A road scene modeling method based on real road surface data

Abstract

The present invention relates to a road scene modeling method based on real road surface data, belonging to the technical field of road modeling, and solves the problems of low efficiency of existing road scene modeling and inability to fit the real road surface conditions. It includes: collecting continuous road scene data according to the sampling frequency to obtain the attributes of each sampling point; selecting segmentation points according to the attributes of each sampling point to segment the road scene data to obtain a plurality of scene primitives; classifying the geometric features of each scene primitive according to the road surface type and average attribute value of each scene primitive to obtain a plurality of geometric primitive sets; performing multi-dimensional feature clustering on each geometric primitive set according to the multi-dimensional attribute features of each scene primitive to obtain a plurality of subdivided primitive sets; according to the received modeling requirement information, selecting several scene primitives from each subdivided scene primitive set, and connecting the selected several scene primitives to obtain a road scene model. It realizes the rapid modeling of a road scene that highly restores the real road surface.

Description

Technical Field

[0001] The present invention relates to the technical field of road modeling, and particularly to a road scene modeling method based on real road surface data. Background Art

[0002] As a crucial part of the modern transportation system, roads have different types in different scenarios, that is, the characteristic attributes of different roads are different. During the vehicle driving process, each attribute of the road affects the vehicle driving performance. During the simulation process, in order to make the simulation effect more in line with the real effect, it is necessary to consider each characteristic attribute of the road and perform modeling.

[0003] In the existing method, road segments are selected for modeling by calculating the cost of road primitives. This method requires the user to explicitly specify the characteristic parameter of each dimension of each required road primitive segment, and then based on the required characteristic parameters, match the most similar scene primitive from the primitive library.

[0004] The existing method is only applicable to the modeling of small-scale specific road scenes. Because if the road scene to be modeled is too long, the required specified characteristic parameters are huge, and it is difficult to achieve accurate specification of the attribute characteristics such as the type, shape, and resistance coefficient of a large number of road segments by the user during the actual modeling process. In addition, although each road point in the road scene model generated by this method retains the real microscopic attribute characteristics and meets the user's specified requirements, it cannot consider the coupling characteristics and distribution characteristics among multiple attributes, so it is difficult to ensure that the attributes of the real road are still similar after the recombination of a large number of scene primitives. Summary of the Invention

[0005] In view of the above analysis, the embodiments of the present invention aim to provide a road scene modeling method based on real road surface data to solve the problems that the existing road scene modeling has low efficiency and cannot be applied to large-scale road scene modeling, and cannot fit the real road surface conditions.

[0006] The embodiments of the present invention provide a road scene modeling method based on real road surface data, including the following steps:

[0007] Collect continuous road scene data according to the sampling frequency to obtain the attributes of each sampling point; select split points according to the attributes of each sampling point to split the road scene data to obtain a plurality of scene primitives;

[0008] Classify the geometric features of each scene primitive according to the road surface type and average attribute value of each scene primitive to obtain a plurality of geometric primitive sets;

[0009] Perform multi-dimensional feature clustering on each geometric primitive set according to the multi-dimensional attribute features of each scene primitive to obtain a plurality of subdivided primitive sets;

[0010] According to the received modeling requirement information, select several scene primitives from each set of subdivision scene primitives, and connect the selected scene primitives to obtain a road scene model.

[0011] Based on a further improvement of the above method, classify the geometric features of each scene primitive according to the road surface type and average attribute value of each scene primitive to obtain multiple geometric primitive sets, including:

[0012] First, divide according to the road surface type to obtain the primitive set corresponding to each road surface type;

[0013] Then, according to the average curvature and average slope of all sampling points included in each scene primitive, perform geometric feature division on the primitive set corresponding to each road surface type to obtain a curved road primitive set, a ramp primitive set, and a straight road primitive set corresponding to each road surface type, respectively serving as a geometric primitive set.

[0014] Based on a further improvement of the above method, the multi-dimensional attribute features of each scene primitive are obtained by calculating the average value and variance of the curvature, slope, rolling resistance coefficient, steering resistance coefficient, and road surface unevenness of all sampling points included in each scene primitive.

[0015] Based on a further improvement of the above method, according to the multi-dimensional attribute features of each scene primitive, use the improved K-means clustering algorithm to perform multi-dimensional feature clustering on each geometric primitive set, and cluster each geometric primitive set into multiple subdivision primitive sets.

[0016] Based on a further improvement of the above method, in the improved K-means clustering algorithm, the distance between each scene primitive in the geometric primitive set and the clustering center is calculated by the following formula:

[0017]

[0018] where Dis(γ j ,μ i ) represents the distance between the j-th scene primitive and the i-th clustering center in the current geometric primitive set; γ j (q) and μ i (q) represent the q-th dimension attributes of the j-th scene primitive and the i-th clustering center; Q represents the total dimension of the attribute features; δ q represents the distance weight of the q-th dimension attribute.

[0019] Based on a further improvement of the above method, the modeling requirement information includes: the total road length and the proportion of the road length of each road surface type.

[0020] Based on further improvements to the above method, several scene primitives are selected from each set of scene primitives for sub - scenarios, including:

[0021] According to the modeling requirement information, the average length of the scene primitives corresponding to each road surface type, and the proportion of the number of scene primitives in each geometric primitive set in each road surface type, the number of scene primitives to be selected for each geometric primitive set in each road surface type is obtained;

[0022] According to the number of scene primitives to be selected for each geometric primitive set, and in accordance with the proportion of the number of scene primitives in each sub - primitive set, the number of scene primitives to be selected for each sub - primitive set is obtained;

[0023] Scene primitives are randomly selected from the number of scene primitives to be selected for each sub - primitive set to obtain the selected scene primitives.

[0024] Based on further improvements to the above method, the selected several scene primitives are connected to obtain a road scene model, including:

[0025] The end slope of the previous scene primitive and the start slope of the adjacent next scene primitive among the selected scene primitives are calculated in sequence; according to the two slopes, and the direction vectors of the end and the start, the rotation angle of the next scene primitive is calculated;

[0026] With the first sampling point of the next scene primitive as the center, all the sampling points included in the next scene primitive are rotated according to the rotation angle, and then all the sampling points of the next scene primitive are translated so that the first sampling point of the next scene primitive coincides with the last sampling point of the previous scene primitive, completing the connection of the two selected scene primitives;

[0027] After the selected several scene primitives are connected in sequence, a road scene model is obtained.

[0028] Based on further improvements to the above method, according to the two slopes, and the direction vectors of the end and the start, the rotation angle of the next scene primitive is calculated, including:

[0029] The difference between the arctangent value of the start slope of the next scene primitive and the arctangent value of the end slope of the previous scene primitive is calculated to obtain the first angle;

[0030] The dot product of the direction vectors of the end and the start is calculated. If the dot product is negative, 180 degrees is added to the first angle to obtain the rotation angle; otherwise, the first angle is the rotation angle.

[0031] Based on further improvements to the above method, the road scene data is segmented by selecting segmentation points according to the attributes of each sampling point to obtain multiple scene primitives. The segmentation points are selected according to the road surface type, curvature, and slope of each sampling point. The road scene data is segmented according to the segmentation points to obtain multiple road segments, and road segments with a length greater than or equal to the length threshold are extracted as each scene primitive.

[0032] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0033] 1. First, the geometric features of each scene primitive are classified from the road surface type, curvature, and slope. Then, multi-dimensional feature clustering is performed on each category by simultaneously considering the coupling features and distribution features of multi-dimensional attributes, so as to better understand the characteristics of different scene primitives, capture real road features, and better model various complex road scenes.

[0034] 2. Simplify the modeling requirement information, improve the user experience, and enhance the modeling efficiency. It is applicable to the dynamic modeling of large-scale road scenes. Determine the number to be selected according to the distribution ratio of each sub-category, ensuring that all selected scene primitives are consistent with the real road scene in terms of the mutual coupling characteristics and distribution characteristics of multi-dimensional attributes, thereby improving the authenticity of the road scene model.

[0035] 3. Perform accurate rotation and translation according to the geometric features of the start and end of the scene primitive to achieve smooth transition of the scene primitive, improve the connection quality, and achieve a better dynamic modeling effect.

[0036] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the subsequent specification. Moreover, some advantages can be made obvious from the specification or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained through the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation to the present invention. Throughout the drawings, the same reference signs represent the same components.

[0038] Figure 1 It is a flowchart of a road scene modeling method based on real road surface data in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following will specifically describe the preferred embodiments of the present invention with reference to the drawings. The drawings constitute a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0040] A specific embodiment of the present invention discloses a road scene modeling method based on real road surface data, as Figure 1 shown, including the following steps:

[0041] S1. Collect continuous road scene data according to the sampling frequency, and obtain the attributes of each sampling point; select segmentation points according to the attributes of each sampling point to segment the road scene data, and obtain multiple scene primitives.

[0042] It should be noted that by arranging sensors such as cameras and radars on the vehicle, and allowing the vehicle to continuously drive on the corresponding road, the data of each sampling point is collected and stored according to the sampling frequency to obtain continuous road scene data.

[0043] Analyze the data of each sampling point to obtain the attributes of each sampling point, including: calculating the curvature, slope and road surface unevenness of each sampling point; constructing an extended Kalman filter or unscented Kalman filter model based on the vehicle dynamics model, and estimating the rolling resistance coefficient and steering resistance coefficient of each sampling point according to the motor torque on the left and right sides of the vehicle, longitudinal vehicle speed and yaw angular velocity.

[0044] Furthermore, the road surface type of each sampling point is obtained by inputting the image captured by the camera into the trained classification neural network, and the road surface type includes but is not limited to: paved road, gravel road and undulating dirt road.

[0045] In this step, segmentation points are selected from each sampling point according to the attributes of each sampling point to segment the collected continuous road scene data, and the continuous road data is segmented into scene primitive segments with as single attributes as possible.

[0046] Specifically, according to the road surface type, curvature and slope of each sampling point, segmentation points are selected in multiple ways, including:

[0047] ① Select the sampling point where the road surface type changes suddenly from the continuous sampling points as the segmentation point.

[0048] Exemplarily, if the road surface type of the first 10 sampling points is paved road and the road surface type of the 11th sampling point is gravel road, then the 11th sampling point is used as the segmentation point.

[0049] ② Divide the continuous road scene data into multiple road intervals on average according to the preset length; calculate the curvature change and slope change of the sampling points in each road interval respectively. If the curvature change exceeds the curvature change threshold, or the slope change exceeds the slope change threshold, then the first sampling point in this road interval is used as the segmentation point.

[0050] Summarize the segmentation points obtained by various methods, and use the segmentation points to segment the collected continuous road scene data to obtain multiple road segments. Considering the uncertainty of the sensor during the data collection process, there will be a certain amount of data jitter, resulting in a large redundancy in the above-obtained segmentation points, and the road segments segmented according to the segmentation points are too fragmented. Therefore, in order to improve the usability of the road segments, discard the road segments with a length less than the length threshold, and only retain the road segments with a length greater than or equal to the length threshold as the primitive elements of each scene.

[0051] S2. Classify the geometric features of each primitive element of each scene according to the road surface type and average attribute value of each primitive element of each scene to obtain multiple geometric primitive element sets.

[0052] It should be noted that the road surface types of each sampling point in each primitive element of each scene obtained in step S1 are the same. In this step, first divide according to the road surface type to obtain the primitive element set corresponding to each road surface type, that is, the primitive element set S corresponding to the road surface type t t , such as the primitive element set S1 corresponding to the paved road, the primitive element set S2 corresponding to the gravel road, and the primitive element set S3 corresponding to the undulating dirt road.

[0053] Then, according to the average curvature and average slope of all sampling points included in each primitive element of each scene, perform geometric feature division on the primitive element set corresponding to each road surface type to obtain the curved road primitive element set, the ramp primitive element set, and the straight road primitive element set corresponding to each road surface type, respectively, as a geometric primitive element set.

[0054] Specifically, take out the primitive element set corresponding to each road surface type in turn. When performing geometric feature division on the primitive element set corresponding to the current road surface type, if the average curvature of all sampling points included in the primitive element of this primitive element set is greater than the set curvature threshold, it is considered that this primitive element is a curve and add it to the curved road primitive element set corresponding to the current road surface type; if the average slope of all sampling points included in the remaining primitive elements of this primitive element set is greater than the set slope threshold, then this primitive element is a ramp and add it to the ramp primitive element set corresponding to the current road surface type; finally, the remaining primitive elements in this primitive element set are straight roads and add them to the straight road primitive element set corresponding to the current road surface type; each obtained curved road primitive element set, ramp primitive element set, and straight road primitive element set is a geometric primitive element set S t,k , where k represents the geometric feature type, that is: curve, ramp, and straight road.

[0055] Exemplarily, the primitive element set S1 corresponding to the paved road is further divided into the curved road primitive element set S 1,1 , the ramp primitive element set S 1,2 and the straight road primitive element set S 1,3 .

[0056] It should be noted that in this embodiment, the scene primitives of each road surface type are further divided into curves, slopes and straight roads, and more accurate geometric feature information of the road is extracted, which is convenient to meet the various modeling needs of users and make the modeled road scene closer to the real road surface data.

[0057] S3. Perform multi-dimensional feature clustering on each geometric primitive set respectively according to the multi-dimensional attribute features of each scene primitive, and obtain multiple subdivided primitive sets.

[0058] It should be noted that when there are a large number of scene primitives, each geometric primitive set obtained in step S2 will still contain many scene primitives. Then, when modeling the road scene, in order to accurately select the scene primitives and ensure that the attributes of the selected scene primitives after modeling are similar to those of the real road, this step further clusters and subdivides each geometric primitive set according to the multi-dimensional attribute features.

[0059] Specifically, the multi-dimensional attributes include: curvature ρ, slope α, rolling resistance coefficient f r , steering resistance coefficient f t and road surface unevenness f f . For each scene primitive, perform statistical analysis on the multi-dimensional attributes of all the sampling points it contains, and calculate the average value and variance of each attribute respectively as the multi-dimensional attribute features of the scene primitive where the average value is used to represent the coupling characteristics of each attribute in the scene primitive; the variance is used to represent the distribution characteristics of each attribute in the scene primitive.

[0060] According to the multi-dimensional attribute features of each scene primitive, use the improved K-means clustering algorithm to perform multi-dimensional feature clustering on each geometric primitive set respectively, and cluster each geometric primitive set into multiple subdivided primitive sets.

[0061] Specifically, select a geometric primitive set S t under the primitive set S t,k corresponding to the road surface type t, which contains m scene primitives [γ1, γ2, γ3,..., γ m as the input samples of the clustering algorithm. Randomly select samples from them as the initial clustering centers [μ1, μ2,..., μ n according to the number of categories n to be divided, and assign each input sample γ j to the category of the nearest clustering center according to its multi-dimensional attribute features.

[0062] It should be noted that the distance between each scene primitive in the geometric primitive set and the clustering center is calculated by the following formula:

[0063]

[0064] Among them, Dis(γ j , μ i ) represents the distance between the j-th scene primitive and the i-th clustering center in the current set of geometric primitives; γ j (q) and μ i (q) represent the q-th dimensional attributes of the j-th scene primitive and the i-th clustering center; Q represents the total number of dimensions of the attribute features; δ q represents the distance weight of the q-th dimensional attribute, which can be preset according to the actual situation, indicating the different influences of each dimensional feature in the clustering process.

[0065] When a clustering iteration is completed, that is, when all scene primitives are allocated, calculate the mean of the dimensional attribute features of all scene primitives in each category as the multi-dimensional attribute features of the new clustering center for the next clustering iteration. When the maximum number of iterations is reached, the clustering process of a set of geometric primitives is completed. The set of geometric primitives S t,k is clustered into n subsets of primitives, and the d-th subset of primitives is denoted as S t,k,d (d = 1, 2,..., n).

[0066] S4. According to the received modeling requirement information, select several scene primitives from each subset of scene primitives, and connect the selected scene primitives to obtain a road scene model.

[0067] It should be noted that the modeling requirement information input by the user includes: the total road length and the proportion of the road length of each road surface type. There is no need to let the user input the attribute feature values of each required scene primitive, which improves the user experience and the modeling efficiency, and is applicable to the dynamic modeling of large-scale road scenes.

[0068] First, according to the modeling requirement information and the average length of the scene primitives corresponding to each road surface type, the following formula is used to obtain the number of scene primitives to be selected corresponding to each road surface type:

[0069]

[0070] Among them, N t represents the number of scene primitives to be selected corresponding to the road surface type t, that is, the number of scene primitives to be selected from the set of primitives S t ; L represents the total road length; P represents the proportion of the road length of the road surface type t; represents the average length of the scene primitives corresponding to the road surface type t, that is, the average length of all scene primitives in the set of primitives S t .

[0071] Then, according to the proportion of the number of scene primitives in each geometric primitive set for each road surface type, obtain the number of scene primitives to be selected N t,k, that is, the number of straight, ramp, and curved road surfaces in each road surface type is obtained.

[0072] It should be noted that if the user has specific requirements for the number of straight, ramp, or curved road surfaces in a certain road surface type, then the specified corresponding quantity is added to the modeling requirement information. At the same time, according to the proportion of the number of scene primitives in each geometric primitive set and the specified quantity, the number of straight, ramp, and curved road surfaces in each road surface type is obtained.

[0073] Next, according to the number of scene primitives to be selected in each geometric primitive set, and in accordance with the proportion of the number of scene primitives in each sub - primitive set, the number of scene primitives to be selected N for each sub - primitive set is obtained. t,k,d ;

[0074] Finally, according to the number of scene primitives to be selected in each sub - primitive set, scene primitives are randomly selected from them to obtain the selected scene primitives. The indices of the selected scene primitives are concatenated to obtain the scene primitive index sequence.

[0075] In this embodiment, different levels of classification of scene primitives are performed through steps S2 and S3. In this step, according to the proportion of the number of each category in each layer, the number of scene primitives to be selected is obtained layer by layer, so as to obtain the number of scene primitives to be selected for each sub - primitive set, ensuring that all selected scene primitives are consistent with the real - road scene in terms of the mutual coupling characteristics and distribution characteristics of multi - dimensional attributes, and improving the authenticity of the modeling scene.

[0076] The indices of the selected scene primitives are concatenated in the order set according to the actual situation or in a random order to obtain the selected scene primitive index sequence. According to the scene primitive index sequence, adjacent subsequent scene primitives are connected in turn on the basis of the previous scene primitive. After several selected scene primitives are connected in sequence, a road scene model is obtained.

[0077] That is to say, the coordinates of the first scene primitive in the scene primitive index sequence remain unchanged, which is used as the previous scene primitive, and the adjacent subsequent scene primitive is connected to it. After connection, the newly connected scene primitive is used as the new previous scene primitive, and the adjacent subsequent scene primitive is continuously connected to it until all connections are completed.

[0078] Specifically, the connection process of the subsequent scene primitive and the previous scene primitive includes:

[0079] ① Calculate the end slope of the previous scene primitive and the start slope of the adjacent subsequent scene primitive in the selected scene primitives in sequence; according to the two slopes, as well as the direction vectors of the end and the start, calculate the rotation angle of the subsequent scene primitive.

[0080] It should be noted that, according to the preset number of sampling points Num, the last Num sampling points of the previous scene primitive are taken as the end, and the first Num sampling points of the next scene primitive are taken as the start. According to the coordinates of the sampling points included in the end and the start respectively, the end slope k is obtained by fitting a linear function. after and the start slope k before .

[0081] Calculate the difference between the arctangent value of the start slope of the next scene primitive and the arctangent value of the end slope of the previous scene primitive, and obtain the first angle θ through the following formula:

[0082] θ = arctan(k after ) - arctan(k before ) Formula (3)

[0083] where arctan(·) represents the arctangent function.

[0084] It should be noted that if the rotation angle is calculated only based on the slope, it may cause the connection direction of the two scene primitives to be opposite. Therefore, further calculate the direction vector of the end according to the coordinates of the last sampling point and the Num-th last sampling point of the previous scene primitive; calculate the direction vector of the start according to the coordinates of the Num-th sampling point and the first sampling point of the next scene primitive; then calculate the dot product of the direction vectors of the end and the start. If the dot product is negative, add 180 degrees to the first angle to obtain the rotation angle; otherwise, the first angle is the rotation angle.

[0085] ② Rotate all the sampling points included in the next scene primitive according to the rotation angle with the first sampling point of the next scene primitive as the center.

[0086] It should be noted that taking the first sampling point (x0, y0) of the next scene primitive as the center, rotate all the sampling points included in the next scene primitive in the XY two-dimensional plane, and calculate the rotated XY coordinates through the following formula:

[0087]

[0088] where (x e , y e ) and (x_rot e , y_ro e ) represent the XY coordinates of the e-th sampling point in the next scene primitive before and after rotation respectively. Since the first sampling point is the rotation center point, the coordinates remain unchanged before and after rotation.

[0089] This step makes the different scene primitives smoothly connected through rotation, improves the connection quality, and achieves a better dynamic modeling effect.

[0090] ③Translate all the sampling points of the latter scene primitive so that the first sampling point of the latter scene primitive coincides with the last sampling point of the former scene primitive, thus completing the connection of the two selected scene primitives.

[0091] It should be noted that the translation is based on the last sampling point of the former scene primitive and is performed according to the relative positions of the sampling points in the latter scene primitive with respect to its first sampling point. The formula is as follows:

[0092]

[0093] where, (x_re e , y_re e , z_re e ) represents the coordinates of the e-th sampling point in the latter scene primitive after translation; z e represents the Z coordinate of the e-th sampling point in the latter scene primitive before translation, z0 represents the Z coordinate of the first sampling point in the latter scene primitive before translation, and (x_before end , y_before end , z_before end ) represents the XYZ coordinates of the last sampling point in the former scene primitive.

[0094] From the above connection process, it can be seen that only the coordinates of each sampling point are changed during the connection of scene primitives, and each sampling point still maintains its original multi-dimensional attributes. When connecting the selected scene primitives, the multi-dimensional attributes of each sampling point included therein can be seamlessly spliced without any modification, and the complete attribute information of the modeled road scene can be obtained.

[0095] Compared with the prior art, a road scene modeling method based on real road surface data provided by this embodiment first classifies the geometric features of each scene primitive according to road surface type, curvature, and slope, and then performs multi-dimensional feature clustering on each category by simultaneously considering the coupling features and distribution features of multi-dimensional attributes, so as to better understand the characteristics of different scene primitives, capture real road features, and better model various complex road scenes; simplify the modeling requirement information, improve the user experience, and increase the modeling efficiency, which is applicable to the dynamic modeling of large-scale road scenes; determine the number of selections to be made according to the distribution ratio of each sub-category, ensuring that all selected scene primitives are consistent with the real road scene in terms of the mutual coupling characteristics and distribution characteristics of multi-dimensional attributes, and improving the authenticity of the road scene model; perform accurate rotation and translation according to the geometric features of the start and end of the scene primitive to achieve a smooth transition of the scene primitive, improve the connection quality, and achieve a better dynamic modeling effect.

[0096] Those skilled in the art can understand that all or part of the processes of implementing the methods of the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a disk, an optical disc, a read-only memory, or a random access memory, etc.

[0097] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A road scene modeling method based on real road surface data, characterized in that: The following steps are involved: Collect continuous road scene data according to the sampling frequency to obtain the attributes of each sampling point; select segmentation points according to the attributes of each sampling point to segment the road scene data to obtain multiple scene primitives; Classify the geometric features of each scene primitive according to the road surface type and average attribute value of each scene primitive to obtain multiple geometric primitive sets; According to the multidimensional attribute characteristics of each scene primitive, each geometric primitive set is clustered by multidimensional characteristics to obtain multiple subdivided primitive sets; According to the received modeling requirement information, a plurality of scene primitives are selected from each subdivided scene primitive set, and the selected plurality of scene primitives are connected to obtain a road scene model, including: sequentially calculating the end slope of a previous scene primitive and the start slope of an adjacent next scene primitive in the selected scene primitives; According to the two slopes and the direction vectors of the end and the start, the rotation angle of the next scene primitive is calculated; with the first sampling point of the next scene primitive as the center, all the sampling points contained in the next scene primitive are rotated according to the rotation angle, and then all the sampling points of the next scene primitive are translated so that the first sampling point of the next scene primitive coincides with the last sampling point of the previous scene primitive, thereby completing the connection of the two selected scene primitives; after the selected several scene primitives are connected in sequence, a road scene model is obtained.

2. The road scene modeling method based on real road surface data according to claim 1 is characterized in that: The geometric feature classification of each scene primitive is performed according to the road surface type and the average attribute value of each scene primitive to obtain multiple geometric primitive sets, including: First, divide the road surface according to its type to obtain the primitive set corresponding to each road surface type; Then, according to the average curvature and average slope of all sampling points contained in each scene primitive, the primitive set corresponding to each road surface type is divided by geometric features to obtain the curve primitive set, ramp primitive set and straight primitive set corresponding to each road surface type, which are respectively regarded as a geometric primitive set.

3. The road scene modeling method based on real road surface data according to claim 2 is characterized in that: The multi-dimensional attribute characteristics of each scene primitive are obtained by calculating the average value and variance of the curvature, slope, rolling resistance coefficient, steering resistance coefficient and road surface roughness of all sampling points contained in each scene primitive.

4. The road scene modeling method based on real road surface data according to claim 3 is characterized in that: According to the multidimensional attribute characteristics of each scene primitive, an improved K-means clustering algorithm is used to perform multidimensional feature clustering on each geometric primitive set, and each geometric primitive set is clustered into multiple subdivided primitive sets.

5. The road scene modeling method based on real road surface data according to claim 4 is characterized in that: In the improved K-means clustering algorithm, the distance between each scene primitive in the geometric primitive set and the cluster center is calculated by the following formula: ， in, Indicates the first The scene primitives and The distance between cluster centers; and Indicates scene primitives and The cluster center Dimensional attributes; Represents the total dimension of attribute features; Indicates The distance weight of the dimension attribute.

6. The road scene modeling method based on real road surface data according to claim 2, characterized in that: The modeling requirement information includes: the total length of the road and the proportion of the road length of each road surface type.

7. The road scene modeling method based on real road surface data according to claim 6, characterized in that: The selecting a plurality of scene primitives from each subdivided scene primitive set includes: According to the modeling requirement information, the average length of the scene primitives corresponding to each road surface type and the proportion of the number of scene primitives in each geometric primitive set in each road surface type, the number of geometric primitive sets to be selected in each road surface type is obtained; According to the number of to-be-selected geometric primitive sets of each set, and according to the proportion of the number of scene primitives in each subdivided primitive set, the number of to-be-selected primitive sets of each set is obtained; A scene primitive is randomly selected from each subdivided primitive set according to the number of primitives to be selected, thereby obtaining a selected scene primitive.

8. The road scene modeling method based on real road surface data according to claim 1, characterized in that: The step of calculating the rotation angle of the next scene primitive according to the two slopes and the direction vectors of the end and the start includes: Calculate the difference between the arc tangent value of the starting slope of the next scene primitive and the arc tangent value of the ending slope of the previous scene primitive to obtain a first angle; The dot product of the direction vectors of the end and the start is calculated. If the dot product is a negative number, 180 degrees is added to the first angle to obtain the rotation angle; otherwise, the first angle is the rotation angle.

9. The road scene modeling method based on real road surface data according to claim 1, characterized in that: The method of selecting segmentation points according to the attributes of each sampling point to segment the road scene data to obtain multiple scene primitives is to select segmentation points according to the road surface type, curvature and slope of each sampling point, segment the road scene data according to the segmentation points to obtain multiple road segments, and extract road segments with a length greater than or equal to a length threshold as each scene primitive.

Images