A radar point cloud coverage detection method, layout rationality detection method and control optimization method

Abstract

The present application discloses a radar point cloud coverage detection method, a layout rationality detection method and a control optimization method; the steps are as follows: construct a three-dimensional simulation model of the area to be measured and perform coordinate system conversion; construct a simulation model of the area to be measured according to the installation position of the radar, and extract a three-dimensional model of the cylindrical detection area that meets the radar perception range; project the radar point cloud onto the surface of the detection area by simulation according to the characteristics of the radar, and construct a rectangular projection area scanned by each projection effective point on the surface of the detection area; count the total effective coverage area of ​​all rectangular projection areas, and calculate the point cloud coverage rate according to the ratio of the total effective coverage area to the total area of ​​all facets in the detection area. The beneficial effects of the present application: the perception efficiency of each radar point is quantified with high precision during the design stage, providing an objective basis for subsequent point screening, and significantly improving the system perception efficiency and layout resource utilization.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a radar point cloud coverage detection method, a layout rationality detection method, and a control optimization method. Background Art

[0002] As a vital component of transportation hubs, vehicle management at highway service areas plays a crucial role in ensuring road safety and smooth traffic flow. Service areas are densely populated with vehicles, especially at key locations like highways, making management challenging. Traditional manual management cannot respond quickly enough during peak hours, leading to traffic jams, improperly parked vehicles, and even oversight of vehicles due to fatigue or negligence, increasing the risk of accidents. The limitations of manual operations make emergency response inefficient, making it difficult to predict and respond promptly.

[0003] To address vehicle classification and tracking in highway service areas, fixed-point LiDAR (lidar) is often used due to its high precision. Compared to traditional cameras, LiDAR can provide three-dimensional information about the service area and vehicles, and can directly track vehicles across the entire area without the need for other technical means. However, due to the numerous obstructions within the service area and the varying specifications of individual LiDARs, some blind spots can occur in the LiDAR scan. These blind spots can affect LiDAR coverage, resulting in areas not being fully covered. These incomplete areas ultimately prevent the LiDAR from identifying vehicles and pedestrians in the corresponding locations, ultimately affecting the overall efficiency of the LiDAR system. Therefore, LiDAR deployment requires optimization. Summary of the Invention

[0004] One of the objectives of the present application is to provide a radar point cloud coverage detection method that can solve at least one of the defects in the above-mentioned background technology.

[0005] Another object of the present application is to provide a radar arrangement rationality detection method that can solve at least one of the defects in the above-mentioned background technology.

[0006] Another object of the present application is to provide a radar control optimization method that can solve at least one of the defects in the above-mentioned background technology.

[0007] To achieve at least one of the above objectives, the present application adopts a technical solution: a radar point cloud coverage detection method, comprising the following steps:

[0008] S100: constructing a three-dimensional simulation model of the area to be measured and performing coordinate system conversion;

[0009] S200: constructing a simulation model of the area to be measured according to the installation position of the radar, and extracting a three-dimensional model of the cylindrical detection area that meets the radar sensing range;

[0010] S300: Projecting the radar point cloud onto the surface of the detection area through simulation according to the characteristics of the radar, and constructing a rectangular projection area scanned by each effective projection point on the surface of the detection area;

[0011] S400: Counting the total effective coverage area of ​​the rectangular projection area corresponding to all the projection valid points, and calculating the point cloud coverage according to the ratio of the total effective coverage area to the total area of ​​all the patches in the detection area.

[0012] Preferably, the construction of the rectangular projection area corresponding to each effective projection point in step S300 includes the following process: in the horizontal direction, the distance L from the effective projection point to the radar is i Multiply by the horizontal angular resolution, calculate the arc length between adjacent points on the same circumference of the projection effective point as the horizontal side length of the rectangular projection area; in the vertical direction, if the current projection effective point is collected by the nth vertical beam, then the two adjacent beams above and below the current projection effective point correspond to angles θ respectively n-1 and θ n+1 , then the longitudinal side length of the rectangular projection area corresponding to the current projection valid point is L i ×(|tan(θ n )-tan(θ n-1 )|+|tan(θ n+1 )-tan(θ n )|); where θ n Indicates the angle corresponding to the collection line bundle of the current projection valid point.

[0013] Preferably, there is an angle between the projection direction of the radar emission point and the normal direction of the surface of the detection area. Therefore, when calculating the area of ​​the rectangular projection surface corresponding to the effective projection point in step S400, a correction coefficient cosΦ is introduced to convert the area of ​​the rectangular projection surface. The calculation formula of the correction coefficient cosΦ is as follows:

[0014] ;

[0015] in, represents the direction vector of the radar emission point, Represents the normal vector of the detection area surface.

[0016] Preferably, in step S400, in order to avoid redundant area statistics caused by the overlap of multiple valid projection points in the rectangular projection area of ​​the same target area, it is necessary to perform de-overlapping and merging processing on all rectangular projection areas in three-dimensional space, which specifically includes the following process: for each valid projection point, a spatial coordinate system ( );in, Represents the normal vector of the detection area surface, Represents the normal vector of the detection area surface Any unit vector that is not collinear with the global direction The orthogonal vectors obtained by cross product are Represents a vector and The vector product of the rectangular projection area; according to the side length a i and b i , get the four corner points P of the rectangular projection area i1 To P i4 The coordinates in the space coordinate system are:

[0017] ;

[0018] ;

[0019] ;

[0020] ;

[0021] in, Indicates the projection effective point in the space coordinate system ( ) in the coordinate points.

[0022] Preferably, in step S400, the calculation of the point cloud coverage includes the following process: dividing each patch of the detection area into equally spaced grids according to the resolution and establishing a two-dimensional Boolean mask; mapping the rectangular projection area of ​​all projected valid points corresponding to each patch to the corresponding equally spaced grid; for each projected valid point, its projection area is marked as 1 if it is on the two-dimensional Boolean mask, and repeated areas are automatically merged; by counting the number of all non-zero mask units in the two-dimensional Boolean mask on each patch and multiplying it by the area of ​​a single grid, the total effective coverage area of ​​the radar point cloud on all patches is obtained; the obtained total effective coverage area is compared and calculated with the total area of ​​the patches of the detection area to obtain the coverage of the radar point cloud.

[0023] A radar layout rationality detection method comprises the following steps:

[0024] S110: Constructing a three-dimensional model of the area to be measured including multiple radars, and extracting a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar;

[0025] S210: Classifying different types of regions in the area to be measured according to their importance, and segmenting the three-dimensional model in step S110 according to the types of regions using a preset algorithm to obtain independent sub-models of the different types of regions in a unified coordinate system;

[0026] S310: Mapping the radar point cloud to the surface of each independent sub-model through simulation according to the characteristics of the radar, and then calculating the point cloud density, point cloud distribution uniformity, and point cloud coverage of each independent sub-model based on the projected valid points; wherein the point cloud coverage is obtained using the above-mentioned radar point cloud coverage detection method;

[0027] S410: Based on the classification of each independent sub-model, the rationality of the layout of the multi-point radar is comprehensively judged through the corresponding point cloud density, point cloud distribution uniformity and point cloud coverage.

[0028] Preferably, a three-dimensional model of the area to be measured including multiple facets is obtained by on-site scanning and data reconstruction; the class areas of the area to be measured include ground, buildings and vegetation, and the identification process of different class areas of the area to be measured is as follows: the normal angle, height and area of ​​each facet are identified; if the angle between the normal of the facet and the Z axis is less than 15°, and the height is lower than the set threshold, the facet is marked as "road surface"; if the angle between the normal of the facet and the Z axis is approximately perpendicular, and the area is greater than the set threshold, the facet is marked as "building surface"; if the facet height is irregular, distributed in the boundary area and has a high curvature, the facet is marked as "vegetation or obstruction".

[0029] Preferably, the calculation of the point cloud density in step S310 includes the following process: dividing each type of area into sub-areas of fixed scale, and counting the actual number of point clouds P falling into each sub-area. q , and then according to the actual number of point clouds P q and the area A of the corresponding sub-region q The actual point cloud density ρ is calculated by the ratio q ; Calculate the distance from the center point of each sub-area to all radar points to obtain the nearest radar point corresponding to each sub-area; For each sub-area, simulate the theoretical number of point clouds P generated under unobstructed conditions based on the parameter information of the nearest radar Gi , and then according to the theoretical point cloud number P Gi and the area A of the corresponding sub-region q The theoretical point cloud density ρ is calculated by the ratio of m ; The actual point cloud density ρ q and the theoretical point cloud density ρ mThe ratio of Y is used as an indicator of the actual point cloud perception ability G The calculation of the uniformity of point cloud distribution in step S310 includes the following process: subdividing each sub-region into multiple sub-grids, calculating the grid point cloud density according to the number of point clouds falling into each sub-grid; calculating the average value of the density of all grid point clouds corresponding to the sub-region and standard deviation σ ρ ; Actual point cloud distribution uniformity index U G The calculation formula is as follows:

[0030] ;

[0031] Here, α represents the normalization coefficient.

[0032] A radar control optimization method comprises the following steps:

[0033] S120: Constructing a three-dimensional model of the area to be measured including multiple radars, and extracting a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar;

[0034] S220: Mapping the point cloud to the detection area through simulation according to the characteristics of the radar, and calculating the completeness of the point cloud in each area of ​​the three-dimensional model based on the mapping results, where the completeness includes point cloud density, point cloud distribution uniformity, and point cloud coverage; wherein the completeness is suitable for being obtained by the radar layout rationality detection method according to any one of claims 6 to 8;

[0035] S320: Comparing the calculated completeness with a set threshold, marking areas that do not meet the threshold as abnormal perception performance areas and projecting them onto the ground of the 3D model to form blind areas;

[0036] S420: Retain the three closest radars for each blind spot and trace back the actual perception contributions of these three radars to the blind spot;

[0037] S500: Based on the backtracking results, perform at least one posture fine-tuning on the radars that have perception potential but insufficient actual contribution. Each posture fine-tuning recalculates the corresponding completeness. If the recalculated completeness meets the set threshold requirements and no new perception blind spots are introduced, the point optimization is considered successful. Otherwise, the radars in the measurement area are redeployed.

[0038] Preferably, the radar redeployment for the area to be measured in step S500 includes the following process: dividing the base map of the area to be measured into several numbered square areas containing complete location information; judging the coverage range of the current radar by the perception distance of the current radar, and selecting several candidate deployment areas based on this; calculating the radar perception projection union area of ​​each point according to the point cloud mapping method of step S220 to obtain the local coverage rate of the point; taking the optimal local coverage rate of the point and the optimal coverage rate point of the blind spot as the optimization target, the optimal deployment point area of ​​the radar is obtained from the candidate deployment area; when the first optimal deployment point is selected, regenerating the candidate deployment point set; simulating the cylindrical perception range A2 of each candidate point in the set and performing a union operation with the perception range A1 of the selected point to obtain the joint coverage area of ​​each candidate point, and then calculating the new joint coverage rate CR 1,2~all =(A1∪A2) / A m ; By introducing the coverage gain indicator ΔCR to judge the overall coverage improvement, ΔCR=CR 1,2~all -max(CR1, CR2); by introducing the overlap ratio indicator Determine the redundancy between points. ; Comprehensively consider the joint coverage rate CR 1,2~all , coverage gain index ΔCR and overlap ratio index , select the best point in the candidate deployment point set as the second deployment point; iteratively select the kth point step by step, each time based on the current deployment point set, and calculate the joint coverage rate CR 1,2,……,k~all , joint gain ΔCR k and overlap rate ; Then select the point with the minimum overlap rate, coverage gain index greater than the set threshold and the best joint coverage rate as the new deployment point; where CR1 represents the local coverage rate corresponding to the first deployment point, CR2 represents the local coverage rate corresponding to each candidate point in the set, and A m Represents the total surface area below the radar installation height in the 3D model, CR 1,2,……,k~all =(A1∪A2∪……∪A k ) / A m , ΔCR k =CR 1,2,……,k~all -CR 1,2,……,(k-1)~all , , A k Indicates the perception range corresponding to the k-th point, A i It represents the sensing range corresponding to the i-th point. The value range of i is {1, 2, ..., k-1}, and k>2.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] (1) This application can quantify the perception efficiency of each radar point with high precision during the design phase, provide an objective basis for subsequent point screening, effectively reduce sensor redundancy caused by misjudgment, overlap or blind spots, and significantly improve the system perception efficiency and deployment resource utilization.

[0041] (2) The radar combination perception coverage index and coverage gain index were introduced. The joint perception area of ​​multiple radar points was projected into the three-dimensional model of the service area to calculate the actual area, and the perception improvement effect after combination was quantified.

[0042] (3) By dividing the entire three-dimensional space of the service area into multiple sub-blocks of the area to be covered, after the initial point position is determined, the sub-block center point extraction and local optimal point position selection operations are gradually performed on the remaining areas, thereby realizing an iterative point layout strategy with target driving capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the workflow of the radar point cloud coverage detection method in this application.

[0044] Figure 2 Schematic diagram of the planar sensing range of the radar in this application.

[0045] Figure 3 Schematic diagram of the workflow of the radar layout rationality detection method in this application.

[0046] Figure 4 Schematic diagram of the workflow of the radar control optimization method in this application. DETAILED DESCRIPTION

[0047] Below, the present application is further described in conjunction with specific implementation methods. It should be noted that, in the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like are intended to mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0048] In the description of this application, it should be noted that for directional words, such as the terms "center", "horizontal", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and so on, indicating the orientation and position relationship are based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing this application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and cannot be understood as limiting the specific scope of protection of this application.

[0049] It should be noted that the terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0050] In this application, unless otherwise expressly specified or limited, terms such as "mounted," "connected," "connect," and "fixed" should be understood in a broad sense. For example, they may refer to connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on specific circumstances.

[0051] In this application, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Moreover, a first feature being "above," "above," and "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may include the first feature being directly below or obliquely below the second feature, or may simply mean that the first feature is lower in level than the second feature.

[0052] The terms "comprises" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units expressly listed, but may include other steps or units not expressly listed or inherent to such process, method, product or apparatus.

[0053] One aspect of the present application provides a radar point cloud coverage detection method, such as Figure 1 As shown, one of the preferred embodiments includes the following steps:

[0054] S100: Construct a three-dimensional model of the area to be measured and perform coordinate system conversion.

[0055] S200: Constructing a simulation model of the area to be measured according to the installation position of the radar, and extracting a three-dimensional model of the cylindrical detection area that meets the radar sensing range.

[0056] S300: Projecting the radar point cloud onto the surface of the detection area through simulation according to the characteristics of the radar, and constructing a rectangular projection area scanned by each projection valid point on the surface of the detection area.

[0057] S400: Counting the total effective coverage area of ​​the rectangular projection area corresponding to all the projection valid points, and calculating the point cloud coverage rate according to the ratio of the total effective coverage area to the total area of ​​all the patches in the detection area.

[0058] It is understandable that the technical solution of this application is applicable to a wide range of scenarios, such as roads, service areas, or parking lots. For ease of understanding, the area to be measured in this embodiment will be described in detail using a service area as an example. This application accurately calculates the perception capability of a single radar in a complex three-dimensional service area environment by combining characteristics such as the radar installation position, angle, measurement distance, and the resolution parameters of the radar itself. This indicator not only takes into account the theoretical perception range of the radar, but also uses the effective projected area it actually generates in the three-dimensional environment as the core measurement basis, thereby avoiding the rough evaluation problem of traditional radar deployment methods that only use coverage radius or blind spot distance as a reference.

[0059] Compared with traditional CAD layout methods or manual experience deduction methods, this method can quantify the perception efficiency of each point with high precision during the design phase, provide an objective basis for subsequent point screening, effectively reduce sensor redundant configuration caused by misjudgment, overlap or blind spots, and significantly improve system perception efficiency and layout resource utilization.

[0060] In this embodiment, in order to simulate and evaluate the coverage area of ​​the radar within the service area during step S100, a high-precision three-dimensional model of the service area must first be constructed. The construction of the three-dimensional model includes the following steps:

[0061] S101: Perform a comprehensive scan of the building structure, terrain, and facilities in the survey area to obtain point cloud data.

[0062] It is understandable that comprehensive scanning data of the area to be measured can be obtained through laser radar measurement data, for example, it can be collected by using a mapping car equipped with a laser radar, or it can be obtained through photogrammetry data.

[0063] S102: Denoising the point cloud data using a statistical outlier removal algorithm.

[0064] It is understandable that, during the actual acquisition process, point cloud data often contains outliers and noise caused by factors such as equipment errors and environmental occlusion (e.g., temporary vehicles, pedestrians, and reflective surfaces). Therefore, preprocessing of the raw point cloud data is required before constructing the 3D service area model. Numerous methods exist for preprocessing point cloud data. In this embodiment, the Statistical Outlier Removal (SOR) algorithm from the Open3D library is preferably used to filter the data. This method analyzes the average distance between each point and its k-nearest neighbors and sets a standard deviation threshold to identify and remove outliers that are far from their neighbors. This method has the advantage of preserving the continuity of the point cloud structure without damaging critical structures such as building edges or road edges. The working process of the SOR algorithm is well known to those skilled in the art and will not be repeated here.

[0065] S103: Convert the denoised point cloud data into a grid structure based on the Poisson reconstruction method.

[0066] It is understood that after noise removal, a surface model must be generated for the point cloud to facilitate subsequent radar projection analysis. While various methods exist for point cloud reconstruction, the Poisson reconstruction method is preferred in this embodiment. This method is suitable for dense and uniform point clouds and can generate closed and continuous triangular meshes. The specific working process of the Poisson reconstruction method is well known to those skilled in the art and will not be repeated here.

[0067] S104: Simplifying the obtained grid structure using a quadratic error measurement method to obtain a three-dimensional model of the area to be measured including multiple facets.

[0068] Understandably, the reconstructed model often contains a large number of redundant facets, which affects the efficiency of subsequent radar projection and intersection calculations. Therefore, model simplification is necessary. For example, some structures such as trees contain a large amount of facet information in their leaves. This information not only consumes a lot of computational space but also results in an excessive number of facets in the model, ultimately affecting the final coverage assessment.

[0069] This embodiment preferably uses a mesh simplification algorithm based on the quadric error metric (QEM) to perform global face reduction on the model. This algorithm can significantly reduce the number of facets while maintaining the model outline and main structure boundaries. The target number of facets for face reduction is set based on the simulation accuracy. For example, 1 million facets can be simplified to about 100,000 (retaining 10%) to reduce the model complexity while retaining the geometric structure as much as possible. During the simplification process, the model will also undergo normal vector re-estimation and vertex merging operations to ensure that the facets are normal and the boundaries are continuous. The specific working process of the quadratic error metric method is well known to those skilled in the art and will not be repeated here.

[0070] In this embodiment, multiple facets are obtained in step S104, and different facets may correspond to different target types. That is, the area to be measured includes different categories of areas, and different categories of areas have different corresponding integrity thresholds during the subsequent radar point placement optimization process based on their priorities. In order to distinguish different semantic objects in the subsequent coverage analysis, the multiple facets obtained need to be semantically annotated. This specifically includes the following process: identifying the normal angle, height, and area of ​​each facet. If the angle between the facet's normal and the Z axis is less than 15°, and the height is below a set threshold, the facet can be considered horizontal and labeled as "road surface." If the angle between the facet's normal and the Z axis is approximately perpendicular, that is, the angle between the facet's normal and the Z axis is close to 90°, and the area is greater than a set threshold, the facet can be labeled as "building surface." If the facet's height is irregular, distributed in a boundary area, and has high curvature, the facet is labeled as "vegetation or obstruction." After the classification is completed, each type of facet is exported as a different model file to facilitate subsequent projection analysis and occlusion evaluation.

[0071] To enhance classification accuracy, simple clustering and voxel segmentation methods can be introduced. For example, the voxel_down_sample_and_trace function in Open3D can be used to voxelize the model. Spatial connectivity analysis can then be performed to extract large, continuous areas of road or building structures while eliminating isolated, misclassified areas. The resulting 3D model will have clear structural divisions and optimized meshes, providing a lightweight and accurate input foundation for radar projection simulation.

[0072] It should be noted that since the original acquisition coordinate system of the service area model may be a local device coordinate system or a local coordinate system with a certain offset, in order to ensure the uniform calculation of the position of subsequent radar projection points, the coordinates of the constructed three-dimensional model need to be uniformly converted to the time coordinate system or the WGS-84 reference coordinate system.

[0073] Specifically, this embodiment uses known control points (such as ground-based calibration points or the center points of fixed facilities) as anchor points and utilizes rigid transformation for coordinate alignment. Least-squares registration or the ICP algorithm is used to fit the control point cloud, determining the rotation matrix and translation vector, thus achieving a global transformation of the point cloud coordinates. This operation ensures that subsequent radar point simulations strictly align with the actual scene coordinates, providing a foundation for radar perception range assessment.

[0074] It is understandable that in order to ensure that the subsequent radar point cloud projection accurately corresponds to the three-dimensional model, after the three-dimensional model is constructed, the installation position and coordinate system of the radar equipment need to be set and unified.

[0075] Specifically, the three-dimensional coordinates O (x, y, z) of the radar's center point are manually assigned within the modeling environment based on the radar's physical installation location. Its heading angles (pitch, yaw, and roll) are then set to form the radar's body coordinate system. The corresponding rotation matrix is ​​then calculated using Euler angles or quaternions, and combined with the translation vector to construct a homogeneous transformation matrix from the radar coordinate system to the world coordinate system. This transformation provides the foundational conversion framework for subsequent simulations of the projection direction of laser scan lines and detection of their intersection with 3D structures in the scene, ensuring that the simulation results are consistent with the actual radar's perception range.

[0076] In this embodiment, step S200 includes the following process: Based on the radar's vertical field of view and maximum ranging capability, the maximum horizontal distance r that the radar can perceive on the ground is calculated. A planar perception range equation for the radar is constructed, with the radar's position in the area to be measured as the center and r as the radius. The planar perception range equation is vertically stretched in space to obtain the radar's perception volume in three-dimensional space. The overlap between this perception volume and the three-dimensional model of the area to be measured is defined as the detection area. The vertical stretch height of the planar perception range equation is the radar's installation height.

[0077] It is understandable that in order to detect the coverage of a single radar, it is necessary to first determine the original coordinate information O(x o ,y o Afterwards, the radar's basic information is extracted, including its installation height H, horizontal and vertical resolutions, and its receptive field R. Next, based on the radar's center point O and receptive field, the corresponding point and radar receptive range are found in the 3D twin model of the service area.

[0078] Specifically, assume that the radar is installed on the top of a fixed pole and the radar height is recorded as H. It is known that the radar has a fixed vertical field of view angle range [θ min ,θ max], and the maximum ranging capability R. At this point, the actual receptive field of the radar on the three-dimensional model can be determined by the trigonometric function method. The radar beam is along the direction angle θ in the three-dimensional space. min The horizontal projection distance r after extending the length R can be obtained by the following expression: r=R×cos(θ min ).

[0079] The result of the above expression represents the maximum horizontal distance that the radar can perceive on the ground under the current installation height and field of view angle. m (x m ,y m ) is the center of the circle and r is the radius. The following two-dimensional perception range equation can be established: . Figure 2 As an example, the service area includes a service building, multiple parking areas, and a gas station. Assuming that the radar is installed in the middle of the parking area, a circular detection area can be obtained through the two-dimensional perception range equation established above, that is, Figure 2 The circular bright area in the .

[0080] Stretching the above equation vertically upward in space forms a cylinder, which can be defined as the radar's reachable perception volume in three-dimensional space. This cylinder is used to crop the 3D twin scene model to obtain the sub-model to be analyzed. This sub-model represents the portion of space that the radar can theoretically observe, i.e., the detection area in step S100.

[0081] It should be noted that after completing the preparation of the three-dimensional scene model of the service area and setting the radar coordinate system, in order to simulate the perception capability of the real radar equipment, it is necessary to accurately model the core parameters of the radar equipment and realize the simulation of the radar scanning process based on this. The goal of this step is to construct an equivalent scanning model based on the input radar equipment technical specifications, and abstract the scanning process into emitting multiple sets of rays in three-dimensional space to simulate its point cloud sampling behavior, and finally obtain the projection position of the point cloud emitted by the radar in the cylindrical detection area through simulation. For ease of understanding, the projection process of the radar point cloud in the cylindrical detection area will be described in detail below.

[0082] Specifically, the projection of the radar point cloud in step S300 includes the following process:

[0083] S301: Calculate the total number of sampling points corresponding to each radar frame according to the radar scanning parameters.

[0084] It is understandable that each frame of radar sampling data is determined by the angular resolution in the horizontal and vertical directions. The horizontal field of view angle range of the radar can be set to , the horizontal resolution of the radar is , the vertical field of view of the radar is [θ min ,θ max ], the horizontal resolution of the radar is Δθ. Then the calculation formula for the total number of sampling points T of the radar in each frame is as follows:

[0085] .

[0086] S302: Construct a three-dimensional vector for each point according to the radar scanning parameters.

[0087] It is understandable that the direction of each point can be expressed by converting spherical coordinates into a three-dimensional vector, and the direction vector set of all points is .

[0088] Among them, θ i Indicates the vertical angle sampling point of the i-th row, Indicates the horizontal angle sampling point of the jth row.

[0089] S303: Extend the three-dimensional vector of each point obtained, and calculate the intersection point between the extension line and the surface of the detection area.

[0090] It can be understood that the extension of the three-dimensional vector is to perform ray extension projection on all direction vectors in the radar coordinate system. Specifically, each direction vector is centered at the radar center point O. m As the starting point, along the direction d i Extensions simulate the laser beam firing behavior and test the intersection of these rays with the cropped 3D sub-model.

[0091] S304: If there is an intersection, it is considered as a valid projection point and summarized.

[0092] It is understandable that after completing the intersection test in step S303, for each point P in the radar coordinate system i (x i ,y i , z i ), find the extension line to the 3D model according to its direction vector, and calculate the exact intersection coordinates Q of the extension line and the 3D sub-model i (x im ,y im , z im ), if there is an intersection, it is regarded as a valid projection point; if there is no intersection or it exceeds the maximum range R, it is regarded as a "missed" point and no point cloud is generated. Finally, the point clouds generated by all valid projection points can be summarized into a set Q.

[0093] After completing the radar point cloud simulation projection, this embodiment provides a radar perception completeness assessment method based on point-level projected area estimation to further measure the radar's coverage of the service area scene model at its current deployment location. This method differs from traditional grid-based statistical methods. Instead, it uses the imaging characteristics of the radar point cloud to accurately estimate the perception area corresponding to each radar sampling point on the 3D model surface, namely the rectangular projected area, thereby providing a more detailed measurement of overall coverage.

[0094] Specifically, the construction of the rectangular projection area corresponding to each effective projection point in step S300 includes the following process: for each effective projection point Q i , according to its actual sampling distance R i The horizontal and vertical resolution parameters of the radar are used to construct a projection matrix area, that is, a rectangular projection surface area, to approximately represent the perception coverage range of the point in space.

[0095] In the lateral direction (i.e. along the horizontal direction of the laser), since the scanning angle of the radar is evenly distributed, it can be simplified as follows: the effective point Q is projected i Distance to radar L i Multiply by the horizontal angular resolution , calculate the arc length between adjacent points on the same circumference of the projection effective point, which can be used as the horizontal side length a of the rectangular projection area i ;Right now Here a i It reflects the difference in lateral spatial distribution caused by the radar's horizontal angular resolution at the current distance.

[0096] In the longitudinal direction, considering that radar often scans vertically through multiple beams at different angles, the spatial distances between a point and the upper and lower adjacent beams are not completely symmetrical, so an asymmetric bidirectional edge is used to construct the longitudinal length of the projection area. Specifically, if the current projection valid point is collected by the nth vertical beam, then the two adjacent beams above and below the current projection valid point correspond to angles θ n-1 and θ n+1 , then the longitudinal side length b of the rectangular projection area corresponding to the current projection effective point is i =L i ×(|tan(θ n )-tan(θ n-1 )|+|tan(θ n+1 )-tan(θ n )|); where θ n Indicates the angle corresponding to the collection line bundle of the current projection valid point.

[0097] In this embodiment, to further improve the accuracy of radar point cloud coverage analysis, the impact of the incident angle between the laser beam direction and the impact surface should be considered. When there is an angle between the projection direction of the radar emission point and the normal vector of the detection area surface, its actual projected area on the detection area surface will be scaled due to the oblique incidence. Therefore, when calculating the area of ​​the rectangular projection surface corresponding to the projected effective point in step S400, a correction coefficient cosΦ is introduced to convert the area of ​​the rectangular projection surface area; the calculation formula of the correction coefficient cosΦ is as follows:

[0098] .

[0099] in, represents the direction vector of the radar emission point, Represents the normal vector of the detection area surface.

[0100] The effective coverage area A of the final radar emission point on the surface of the detection area is i It can be expressed as:

[0101] A i =a i ×b i ×|cosΦ|.

[0102] In this embodiment, during step S400, to avoid redundant area statistics caused by overlapping rectangular projections of multiple valid projected points within the same target area, all rectangular projections are subjected to a three-dimensional de-overlapping and merging process. Unlike conventional two-dimensional planar Boolean operations, this method fully utilizes the normal vector information of the surface to which each point belongs, directly constructing a three-dimensional projected rectangular patch for each point on the three-dimensional model, and then calculating the joint coverage area of ​​these three-dimensional patches using a geometric Boolean algorithm.

[0103] Specifically, for each projection valid point Q i There is a local plane coordinate system for the formation of rectangular projection patches, so a spatial coordinate system orthogonal to the corresponding local plane coordinate system can be constructed in three-dimensional space ( ).in, Represents the normal vector of the detection area surface, Represents the normal vector of the detection area surface Any unit vector that is not collinear with the global direction The orthogonal vectors obtained by cross product are Represents a vector and The vector product of . and vector product The expression is as follows:

[0104] .

[0105] Then, the side length a of the rectangular projection area calculated according to the above content i and b i , construct the four corner points P of the rectangular projection area i1 To P i4 The coordinates in the space coordinate system are:

[0106] .

[0107] .

[0108] .

[0109] .

[0110] in, Represents the projection effective point Q i In the spatial coordinate system ( ). These four points constitute the rectangular projection patch of the radar emission point on the surface of the detection area in three-dimensional space, which can be regarded as a local plane quadrilateral with a clear spatial position and geometric boundary.

[0111] In this embodiment, after obtaining the rectangular projection patches corresponding to all valid points projected by the radar on the detection area surface, the 3D service area model has been simplified, resulting in relatively regular patches and a large concentration of points in the plane. Therefore, in this embodiment, a 3D surface mapping algorithm combined with a bitmap overlay algorithm is preferred for area union. This method offers the advantages of simple implementation, controllable accuracy, and high computational efficiency, making it particularly suitable for scenarios with numerous overlapping rectangles and densely repetitive patches.

[0112] Specifically, in step S400, the calculation of the point cloud coverage includes the following process: First, each patch in the detection area is divided into equally spaced grids based on resolution and a two-dimensional Boolean mask is created. Then, the rectangular projection area of ​​all projected valid points corresponding to each patch is mapped to the corresponding equally spaced grid. For each projected valid point, if its projected area falls on the two-dimensional Boolean mask, it is marked as 1, and duplicate areas are automatically merged. Finally, the total effective coverage area of ​​the radar point cloud for all patches is calculated by counting the number of non-zero mask cells in the two-dimensional Boolean mask on each patch and multiplying it by the area of ​​a single grid. The obtained total effective coverage area is then compared with the total area of ​​the patches in the detection area to calculate the coverage of the radar point cloud.

[0113] It's important to note that when calculating the total area of ​​the patches forming the 3D sub-model of the detection area, only the area corresponding to the patches below the radar height is calculated. When calculating radar point cloud coverage, a higher value indicates more complete coverage, while a lower value indicates blind spots or occlusion.

[0114] In this embodiment, after completing the calculation of the above-mentioned radar point cloud coverage, the local point cloud coverage corresponding to each facet can also be obtained. Then, based on the constructed three-dimensional model of the area to be measured, a digital twin three-dimensional model can be generated for interface display. When displaying the digital twin model, different facets in the detection area can be assigned different color values ​​or transparency according to the corresponding radar point cloud coverage. A color coverage rendering can then be generated so that users can intuitively view the model coverage from any perspective.

[0115] Another aspect of the present application provides a radar arrangement rationality detection method, such as Figure 3 As shown, one of the preferred embodiments includes the following steps:

[0116] S110: Construct a three-dimensional model of the area to be measured including multiple radars, and extract a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar.

[0117] It is understandable that the specific construction method of the three-dimensional model and the detection area can refer to the aforementioned step S100; the difference is that the constructed three-dimensional model includes radars at multiple points; therefore, it will not be repeated here.

[0118] S210: Classify different types of areas in the area to be measured according to their importance, and segment the three-dimensional model in step S110 according to the types of areas using a preset algorithm to obtain independent sub-models of different types of areas in a unified coordinate system.

[0119] It is understandable that the specific division process of the class area or the independent sub-model can refer to the aforementioned step S104; therefore, it will not be repeated here.

[0120] S310: Mapping the radar point cloud to the surface of each independent sub-model through simulation according to the characteristics of the radar, and then calculating the point cloud density, point cloud distribution uniformity and point cloud coverage of each independent sub-model based on the projected valid points.

[0121] It is understandable that the acquisition of effective projection points and the specific mapping method can refer to the aforementioned step S300; the calculation of point cloud coverage can refer to the aforementioned step S400; therefore, they will not be repeated here.

[0122] S410: Based on the classification of each independent sub-model, the rationality of the layout of the multi-point radar is comprehensively judged through the corresponding point cloud density, point cloud distribution uniformity and point cloud coverage.

[0123] In this embodiment, the point cloud density is used to verify the actual point cloud perception capability of each type of area, which specifically includes the following process: a fixed-scale sub-area G of each type of area is used to verify the actual point cloud perception capability of each type of area. i Divide and count the actual number of point clouds P that fall into each sub-area q , and then according to the actual number of point clouds P q and the area A of the corresponding sub-region q The actual point cloud density ρ is calculated by the ratio q Calculate the distance from the center point of each sub-area to all radar points to obtain the nearest radar point corresponding to each sub-area. For each sub-area, simulate the theoretical number of point clouds generated under unobstructed conditions based on the parameter information of the nearest radar. Gi , and then according to the theoretical point cloud number P Gi and the area A of the corresponding sub-region q The theoretical point cloud density ρ is calculated by the ratio of m . The actual point cloud density ρ q and the theoretical point cloud density ρ m The ratio of Y is used as an indicator of the actual point cloud perception ability G .

[0124] It can be understood that for the sub-region G i The fixed scale of the sub-area G can be selected according to the actual needs of those skilled in the art. i The scale of the sub-area G can be 0.1m×0.1m, 0.2m×0.2m, 0.3m×0.3m, etc. i The distance calculation process from the center to each radar point is as follows: Get the sub-area G i The coordinates of the center point O Gi (x Gi ,y Gi , z Gi ), and the radar point O R (x Ri ,y Ri , z Ri ) ; Substitute the obtained coordinates into the distance d GR In the calculation formula, the distance d GR The calculation formula is as follows:

[0125] .

[0126] Find the distance from the current sub-area G from the distances of all sub-areas to the radar point i The nearest radar point R i, and obtain the radar parameter information of the radar point, including the horizontal resolution And the vertical resolution Δθ. Then, according to the resolution, the number of points P that should be generated in this sub-area theoretically under unobstructed conditions is simulated. Gi .

[0127] It should be known that a threshold can be set based on experience: if Y G >0.8, it is considered that the density of the sub-area is sufficient and the point cloud coverage meets the perception requirements; that is, it is considered that the point cloud density of the sub-area can meet the current layout requirements. Specifically, the perception range of the radar close to the sub-area can basically cover the current sub-area. Even if there are some areas in the current sub-area that are partially blocked or cannot be perceived, the radar at the remaining points can complete the data of this area. In this way, although the sub-area is fully covered, the radar point cloud density will be reduced, so a threshold of 0.8 is set to retain a certain degree of computational redundancy. If 0.4 <Y G <0.8, it indicates that the point cloud density of the current sub-area is low, and there is a certain risk of perception blind spots; if Y G <0.4, it indicates that there is severe occlusion or point cloud missing in the sub-area, and the radar point layout needs to be re-optimized.

[0128] In this embodiment, judging the coverage adequacy of a sub-region only by the density index may not reflect the uniformity of the density distribution within the sub-region. For example, even if the overall density of a sub-region is high, if there is a local area with complete lack of point cloud data, it will still affect the perception effect. Therefore, it is necessary to verify the uniformity of the actual point cloud distribution of each type of area, which specifically includes the following process: i Subdivided into n×n subgrids g ij , n>1; according to each sub-grid g ij The number of point clouds P gij Calculate the grid point cloud density ρ gij =P gij / A gij Among them, A gij Represents the grid area. After completing the calculation of the grid point cloud density corresponding to all grids in the sub-area, calculate the average value of the grid point cloud density corresponding to the sub-area and standard deviation σ ρ ; Among them, the average and standard deviation σ ρ The specific calculation formula is as follows:

[0129] .

[0130] .

[0131] Based on the average value of all grid point cloud densities corresponding to the sub-region and standard deviation σ ρ , introducing density uniformity index .

[0132] .

[0133] In order to quantify this indicator, it is necessary to normalize the indicator to a range of [0, 1] and define the actual point cloud distribution uniformity index U G ;U G The calculation formula is as follows:

[0134] .

[0135] Where α is the normalization coefficient, which represents the tolerable density fluctuation. Approaching 0, it means that the point cloud is very evenly distributed, then U G Approaches 1; otherwise, U G It approaches 0, which means that the density distribution is very uneven.

[0136] It is understandable that for the ground area, as the core target area of ​​the radar perception system, it is required to be fully perceived and covered by the point cloud; therefore, it is necessary to meet the following requirements: G >0.8, U G >0.8; if a sub-region does not meet any of the above conditions, it will be marked as an "insufficient coverage area" and will be given priority in subsequent optimization.

[0137] For the building wall area, as an auxiliary perception area, its perception targets include vehicle rear occlusion, residual detection of human targets in a static state, etc. Therefore, it is necessary to meet the following requirements: G >0.5, U G >0.5.

[0138] Green areas such as vegetation are generally inaccessible and have natural occlusion characteristics, making them less important in the point cloud perception system. Therefore, this method allows for sparse point clouds or even no perception in these areas; they are only considered as perception redundancy when the area is adjacent to a critical path.

[0139] It should be noted that after initially assessing the point cloud distribution and density of various areas, further geometric coverage measurement methods based on the projected area of ​​the point cloud are required to accurately quantify the perceived quality. While the aforementioned methods based on point count distribution and density can reflect point cloud distribution trends, they cannot truly reflect the actual spatial area covered by the point cloud. Therefore, a more rigorous coverage assessment method based on 3D scene models is needed.

[0140] In this embodiment, after completing the radar point cloud simulation projection, to further measure the radar's coverage capability of the service area scene model at its current deployment location, this application provides a radar point cloud coverage detection method based on point-level projection area estimation. This method differs from traditional grid-accumulation-based statistical methods. Instead, it uses the imaging characteristics of the radar point cloud to accurately estimate the corresponding perception area (i.e., the rectangular projection area) of each radar sampling point on the 3D model surface, thereby more meticulously measuring the overall coverage.

[0141] Understandably, for service area scenarios, high-density, high-coverage perception is required for ground areas, where both people and vehicles move. Therefore, radar point cloud coverage must be above 98%. Building walls close to the ground are used to capture moving targets or sudden behaviors in potentially obstructed areas. Radar point cloud coverage of building walls must be above 60%. A moderate reduction in coverage is permitted for areas such as green belts and bushes away from traffic paths, meaning no radar point cloud coverage requirement is required for vegetation areas.

[0142] In summary, to determine the rationality of multi-point radar deployment for each independent sub-model within the measurement area, it is necessary to examine multi-point radar coverage from multiple perspectives, including 3D model mapping, density assessment, and uniformity analysis. This not only enables an accurate assessment of global coverage but also provides a quantitative basis and optimization direction for radar deployment strategies through regional classification and dynamic indicators.

[0143] Another aspect of the present application provides a radar deployment optimization method, such as Figure 4 As shown, one of the preferred embodiments includes the following steps:

[0144] S120: Construct a three-dimensional model of the area to be measured including multiple radars, and extract a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar.

[0145] It is understandable that this step is the same as step S110 and thus will not be repeated here.

[0146] S220: Mapping the point cloud to the detection area through simulation according to the characteristics of the radar, and calculating the completeness of the point cloud in each area of ​​the three-dimensional model based on the mapping results. The completeness includes point cloud density, point cloud distribution uniformity, and point cloud coverage.

[0147] It is understandable that the specific method of corresponding point cloud mapping can refer to the above-mentioned step S300, and the calculation of point cloud density, point cloud distribution uniformity and point cloud coverage included in the completeness can refer to the above-mentioned step S310, so it will not be repeated here.

[0148] S320: Compare the calculated completeness with a set threshold, mark the area that does not meet the threshold requirement as a perception performance abnormal area and project it onto the ground of the three-dimensional model to form a blind area.

[0149] S420: Retain the three closest radars for each blind spot, and trace back the actual perception contributions of the three radars to the blind spot.

[0150] S500: Based on the backtracking results, perform at least one posture fine-tuning on the radars that have perception potential but insufficient actual contribution; each posture fine-tuning recalculates the corresponding completeness. If the recalculated completeness meets the set threshold requirements and does not introduce new perception blind spots, the point optimization is considered successful; otherwise, the radars in the measurement area are re-deployed.

[0151] It is understood that after completing the radar layout rationality assessment for the three-dimensional service area model, the current perception coverage of multiple radar points within the service area model can be obtained, including the point cloud density, density uniformity, and regional coverage level of each area. Based on this coverage assessment result, this application aims to propose a highly targeted and deployable radar deployment optimization method. The main objectives include: 1) addressing coverage blind spots caused by radar obstruction or inadequate point layout; 2) improving point cloud density and uniformity in key areas (such as floors and walls); 3) reducing the system burden and deployment costs caused by redundant perception; and 4) optimizing radar parameters (angle, height, and orientation) to improve unit radar coverage efficiency. To achieve these goals, this embodiment proposes a two-level optimization strategy: a point addition judgment mechanism and a parameter fine-tuning optimization mechanism. Supported by a multi-objective optimization algorithm, this establishes a comprehensive radar deployment improvement process. For ease of understanding, a detailed description is provided below.

[0152] In this embodiment, first, in step S320, based on the radar point cloud integrity evaluation results of the aforementioned service area 3D model, combined with the engineering tolerance threshold, the minimum reception thresholds for the three indicators of point cloud density, point cloud distribution uniformity, and point cloud coverage are set. Among them, the point cloud coverage is not less than 95%, the point cloud density is greater than 0.8, and the point cloud distribution uniformity needs to be greater than 0.7. Areas that do not meet any of the indicators are marked as abnormal perception performance areas and are included in the set of areas to be optimized G. u The set of regions to be optimized G u Consists of all sub-areas that are under-covered, under-densified, or unevenly distributed.

[0153] In this embodiment, after obtaining the set of regions to be optimized G u Then, for the set G uAll abnormal subgrids in the model are spatially extracted and aggregated. Specifically, the coordinates of the boundary corner points of each abnormal perception performance area are first extracted and projected onto the ground area of ​​the three-dimensional model to form a two-dimensional defect distribution map. The connected region clustering method is used to merge the projection areas of adjacent abnormal perception performance areas into several continuous perception blind areas C1, C2, C3, ..., C k ; Assign a unique number to each blind area and extract its area and boundary information. By traversing each blind area C j The coordinates of the boundary corners are extracted, and the four corners corresponding to the maximum distance are used to construct two diagonal lines and find their intersection point O j As the blind spot C j The geometric center point of .

[0154] At the midpoint O of the blind zone j Afterwards, the spatial distances to all deployed radar locations are calculated, and the three closest radars are retained. The three radars' actual perception contributions to the blind spot are then retrospectively checked, including whether their perception ranges fully or partially cover the blind spot and whether there is redundant coverage outside their ranges. If a radar is found to have potential perception but insufficient contribution (e.g., due to limited angles or critical distances), the radar pose fine-tuning phase begins.

[0155] In this embodiment, the radar posture fine-tuning includes the following process: the base map of the three-dimensional model is divided into several candidate sub-block areas. And the radar is translated or fine-tuned in a small range; for radar translation, its sensing radius r range can be divided into a 3×3 pattern, and the translation is performed within the 3×3 pattern, with each translation being one grid distance; for radar posture fine-tuning, its adjustment range includes pitch angle ±3° and yaw angle ±5°, and each adjustment angle can be 1°, which can be selected according to actual needs. For each posture fine-tuning process, the radar is simulated to detect the blind area C. j The projection coverage effect of the three radars is calculated and the updated blind area C is calculated. j If the condition CR>98% is met, Y G If the value is >0.8 and no new blind spots are introduced, the point is considered successfully optimized and the current parameters are recorded as candidate replacement points. If all pose fine-tuning actions fail to meet the perception requirements, the point re-positioning mechanism is activated.

[0156] In this embodiment, the radar redeployment of the area to be measured in step S500 includes the following process: dividing the base map of the area to be measured into several numbered square areas containing complete location information; judging its coverage range by the perception distance of the current radar, and determining the coverage range according to the number of covered square areas, and selecting several candidate deployment areas based on this. After obtaining the candidate deployment areas, the three-dimensional coordinates of the center point of each candidate area are calculated as the initial deployment point. Combined with the point cloud mapping method of the aforementioned step S220, the radar perception projection union area A of each point is calculated respectively. proj To obtain the local coverage CR of the point i~all , and its share relative to the overall 3D model of the server. i~all =A proj / A m Finally, the optimal local coverage of the point and the optimal coverage of the blind area are used as optimization targets to obtain the optimal radar deployment area from the candidate deployment area.

[0157] Understandably, during radar point optimization, the primary goal is to ensure maximum radar coverage at each location. Furthermore, for the aforementioned areas of abnormal perception performance, it's crucial to ensure that the newly deployed radars effectively cover the blind spots created by their projections. Otherwise, radar relocation will be meaningless. For ease of understanding, the following describes the radar point relocation process in detail.

[0158] Specifically, the entrance and exit of the service area is the core area of ​​the server and needs to be given priority. Then, the first optimal layout point can be selected based on the optimal coverage of the entrance and exit of the service area. After the first optimal layout point is selected, the remaining uncovered area of ​​the service area is divided into several independent sub-areas, and the candidate layout point set is regenerated according to the area number. For each candidate point in the set, its own cylindrical perception range A2 is simulated and combined with the perception range A1 of the selected point to obtain the joint coverage area A of each candidate point. 1∪2 =A1∪A2, and then calculate the new joint coverage rate CR 1,2~all .

[0159] Among them, CR 1,2~all =(A1∪A2) / A m .

[0160] In order to judge whether the introduction of new deployment points has marginal optimization significance, the coverage gain indicator ΔCR is introduced to judge the overall coverage improvement.

[0161] Where ΔCR=CR 1,2~all-max(CR1, CR2); CR1 represents the local coverage corresponding to the first deployment point, and CR2 represents the local coverage corresponding to each candidate point in the set.

[0162] It is understood that if ΔCR is significantly greater than the set threshold, it means that the new deployment point has substantially improved the overall coverage. The specific value of the threshold can be set according to the actual needs of those skilled in the art, for example, 1.5% to 3%.

[0163] At the same time, considering the redundancy problem between the layout points, the overlap rate index is introduced Determine the redundancy between points.

[0164] in, .

[0165] It is understandable that the overlap ratio indicator is incorporated into the joint objective function as a penalty factor to control the waste of resources caused by repeated coverage among multiple radars.

[0166] Taking into account the combined coverage ratio CR 1,2~all , coverage gain index ΔCR and overlap ratio index , select the best point in the candidate layout point set as the second layout point.

[0167] As the number of deployed points increases, the kth point is gradually selected in an iterative manner, k>2; each time the joint coverage rate CR is calculated based on the current set of deployed points. 1,2,……,k~all , joint gain ΔCR k and overlap rate .

[0168] CR 1,2,……,k~all =(A1∪A2∪……∪A k ) / A m .

[0169] ΔCR k = CR 1,2,……,k~all -CR 1,2,……,(k-1)~all .

[0170] .

[0171] Among them, A k Indicates the perception range corresponding to the k-th point, A i It represents the perception range corresponding to the i-th point, and the value range of i is {1, 2, ..., k-1}.

[0172] Finally, the points with the lowest overlap, coverage gain greater than a set threshold, and optimal combined coverage are selected as new deployment points. The entire deployment process ends when coverage reaches 99% of the entire service area and all key areas have no blind spots, thus forming a multi-radar deployment strategy that minimizes the number of deployment points and maximizes perception.

[0173] It can be understood that by introducing the radar combination perception coverage index and coverage gain index, the joint perception area of ​​multiple radar points is projected into the three-dimensional model of the service area for actual area calculation, and the perception improvement effect after the combination is quantified. This mechanism allows the system to not only focus on the coverage capability of the newly added points themselves during the point selection process, but also evaluate their marginal contribution to the existing deployed points, thereby achieving the optimal incremental selection. In addition, by introducing the overlap penalty term for redundant coverage between points, the problem of excessive overlap in coverage areas caused by multi-point collaboration can be effectively avoided, and coverage integrity and resource efficiency can be coordinated from a global perspective. This optimization model is significantly superior to existing "equidistant deployment" or "area uniform distribution" algorithms, and can effectively improve the overall coverage integrity, stability and intelligent adaptability of the radar network.

[0174] By dividing the entire three-dimensional space of the service area into multiple sub-blocks of the area to be covered, after the initial point location is determined, the sub-block center of gravity point extraction and local optimal point selection operations are gradually performed on the remaining areas, thereby realizing an iterative point deployment strategy with target-driven capabilities. This deployment process from "maximum coverage area → local blind spot filling → global optimization termination" is significantly different from the traditional global one-time point generation method. Its advantage is that each decision step is based on the actual coverage feedback of the previous stage, which is highly adaptable and dynamically optimal. In addition, during the deployment process, the system continuously updates multiple constraint information such as the distribution of perception blind spots, joint coverage indicators, and redundant overlap between points. It can comprehensively consider multiple factors such as perception integrity, point redundancy rate, cost constraints, and regional priority to achieve flexible, intelligent, and actual demand-oriented point generation and optimization. This method is highly scalable and suitable for service area deployment tasks of different scales and forms.

[0175] The above describes the basic principles, main features, and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-described embodiments. The above-described embodiments and the specification merely illustrate the principles of the present application. Various changes and improvements may be made to the present application without departing from the spirit and scope of the present application. These changes and improvements fall within the scope of the present application for which protection is sought. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A radar point cloud coverage detection method, characterized in that: The steps include: S100: constructing a three-dimensional simulation model of the area to be measured and performing coordinate system conversion; S200: constructing a simulation model of the area to be measured according to the installation position of the radar, and extracting a three-dimensional model of the cylindrical detection area that meets the radar sensing range; S3 00: The radar point cloud is projected onto the surface of the detection area through simulation according to the characteristics of the radar, and a rectangular projection area is constructed where each effective projection point scans the surface of the detection area. S400: Counting the total effective coverage area of ​​the rectangular projection area corresponding to all the valid projection points, and calculating the point cloud coverage rate according to the ratio of the total effective coverage area to the total area of ​​all the patches in the detection area; The construction of the rectangular projection area corresponding to each effective projection point in step S300 includes the following process: In the lateral direction, the distance L from the projected effective point to the radar i Multiply by the horizontal angular resolution and calculate the arc length between adjacent points on the same circumference of the projection effective point as the horizontal side length of the rectangular projection area; In the longitudinal direction, if the current projection valid point is collected by the nth vertical beam, then the two adjacent beams above and below the current projection valid point correspond to angles θ n-1 and θ n+1 , then the longitudinal side length of the rectangular projection area corresponding to the current projection valid point is L i ×(|tan(θ n )-tan(θ n-1 )|+|tan(θ n+1 )-tan(θ n )|); Among them, θ n Indicates the angle corresponding to the current projection valid point acquisition line bundle; In step S400, the calculation of point cloud coverage includes the following process: Each patch of the detection area is divided into equally spaced grids according to the resolution and a two-dimensional Boolean mask is created; Map the rectangular projection area of ​​all valid projection points corresponding to each patch to the corresponding equidistant grid; For each valid projection point, if its projection area is on the two-dimensional Boolean mask, it is marked as 1, and the repeated areas are automatically merged; The total effective coverage area of ​​the radar point cloud on all patches is obtained by counting the number of all non-zero mask cells in the two-dimensional Boolean mask on each patch and multiplying it by the area of ​​a single grid. The obtained total effective coverage area is compared with the total area of ​​the patches in the detection area to obtain the coverage rate of the radar point cloud.

2. The radar point cloud coverage detection method according to claim 1, wherein: There is an angle between the projection direction of the radar emission point and the normal direction of the detection area surface. Therefore, when calculating the area of ​​the rectangular projection area corresponding to the projection effective point in step S400, a correction coefficient needs to be introduced. Convert the area of ​​the rectangular projection surface; Correction factor The calculation formula is as follows: ； in, represents the direction vector of the radar emission point, Represents the normal vector of the detection area surface.

3. The radar point cloud coverage detection method according to claim 2, wherein: In step S400, in order to avoid redundant area statistics caused by the overlap of multiple valid projection points in the rectangular projection area of ​​the same target area, it is necessary to perform de-overlapping and merging processing on all rectangular projection areas in three-dimensional space, which specifically includes the following steps: For each effective projection point, a spatial coordinate system orthogonal to the corresponding local plane coordinate system is constructed ( ); in, Represents the normal vector of the detection area surface, Represents the normal vector of the detection area surface Any unit vector that is not collinear with the global direction The orthogonal vectors obtained by cross product are Represents a vector and The vector product of According to the side length a of the rectangular projection area i and b i , get the four corner points P of the rectangular projection area i1 To P i4 The coordinates in the space coordinate system are: ； ； ； ； in, Indicates the projection effective point in the space coordinate system ( ) in the coordinate points.

4. A radar layout rationality detection method, characterized in that: The steps include: S110: Constructing a three-dimensional model of the area to be measured including multiple radars, and extracting a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar; S210: Classifying different types of regions in the area to be measured according to their importance, and segmenting the three-dimensional model in step S110 according to the types of regions using a preset algorithm to obtain independent sub-models of the different types of regions in a unified coordinate system; S310: Mapping the radar point cloud to the surface of each independent sub-model through simulation according to the characteristics of the radar, and then calculating the point cloud density, point cloud distribution uniformity, and point cloud coverage of each independent sub-model based on the projected valid points; wherein the point cloud coverage is obtained by the radar point cloud coverage detection method according to any one of claims 1 to 3; S410: Based on the classification of each independent sub-model, the rationality of the layout of the multi-point radar is comprehensively judged through the corresponding point cloud density, point cloud distribution uniformity and point cloud coverage.

5. The radar layout rationality detection method according to claim 4, characterized in that: Obtaining a three-dimensional model of the area to be measured including multiple facets by on-site scanning and data reconstruction; The categories of the area to be measured include ground, buildings, and vegetation. The identification process of different categories of the area to be measured is as follows: Identify the normal angle, height and area of ​​each facet; If the angle between the normal of a patch and the Z axis is less than 15° and the height is lower than the set threshold, the patch is labeled as "road surface"; if the angle between the normal of a patch and the Z axis is approximately perpendicular and the area is greater than the set threshold, the patch is labeled as "building surface"; if the patch height is irregular, distributed in the boundary area and has high curvature, the patch is labeled as "vegetation or obstruction".

6. The radar layout rationality detection method according to claim 4, characterized in that: The calculation of the point cloud density in step S310 includes the following process: Divide each type of area into sub-regions of fixed scale and count the actual number of point clouds P falling into each sub-region q , and then according to the actual number of point clouds P q and the area A of the corresponding sub-region q The actual point cloud density is calculated by the ratio of ; Calculate the distance from the center point of each sub-area to all radar points to obtain the nearest radar point corresponding to each sub-area; For each sub-area, the theoretical number of point clouds P generated under unobstructed conditions is simulated based on the parameter information of the nearest radar. Gi , and then according to the theoretical point cloud number P Gi and the area A of the corresponding sub-region q The theoretical point cloud density is calculated by the ratio of ; The actual point cloud density Compared with the theoretical point cloud density The ratio of Y is used as an indicator of the actual point cloud perception ability G ; The calculation of the point cloud distribution uniformity in step S310 includes the following process: Each sub-region is subdivided into multiple sub-grids, and the grid point cloud density is calculated according to the number of point clouds falling into each sub-grid; Calculate the average value of all grid point cloud densities corresponding to the sub-region and standard deviation ; Actual point cloud distribution uniformity index U G The calculation formula is as follows: ； Here, α represents the normalization coefficient.

7. A radar control optimization method, characterized in that: The steps include: S120: Constructing a three-dimensional model of the area to be measured including multiple radars, and extracting a cylindrical detection area that meets the sensing range of each radar according to the installation position of each radar; S220: Mapping the point cloud to the detection area through simulation according to the characteristics of the radar, and calculating the completeness of the point cloud in each area of ​​the three-dimensional model based on the mapping results, where the completeness includes point cloud density, point cloud distribution uniformity, and point cloud coverage; wherein the completeness is suitable for being obtained by the radar layout rationality detection method according to any one of claims 4 to 6; S320: Comparing the calculated completeness with a set threshold, marking areas that do not meet the threshold as abnormal perception performance areas and projecting them onto the ground of the 3D model to form blind areas; S420: Retain the three closest radars for each blind spot and trace back the actual perception contributions of these three radars to the blind spot; S500: Based on the backtracking results, perform at least one posture fine-tuning on the radars that have perception potential but insufficient actual contribution. Each posture fine-tuning recalculates the corresponding completeness. If the recalculated completeness meets the set threshold requirements and no new perception blind spots are introduced, the point optimization is considered successful. Otherwise, the radars in the measurement area are redeployed.

8. The radar control optimization method according to claim 7, characterized in that: The radar redeployment in the area to be measured in step S500 includes the following process: Divide the base map of the area to be measured into several numbered square areas containing complete location information; determine its coverage range based on the current radar's sensing distance, and select several candidate deployment areas based on this; According to the point cloud mapping method of step S220, the radar perception projection union area of ​​each point is calculated to obtain the local coverage rate of the point; Taking the optimal local coverage of the point and the optimal coverage of the blind area as the optimization target, the optimal radar deployment area is obtained from the candidate deployment area; When the first optimal layout point is selected, the candidate layout point set is regenerated; Simulate the cylindrical perception range A2 of each candidate point in the set and perform a union operation with the perception range A1 of the selected point to obtain the joint coverage area of ​​each candidate point, and then calculate the new joint coverage rate CR 1,2~all =(A1∪A2) / A m ; By introducing the coverage gain indicator CR judges the overall coverage improvement. CR=CR 1,2~all -max(CR1,CR2); By introducing the overlap ratio indicator Determine the redundancy between points. =(A1∩A2) / min(A1,A2); Comprehensive consideration of the joint coverage ratio CR 1,2~all , coverage gain index CR and overlap rate indicators , select the best point in the candidate deployment point set as the second deployment point; Iteratively select the kth point step by step, each time based on the current set of deployed points, and calculate the joint coverage rate CR 1,2,……,k~all , joint gain CR k and overlap rate ; Then select the point with the minimum overlap rate, coverage gain index greater than the set threshold and the best joint coverage rate as the new deployment point; Among them, CR1 represents the local coverage rate corresponding to the first deployment point, CR2 represents the local coverage rate corresponding to each candidate point in the set, and A m Represents the total surface area below the radar installation height in the 3D model; CR 1,2,……,k~all =(A1∪A2∪……∪A k ) / A m ; CR k = CR 1,2,……,k~all - CR 1,2,……,(k-1)~all ; ； A k Indicates the perception range corresponding to the k-th point, A i It represents the sensing range corresponding to the i-th point. The value range of i is {1, 2, ..., k-1}, and k>2.

Images