larger or smaller than the scanning radius. This interference can be eliminated by taking several point clouds from both sides and calculating the angle together. Figure 4 As shown in the dashed box A1, when point cloud interference occurs, the smaller n is, the greater the curvature angle θ formed. i The smaller the value of n, the easier it is to misidentify it as a curb bend; the larger the value of n, the greater the angle θ. j The larger the value of n, the lower the possibility of false detection. However, for longer scan lines, the number of point clouds along the roadside is relatively small. If n is too large, exceeding the number of point clouds along the roadside, then the calculation will not only include the point clouds along the roadside, easily leading to overestimation of the included angle and loss of roadside inflection points, such as... Figure 4 The top view is shown in the box.

[0084] Based on the above reasons, the design of n in this embodiment is as follows:

[0085] It is known that the distance between two point clouds on different beams is related to the scanning radius of the beam. The product of the scanning radius and the horizontal resolution is the arc length between two point clouds on the same beam. The larger the arc length, the sparser the point cloud, and the fewer point clouds are distributed on the curb. Therefore, the sparseness variation of the point cloud can be approximated by judging the change in the scanning radius.

[0086] Figure 5 In the diagram, H represents the installation height, l0 represents the gap between the -15° laser and the -13° laser, l1 represents the gap between the -13° laser and the -11° laser, and so on, with l6 representing the gap between the -3° laser and the -1° laser. When the angle value is positive, the laser scans upwards along the horizontal plane of the installation and cannot scan the ground. Therefore, the gap is not calculated for the scanning line that has a positive angle with the horizontal plane.

[0087] Therefore, the formula relating the gap and the angle is:

[0088]

[0089] We can deduce that:

[0090] l i =H×tan(90°-α) i+1 )-tan(90°-αi (9)

[0091] Where i = 0, 1, 2, 3, 4, 5, 6 represent the 7 gaps between 8 scan lines with negative vertical angles; α i Let α0 = -15°, α1 = -13°, α2 = -11°, and so on, α6 = -1°, when the vertical angles of the lidar are arranged from smallest to largest.

[0092] Substituting the actual installation height H = 0.6 and the corresponding angle into equation (9), we get l0 = 0.36, l1 = 0.48, l2 = 0.70, l3 = 0.87, l4 = 1.90, l5 = 4.56, and l6 = 22.90, in meters. It can be observed that when i = 4 and i = 5, the wire harness gap l4 suddenly increases significantly, while the gap increases of the four wire harnesses of the first five wire harnesses are gradual. Therefore, it can be inferred that the density of the scanning point cloud of the sixth wire harness is much different from that of the first five wire harnesses. Based on this, the first five wire harnesses are one segment with the same n value; i = 5 and i = 6 are another segment with different n values.

[0093] This embodiment uses dual lidars. During point cloud fusion, the beam values ​​of lidar #2 were all increased by 16. Therefore, for scan lines with beam values ​​greater than 16, the n value and the beam value corresponding to the beam value minus 16 are the same. After multiple experiments, the relationship between the value of n and the beam marker ring is shown in Table 1:

[0094] Table 1. Correspondence between LiDAR beam and n value

[0095]

[0096] In practical engineering, to reduce the amount of calculation, we first take n=1 and quickly calculate the curvature angle between three adjacent points. If the angle is less than 150°, it is a potential roadside turning point. For this type of point cloud, we use the method of selecting n values ​​in segments to calculate the accurate curvature angle and then make an accurate judgment again.

[0097] Furthermore, because each edge point on the scan line exhibits two curvature abrupt changes, a unique characteristic of edge point clouds, when the curvature angle of a detected point cloud satisfies the edge inflection point curvature condition, it cannot be determined which side the inflection point is on. If detection is only performed in a single direction, and the first edge inflection point is not detected, but only the second inflection point is detected, and subsequent point cloud detections of the second inflection point do not conform to the straight-line feature, then a missed detection will occur. Figure 6 As shown, when traversing the point cloud counterclockwise on the same line bundle, if the lower curb inflection point is not detected and only the upper curb inflection point is detected, according to the rules, only the linear features of the subsequent point cloud of the upper curb inflection point will be calculated, which will lead to the missed detection of the real curb inflection point.

[0098] Therefore, this embodiment employs a bidirectional detection method. When a potential curb inflection point is detected, straight line features are detected in both forward and backward directions. Thus, even if only a second curb inflection point is detected, the missed curb point cloud can be identified through a second backward detection. The direction for calculating straight line features at curb inflection points is as follows: Figure 6 The arrows in the image indicate the direction.

[0099] In this embodiment, the detected curb point cloud has two straight line features: first, the curb point cloud appears as an inclined straight line in the XOZ plane, satisfying equations (10) and (11); second, the curb point cloud also appears as a straight line in the XOY plane, satisfying equations (12) and (13).

[0100]

[0101] |z i+2 -z i |≥|z i+2 -z i+1 |(11)

[0102]

[0103] |y i+1 -y i |<δ y (13)

[0104] Where, δ xy The slope threshold of the curb point in the XOY plane; δ y This is the Y-axis distance between adjacent roadside points, which is less than the Y-axis distance between non-roadside points; x i x i+1 Let y be the X-axis coordinates of the i-th and i+1-th road edge point clouds; i y i+1 Let z be the Y-axis coordinates of the i-th and i+1-th road edge point clouds; i z i+1 z i+2 Let be the Z-axis coordinates of the i-th, i+1-th, and i+2-th roadside point clouds.

[0105] Example 2

[0106] This embodiment provides a curb detection system based on dual lidar, including:

[0107] The point cloud fusion module is configured to acquire point cloud data from the two lidars within their respective detection angles, and then calibrate and fuse the point cloud data from the two lidars to obtain a fused point cloud.

[0108] The point cloud projection module is configured to divide the fused point cloud into a grid, sample the minimum point cloud along the Z-axis in each grid, and project the minimum point cloud along the Z-axis onto the midline of the included angle of the corresponding grid to obtain a point cloud group.

[0109] The curvature feature judgment module is configured to calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line;

[0110] The straight line feature detection module is configured to determine whether the curvature angle of the point cloud meets the curvature condition of the road edge inflection point, and to perform bidirectional straight line feature detection on the point cloud that meets the curvature condition of the road edge inflection point, thereby obtaining the road edge detection result.

[0111] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0112] In further embodiments, the following is also provided:

[0113] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0114] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0115] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0116] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0117] The method in Example 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0118] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0119] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A curb detection method based on dual lidar, characterized in that, include: The point cloud data of the two lidars within their respective detection angles are acquired, and the point cloud data of the two lidars are calibrated and fused to obtain the fused point cloud; The fused point cloud is divided into grids, the minimum point cloud along the Z-axis in each grid is sampled, and the minimum point cloud along the Z-axis is projected onto the midline of the included angle of the corresponding grid to obtain a point cloud group. Calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line; When calculating the curvature angle, the n value is selected piecewise. Specifically, the gap between scan lines with negative vertical angles is determined. Then, select segments based on the smoothness of the gap increment; in, When the vertical angles of the lidars are arranged from smallest to largest, the first... The angle between the scan line and the installation horizontal line. Installation height; Determine whether the curvature angle of the point cloud satisfies the curvature condition of the road edge inflection point. For point clouds that satisfy the curvature condition of the road edge inflection point, perform bidirectional straight line feature detection to obtain the road edge detection result.

2. The curb detection method based on dual lidar as described in claim 1, characterized in that, The detection angles of the two lidars are 105° to 270° and 90° to 255°, respectively, and the termination threshold of one lidar must intersect at least with the point cloud farthest from the other lidar on the starting threshold of the other lidar.

3. The curb detection method based on dual lidar as described in claim 1, characterized in that, The calibration and fusion process includes: calibrating the overlapping parts of the point cloud data of the two lidars, converting the point cloud data of one lidar to the coordinate system of the other lidar, and then stitching the point clouds together to obtain the fused point cloud.

4. The curb detection method based on dual lidar as described in claim 1, characterized in that, The detection conditions for bidirectional straight line features are: the roadside point cloud that meets the roadside inflection point curvature condition is a slanted straight line on the XOZ plane, or the roadside point cloud is a straight line in the XOY plane.

5. The curb detection method based on dual lidar as described in claim 4, characterized in that, When the curb point cloud is a sloping straight line on the XOZ plane, it satisfies: in, For the first i、i+ The X-axis coordinate of a single curb point cloud; For the first i、i+ 1 、i+ Z-axis coordinates of two curb point clouds; The threshold value for the slope of the curb point cloud in the XOZ plane.

6. The curb detection method based on dual lidar as described in claim 4, characterized in that, When the point cloud of the curb is a straight line in the XOY plane, it satisfies: in, For the first i、i+ The X-axis coordinate of a single curb point cloud; For the first i、i+ The Y-axis coordinate of a single curb point cloud; The slope threshold of the roadside point cloud in the XOY plane; The distance between adjacent roadside point clouds along the Y-axis.

7. A curb detection system based on dual lidar, characterized in that, include: The point cloud fusion module is configured to acquire point cloud data from the two lidars within their respective detection angles, and then calibrate and fuse the point cloud data from the two lidars to obtain a fused point cloud. The point cloud projection module is configured to divide the fused point cloud into a grid, sample the minimum point cloud along the Z-axis in each grid, and project the minimum point cloud along the Z-axis onto the midline of the included angle of the corresponding grid to obtain a point cloud group. The curvature feature judgment module is configured to calculate the curvature angle of each point cloud in the point cloud group; the curvature angle is the angle between the current point cloud and the vector formed by the n point clouds on both sides of the same scan line; When calculating the curvature angle, the n value is selected piecewise. Specifically, the gap between scan lines with negative vertical angles is determined. Then, select segments based on the smoothness of the gap increment; in, When the vertical angles of the lidars are arranged from smallest to largest, the first... The angle between the scan line and the installation horizontal line. Installation height; The straight line feature detection module is configured to determine whether the curvature angle of the point cloud meets the curvature condition of the road edge inflection point, and to perform bidirectional straight line feature detection on the point cloud that meets the curvature condition of the road edge inflection point, thereby obtaining the road edge detection result.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

Images