Map updating method and device
By vectorizing the point cloud data, matching points with the same name to update the high-precision map, the problem of difficulty in matching point cloud data and vector data is solved, and the efficiency and accuracy of map updates are improved.
Patent Information
- Application Number
- CN201910272509.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-04-04
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2039-04-04
AI Technical Summary
The efficiency of high-precision map update is inefficient, mainly due to the difficulty of extracting features from point cloud data, which leads to the difficulty of matching points with the same name, requiring a lot of manual intervention, and the existing technology cannot directly match point cloud data and vector data.
By vectorizing the collected point cloud data, point cloud vector data is generated, and multiple points of the same name are determined based on the original map data and the point cloud vector data to be matched to the original map data to be updated efficiently.
It improves the efficiency of high-precision map updates, reduces manual intervention, realizes direct matching and updates between vector data, and improves the accuracy and efficiency of map updates.
Smart Images

Figure CN111782739B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to computer technology, and in particular to a map updating method and device. Background Art
[0002] With the continuous advancement and development of vehicle autonomous driving technology, high-precision maps with detailed map elements have also been widely used as one of the core technologies of autonomous driving. In order to ensure the safety of autonomous driving, it is usually necessary to continuously update high-precision map data.
[0003] At present, when updating high-precision maps, point cloud data is usually obtained by collecting data on site, and features are extracted from the point cloud data for similarity matching to generate homonymous points. Secondly, the new map data and the old map data are aligned and differentiated according to the homonymous points to obtain the difference between the new and old map data, and then the difference is updated.
[0004] However, extracting features from point cloud data to generate homonymous points is very difficult, which leads to inefficient updating of high-precision maps. Summary of the invention
[0005] The embodiments of the present invention provide a map updating method and device to overcome the problem of low updating efficiency of high-precision maps.
[0006] In a first aspect, an embodiment of the present invention provides a map updating method, comprising:
[0007] Performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data;
[0008] Obtaining a plurality of points with the same name according to the original map data and the point cloud vector data, wherein the original map data is vector data;
[0009] The original map data is updated according to the points with the same name to obtain updated map data.
[0010] In a second aspect, an embodiment of the present invention provides a map updating device, including:
[0011] A processing module, used for performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data;
[0012] A homonymous point generation module, used to obtain a plurality of homonymous points according to original map data and the point cloud vector data, wherein the original map data is vector data;
[0013] The updating module is used to update the original map data according to the points with the same name to obtain updated map data.
[0014] In a third aspect, an embodiment of the present invention provides a (device subject), including:
[0015] Memory, used to store programs;
[0016] A processor is used to execute the program stored in the memory. When the program is executed, the processor is used to execute the method described in the first aspect and any one of the various possible designs of the first aspect.
[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute the method described in the first aspect and any one of the various possible designs of the first aspect.
[0018] The embodiment of the present invention provides a map updating method and device, the method comprising: performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data; obtaining multiple points of the same name according to the original map data and the point cloud vector data, wherein the original map data is vector data; updating the original map data according to the points of the same name to obtain updated map data. By processing the collected point cloud data into point cloud vector data, and then obtaining multiple points of the same name according to the original map data and the point cloud vector data, wherein the original map data and the point cloud vector data are both vector data, multiple points of the same name can be determined efficiently and conveniently, and then the original map data is updated according to the points of the same name, thereby effectively improving the efficiency of map updating. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0020] Figure 1 Scenario diagram of the map updating method provided by the present invention Figure 1 ;
[0021] Figure 2 A flowchart of a map updating method provided by a first embodiment of the present invention;
[0022] Figure 3 A flowchart of a map updating method provided by a second embodiment of the present invention;
[0023] Figure 4 A flowchart of a map updating method provided by a third embodiment of the present invention;
[0024] Figure 5 Scenario diagram of the map updating method provided by the present invention Figure 2 ;
[0025] Figure 6 A schematic diagram of the structure of a map updating device provided by the present invention;
[0026] Figure 7 A schematic diagram of the hardware structure of the cloud platform provided by the present invention. DETAILED DESCRIPTION
[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0028] With the continuous advancement and development of electronic map navigation technology, vehicle autonomous driving technology has also been widely studied. Vehicle autonomous driving technology must rely on high-precision maps. Compared with ordinary navigation electronic maps, high-precision maps suitable for autonomous driving have higher accuracy, more detailed map elements, and richer attributes. They are crucial to the positioning, navigation, control and safety of autonomous driving unmanned vehicles.
[0029] To ensure that high-precision maps can reflect the latest road conditions, the timeliness of high-precision maps is very important. Therefore, high-precision maps need to be continuously updated to meet actual needs. When updating maps, high-precision map update collection vehicles are usually used to collect data again on a certain section of road or in a certain area. For example, point cloud data can be obtained through lidar collection. Optionally, the data collected by a binocular camera can be converted into point cloud data. Secondly, the point cloud data is vectorized to form geometric elements such as lane edges, curbs, guardrails, etc. to replace the original vector data to achieve updates.
[0030] Furthermore, the newly collected vector data is aligned with the original vector data, and then the new and old data are differentiated to generate the vector difference part of the new and old data. Only the vector difference part is updated, which can improve the efficiency of the map update operation. Specifically, a large number of points with the same name need to be found before the alignment operation.
[0031] At present, the existing technology usually extracts features from point cloud data when searching for homonymous points, and then performs similarity matching between the extracted features and the original vector data to obtain homonymous points. However, the operation of searching for homonymous points is very difficult due to the following factors:
[0032] 1. Update operations often occur many years after the last operation, and the natural environment has changed (such as repainting road markings, re-repairing guardrails, etc.), making it very difficult to extract the same features in point cloud data and original vector data
[0033] 2. Point cloud data may be acquired by different means, such as lidar and binocular vision. There are differences in the technical means of extracting features (for example, lidar can determine features by emission intensity, while binocular vision has no intensity information but can extract features through image segmentation), which makes it difficult to extract features.
[0034] 3. It is difficult to automatically extract features from point cloud data. The number of points with the same name that can be automatically extracted is small. Usually, after automatically extracting features, some manual points need to be added to supplement the points with the same name.
[0035] 4. The original map data is vector data, while the newly collected data is point cloud data. The two do not have the same features and cannot be directly matched. It is necessary to find the point cloud data corresponding to the original map data in order to find the points with the same name.
[0036] The above factors will lead to low update efficiency of high-precision maps. Based on this problem, the embodiment of the present invention provides a map update method and device. Figure 1 First, the application scenarios of the present invention are briefly introduced.
[0037] Figure 1 Scenario diagram of the map updating method provided by the present invention Figure 1 .like Figure 1 As shown, the original map data 101 includes road information of a certain area, such as traffic lights, curbs, road markings, guardrails and other information. The specific road information included in the original map data is determined according to the actual natural scene, and is not limited here.
[0038] Furthermore, for example, if a no-entry sign is added at a certain intersection in the current area in a natural scene, the original map data needs to be updated so that the autonomous driving unmanned vehicle can obtain new road information in a timely manner. The updated map is shown in 102, and a no-entry sign 103 is added to the map.
[0039] Those skilled in the art will understand that Figure 1The map in the figure is only for illustration and is not a high-precision map in actual application. Specifically, the map data of the high-precision map is vector data. Figure 2 The map updating method provided by the present invention is described in detail with reference to specific embodiments. Figure 2 This is a flowchart of a map updating method provided by the first embodiment of the present invention.
[0040] like Figure 2 As shown, the method includes:
[0041] S201. Perform vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data.
[0042] Specifically, first, point cloud data is collected by a high-precision map collection vehicle, where a point cloud refers to a massive point set of target surface characteristics. Optionally, for example, a point cloud is obtained by the laser measurement principle of a lidar, and its point cloud data includes three-dimensional coordinates and laser reflection intensity.
[0043] Optionally, the point cloud obtained by the photogrammetry principle of binocular vision, the point cloud data includes three-dimensional coordinates and color information. Those skilled in the art can understand that the specific method of collecting data can be selected according to needs, and there is no special limitation on the point cloud data here.
[0044] Furthermore, the collected point cloud data is vectorized to obtain point cloud vector data, wherein the specific implementation method of the vectorization processing may be, for example, vectorizing the point cloud data using an automatic vectorization algorithm. This embodiment does not limit the automatic vectorization algorithm. The specific implementation method of the vectorization processing may also be, for example, manually vectorizing the point cloud data in sequence. The specific implementation method may be selected according to demand.
[0045] In this embodiment, vector data is data corresponding to a data organization method that uses points, lines, surfaces and their combinations in Euclidean geometry to represent the spatial distribution of geographic entities. Vector data can represent the spatial position of geographic entities by recording coordinates.
[0046] S202. Obtain at least one point with the same name according to original map data and point cloud vector data, wherein the original map data is vector data.
[0047] Furthermore, both the original map data and the point cloud vector data are vector data, so the original map data and the point cloud vector data can be directly matched to obtain at least one point with the same name.
[0048] Among them, the same-name point refers to the image point of the same point on the ground in different images. In this embodiment, the same-name point can be, for example, the point in the upper left corner of the same speed limit sign at the same position in the original map data and the newly collected point cloud vector data, that is, the point corresponding to the same object in the original map data and the newly collected point cloud vector data.
[0049] Specifically, the implementation method for matching multiple points with the same name can be, for example, determining the distance between each feature point in the original map data and each vector point in the newly collected point cloud vector data based on the position of the feature point in the original map data and the position of the vector point in the newly collected point cloud vector data, wherein the position of the feature point in the original map data and the position of the vector point in the newly collected point cloud vector data correspond to the same position of the same object.
[0050] Furthermore, for each feature point in the original map data, a vector point in the newly collected point cloud vector data having the shortest distance thereto is determined as the corresponding point of the same name, thereby obtaining a plurality of points of the same name.
[0051] Optionally, for example, artificial puncture points may be manually selected to obtain a plurality of points with the same name. The specific implementation method of generating points with the same name is not limited in this embodiment.
[0052] S203: Update the original map data according to the points with the same name to obtain updated map data.
[0053] Furthermore, the original map data is updated according to the points with the same name. The specific implementation method may be, for example, to determine the difference between the current original map data and the point cloud vector data according to the positions of the points with the same name in the original map data and in the point cloud vector data, and then update the difference to obtain updated map data.
[0054] Optionally, the implementation method may also be, for example, to compare the original map data and the point cloud vector data according to the semantic information of the object indicated by the current point of the same name, thereby updating the map, etc., which is not limited in this embodiment.
[0055] The map updating method provided by the embodiment of the present invention includes: performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data; obtaining multiple points of the same name based on the original map data and the point cloud vector data, wherein the original map data is vector data; updating the original map data based on the points of the same name to obtain updated map data. By processing the collected point cloud data into point cloud vector data, and then obtaining multiple points of the same name based on the original map data and the point cloud vector data, wherein both the original map data and the point cloud vector data are vector data, multiple points of the same name can be determined efficiently and conveniently, and then the original map data is updated based on the points of the same name, thereby effectively improving the efficiency of map updating.
[0056] Based on the above embodiments, the specific implementation method of generating the same-name points is further described in detail below in conjunction with the specific embodiments. Figure 3 To explain, Figure 3 This is a flowchart of a map updating method provided by the second embodiment of the present invention.
[0057] like Figure 3 As shown, the method includes:
[0058] S301. Classify the point cloud data according to a preset classification type to obtain the classified point cloud data, where the classified point cloud data contains semantic information.
[0059] Specifically, the classification types of the preset classification include at least one of the following: point data class, line data class and surface data class. Optionally, the line data class can be further subdivided into curb class, guardrail class, etc., and the surface data class can be further subdivided into traffic sign class, wall class, etc. This embodiment does not limit the specific classification type of the preset classification, which can be set according to needs.
[0060] Furthermore, the point cloud data is classified and processed, wherein the classification processing can be implemented, for example, according to the automatic vectorization algorithm introduced above, and can also be implemented, for example, according to a separate classification algorithm, to obtain the classified point cloud data, and divide the point cloud data into at least point cloud data of a point data class, point cloud data of a line data class, and point cloud data of a surface data class.
[0061] In this embodiment, the classified point cloud data also contains semantic information, wherein the semantic information is used to indicate the specific object content corresponding to the point cloud data. Specifically, the semantic information may be, for example, lane boundaries, curbs, guardrails, walls, etc., thereby enabling the high-precision map data to have rich actual scene information.
[0062] S302: performing vectorization processing on the classified point cloud data to obtain point cloud vector data.
[0063] The vectorization process has been introduced in detail in the above embodiments and will not be described again here.
[0064] S303: Obtain the classification type to which the point cloud vector data belongs, wherein the classification type includes at least one of the following: a point data class, a line data class, and a surface data class.
[0065] Specifically, vector data is data composed of points, lines, and surfaces. This embodiment generates points of the same name in different ways for point cloud vector data of different classification types, so the classification type of the point cloud vector data is first obtained.
[0066] S304. Acquire feature point data corresponding to the point cloud vector data in the original map data according to the classification type to which the point cloud vector data belongs; wherein the feature point data and the point cloud vector data correspond to the same object.
[0067] Furthermore, in the process of generating points with the same name, it is necessary to match the points with the same name based on points as the basic unit. Therefore, firstly, the feature point data corresponding to the point cloud vector data is obtained in the original map data. The feature point data may correspond to multiple feature points, for example, wherein the feature point data and the point cloud vector data correspond to the same object. For example, there is vector data of a target traffic light in the point cloud vector data. Secondly, the vector data corresponding to the target traffic light is obtained in the original map data. The vector data corresponding to the target traffic light in the original map data is the feature point data.
[0068] In this embodiment, for point cloud vector data of different classification types, the implementation methods for determining the feature point data are different, and the acquisition of the feature point data is specifically introduced below.
[0069] Optionally, if the classification type to which the point cloud vector data belongs is a point data class, at least one point data corresponding to the point cloud vector data in the original map data is determined as feature point data.
[0070] Specifically, the point cloud vector data of the point data class is a specific vector point, which can be directly matched to generate points with the same name. Furthermore, due to the influence of systematic errors and random errors, the vector points may not be completely aligned. There may be multiple feature points for the point cloud vector data of a point data class. Therefore, at least one point data corresponding to the point cloud vector data is directly determined as the feature point data.
[0071] Further optionally, if the classification to which the point cloud vector data belongs is a linear data class, a shape point of at least one linear data corresponding to the point cloud vector data in the original map data is determined as feature point data.
[0072] Specifically, the vector data of the linear data class cannot directly match the points of the same name, so the vector data of the linear data needs to be discretized to obtain the shape points of the linear data. In this embodiment, first, at least one linear data corresponding to the point cloud vector data in the original map data is obtained, and then the obtained linear data is discretized to obtain the shape points of the linear data, wherein the shape points of the linear data are the feature points of the linear data, and the feature point data is determined based on the obtained shape points.
[0073] Further optionally, if the point cloud vector data belongs to a category of planar data, a centroid and a corner point of at least one planar data corresponding to the point cloud vector data in the original map data are determined as feature point data.
[0074] Specifically, the vector data of the surface data class cannot be matched directly either, because the center of mass and corner points of the same object are certain. Therefore, the center of mass and corner points of the vector data of the surface data class are first obtained, and then the center of mass and corner points of at least one surface data corresponding to the point cloud vector data in the original map data are determined as feature point data.
[0075] Those skilled in the art will appreciate that the above-described method of determining feature points is not the only implementation method, and the method of determining feature points for different classification types can be selected according to actual needs. Optionally, for example, feature points can also be determined manually, etc. This embodiment does not limit this.
[0076] S305: Obtain a first distance between each vector point corresponding to the point cloud vector data and a feature point corresponding to the feature point data.
[0077] In this embodiment, the point cloud vector data includes vector data composed of points, lines, and surfaces. First, the vector data of the linear data class and the surface data class are processed into the corresponding vector data of the point data class, and the point cloud vector data is processed in units of vector points.
[0078] Secondly, for each vector point corresponding to the point cloud vector data, the first distance between it and the feature point corresponding to the feature point data is obtained. For example, one vector point may correspond to multiple feature points, so the first distance of a vector point may be multiple distance values.
[0079] Optionally, when the point cloud vector data is of a linear data type, the first distance may be obtained by projecting at least one shape point of the linear data corresponding to the point cloud vector data in the original map data onto a vector point of the point cloud vector data.
[0080] S306: For each vector point, determine a minimum first distance corresponding to the vector point.
[0081] S307: If the minimum first distance is within a preset distance range, the feature point corresponding to the minimum first distance is used as a point of the same name as the vector point.
[0082] Furthermore, for each vector point in the point cloud vector data, the minimum first distance in the first distance of the vector point is determined, and then it is determined whether the minimum first distance is within a preset distance range. If so, the feature point corresponding to the minimum first distance is used as a homonymous point of the vector point, wherein the preset distance range is used to limit the distance between the selected homonymous points to not be too far, and its specific value can be selected according to needs.
[0083] In this embodiment, for each vector point, the feature point closest to the vector point is selected as the to-be-selected homonymous point, and then the to-be-selected homonymous points are further screened by setting a preset distance range, thereby avoiding excessive errors in the selected homonymous points and improving the reliability of the determination of homonymous points.
[0084] S308: Update the original map data according to the points with the same name to obtain updated map data.
[0085] Specifically, the implementation method of S308 is similar to that of S203, and will not be repeated here.
[0086] The map updating method provided by the embodiment of the present invention includes: classifying point cloud data according to a classification type of a preset classification to obtain point cloud data after classification, wherein the point cloud data after classification contains semantic information. Vectorizing the point cloud data after classification to obtain point cloud vector data. Obtaining the classification type to which the point cloud vector data belongs, wherein the classification type includes at least one of the following: a point data class, a line data class, and a surface data class. According to the classification type to which the point cloud vector data belongs, obtaining feature point data corresponding to the point cloud vector data in the original map data; wherein the feature point data and the point cloud vector data correspond to the same object. Obtaining the first distance between each vector point corresponding to the point cloud vector data and the feature point corresponding to the feature point data. For each vector point, determining the minimum first distance corresponding to the vector point. If the minimum first distance is within the preset distance range, the feature point corresponding to the minimum first distance is used as a synonym of the vector point. The original map data is updated according to the synonym to obtain updated map data. By first determining the feature point data according to the classification type of the point cloud data, and then determining the homonymous points according to the distance between the vector point and the feature point, the accuracy of determining the homonymous points is guaranteed. A large number of homonymous points can be generated according to the linear data class, avoiding the need to manually supplement the homonymous points, and ensuring the efficiency of determining the homonymous points.
[0087] Based on the above embodiments, the specific implementation method of updating the map according to the same-name points is described in detail below in conjunction with the specific embodiments. Figure 4 and Figure 5 To explain, Figure 4 A flowchart of a map updating method provided in a third embodiment of the present invention. Figure 5 Scenario diagram of the map updating method provided by the present invention Figure 2 .
[0088] like Figure 4 As shown, the method includes:
[0089] S401. Perform vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data.
[0090] S402. Obtain a plurality of points with the same name according to original map data and point cloud vector data, wherein the original map data is vector data.
[0091] Among them, the implementation method of S401 and S402 is similar to that of S201 and S202, and will not be repeated here.
[0092] S403 . For each feature point in the original map data, select a preset number of points with the same name around each feature point as control points.
[0093] Among them, the above-mentioned embodiment obtains multiple feature point data in the original map data when matching the same-name points. Furthermore, this embodiment aligns each feature point and the same-name points around it to achieve relative smoothness between the original map data and the newly collected map data, so that further differential processing can be performed while ensuring data accuracy.
[0094] Specifically, for each feature point in the original map data, a preset number of points with the same name are selected around each feature point, where the specific value of the preset number can be set according to needs and is not limited here. Secondly, the periphery of each feature point can be, for example, a circular range with the shape point as the center and a preset distance as the radius, or a rectangular range with the feature point as the center, etc. This embodiment does not impose any restrictions on this.
[0095] Furthermore, the selected points with the same name are used as control points, wherein the control points are used to indicate changes in position information of feature points during registration.
[0096] S404: For each control point, obtain a second distance between the control point and the remaining control points.
[0097] Furthermore, a preset number of points with the same name are selected as control points, wherein the selected points with the same name may be points with higher precision, that is, the correspondence between the original map data and the newly collected point cloud data of the points with the same name is relatively accurate.
[0098] Optionally, the selected homonymous points may also be homonymous points with poor accuracy, that is, there is a certain deviation in the correspondence between the original map data and the newly collected point cloud data. In this embodiment, by assigning weights to each control point, the homonymous points with different accuracies have different degrees of influence in the alignment process.
[0099] Specifically, for each control point, the second distance between the control point and the remaining control points is obtained. For example, there are currently three control points A, B, and C. First, for control point A, the second distances between AB and AC can be obtained. For control point B, the second distances between BA and BC can be obtained. The remaining implementation methods are similar and will not be repeated here.
[0100] S405. Determine the weight of each control point according to a preset weight calculation model and each second distance, wherein the weight is used to indicate the influence range of the control point, and the preset weight calculation model is used to characterize that the second distance and the weight of the control point are negatively correlated.
[0101] In this embodiment, the preset weight calculation model is shown in the following formula 1:
[0102]
[0103] Among them, i is used to represent each control point, Wi(R) is used to represent the weight of the control point, and R is the weight indicator parameter of the control point. Specifically, R is calculated as shown in the following formula 2:
[0104]
[0105] Among them, R iN is the distance between the current control point x and the farthest control point among the k remaining control points, dist(x,x i ) is the second distance between the current control point x and the k remaining control points, corresponding to k values, then R is the ratio of the second distance between the current control point x and the k remaining control points to the farthest distance.
[0106] In this embodiment, the preset weight calculation model is used to characterize that the second distance and the weight of the control point are negatively correlated. Specifically, the weight of each control point is calculated according to Formula 1 and Formula 2, wherein the weight is used to indicate the influence range of the control point. Specifically, when R>1, it indicates that the distance between the current control point and a certain control point is greater than the farthest distance, indicating that the selection deviation of the control point is very large, and the weight of the control point is set to 0 so that it has no influence.
[0107] Furthermore, when 0≤R≤1, the weight of the control point is set to 1-R 2, the closer the distance to the shape point is, the larger the influence range in the registration process is, and the farther the distance to the shape point is, the smaller the influence range in the registration process is, that is, the second distance is negatively correlated with the weight of the control point, thereby eliminating some data unevenness caused by inaccurate selection of points with the same name.
[0108] S406 , determining the position of the feature point after updating according to the weight of each control point and the thin plate spline deformation model, wherein the thin plate spline deformation model is a model corresponding to the thin plate spline function.
[0109] In this embodiment, registration is achieved based on a thin plate spline deformation model, wherein the thin plate spline deformation model is a model corresponding to a thin plate spline function. The thin plate spline function is an interpolation function, which is usually interpolated based on two dimensions, and is therefore often used in image registration. Specifically, the thin plate spline function can be used to deform each feature point to the corresponding position of each control point, while achieving deformation of the entire space, thereby achieving registration.
[0110] Let's first combine Figure 5 The thin plate spline deformation model in this embodiment is briefly described as follows: Figure 5 As shown, the cross 501 is used to indicate each feature point. For example, a point with the same name is selected around each feature point 501 as a control point, that is, Figure 5 The object indicated by the middle circle 502 is the control point. Next, each feature point is deformed to the position of the control point according to the thin plate spline deformation model, wherein the space is also deformed, thereby achieving registration between the original map data and the newly collected map data.
[0111] Optional, for example, when multiple points with the same name are selected as control points, Figure 5 The object indicated by the middle circle 502 is the deformation target position obtained by synthesizing the weights of each control point, so as to determine the position of the feature point after updating according to the weights of each control point and the thin plate spline deformation model.
[0112] The thin plate spline deformation model is specifically introduced below, wherein the expression of the thin plate spline deformation model is shown in the following formula 3:
[0113]
[0114] Among them, x and y are the two-dimensional coordinates of the feature points, i represents each control point, represents the distance between control point i and feature point, w i and a, a x 、a yare the parameters to be determined, f(x, y) are the coordinate numerical components after deformation (x value component, y value component, z value component). In this embodiment, there are N control points, and N control points can form N expressions of f(x, y). The above formula three indicates the coordinate changes of the feature points during the alignment process.
[0115] Furthermore, the above formula 3 is expressed in a matrix as shown in the following formula 4:
[0116]
[0117] Among them, the matrix K is an N×N matrix, and its expression is shown in Formula 5:
[0118]
[0119] in, Compared with the formula 3 Partially corresponding, among which σ is the regularization parameter.
[0120] Specifically, since the traditional thin plate spline deformation model requires all control points to participate in the transformation during deformation, the amount of calculation is large and the local transformation is inconsistent, which will lead to poor alignment effect. In order to avoid this problem and the road elements of high-precision maps generally change smoothly, this embodiment introduces a regularization parameter in the thin plate spline deformation model, wherein the regularization parameter is negatively correlated with the bending energy of the thin plate spline, and the bending ability is used to indicate the degree of spatial deformation. Specifically, the larger the regularization parameter, the smaller the bending energy, and the flatter the fitted surface.
[0121] Furthermore, the matrix M is an N-dimensional column matrix, and its expression is shown in Formula 6, which is the same as in Formula 1. Partially corresponding:
[0122]
[0123] Furthermore, A is a 3D column matrix, which is expressed as shown in Formula 7, which is the same as a+a in Formula 1. x x+a y a, a in the y part x 、a y Partially corresponding:
[0124]
[0125] Furthermore, the matrix P is an N×3 matrix, and its expression is shown in Formula 8, which is the same as a+a in Formula 1. x x+a y The 1, x, and y in the y part correspond to:
[0126]
[0127] Furthermore, the matrix N is an N-dimensional column matrix, and its expression is shown in Formula 9, which corresponds to the f(x,y) part in Formula 1:
[0128]
[0129] Optionally, according to Formula 2, P T M=0, where this formula is a boundary constraint condition, ensuring that the rigid transformation is 0 at infinity.
[0130] Specifically, in formula 3, w i and a, a x 、a y is the parameter to be determined, and the corresponding matrix M and matrix A are the matrices to be determined. In this embodiment, the coordinates of the points that have been deformed and the positions after deformation are used to solve the thin plate spline deformation model, where K, P, and V are all known matrices. Then, the matrix M and the matrix A can be solved by formula 4. Correspondingly, w in formula 3 can be obtained. i and a, a x 、a y , thus obtaining a complete thin plate spline deformation model.
[0131] Optionally, this embodiment adjusts the bending energy by setting a regularization parameter σ. Specifically, a convergence threshold of the bending energy is first set. If the bending energy is greater than the set convergence threshold, the regularization parameter is adjusted, and iteration is performed again until the bending energy is less than the convergence threshold. The iteration ends. The bending energy can be obtained by Formula 10:
[0132] I f =M T KM Formula 10
[0133] Among them, I f This is the bending energy.
[0134] Furthermore, in this embodiment, a weighted average method is used to take a weighted average of all thin plate spline functions passing through any point P (x, y) to determine the corresponding components.
[0135] The calculation of the X-value component mapping function f(x, y) is shown in the following formula 11:
[0136]
[0137] The calculation of the Y value component mapping function g(x, y) is shown in the following formula 11:
[0138]
[0139] The calculation of the Z value component mapping function h(x, y) is shown in the following formula 13:
[0140]
[0141] The purpose of the thin plate spline deformation model based on local weighting is to make the control point closest to the point P(x, y) have the greatest influence, and the farther away it is, the smaller the influence. i ,y i ) is greater than R iN , the impact is not taken into account.
[0142] S407 , performing differential processing on the position of each feature point after being updated and the position of each feature point before being updated to obtain data to be updated.
[0143] Furthermore, after alignment, the position of each feature point after update is obtained, and then differential processing is performed based on the position of each feature point after update and the position of each feature point before update to obtain the data to be updated, wherein the data to be updated is the part that needs to be added, deleted, or modified. In this embodiment, the differential processing can, for example, be to compare the coordinate positions of each feature point, or can also be to manually compare the data, etc., which is not limited here.
[0144] S408. Update the data to be updated according to the point cloud data to obtain updated map data.
[0145] Further, the part of the data to be updated is updated, for example, for the part that needs to be added, the part is added to the original map data according to the point cloud data, for example, for the part that needs to be modified, the part is modified according to the point cloud data, so as to obtain an updated map.
[0146] The map updating method provided by the embodiment of the present invention includes: performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data. According to the original map data and the point cloud vector data, at least one point of the same name is obtained, wherein the original map data is vector data. For each feature point in the original map data, a preset number of points of the same name around each feature point are selected as control points. For each control point, a second distance between the control point and the remaining control points is obtained. According to a preset weight calculation model and each second distance, the weight of each control point is determined, wherein the weight is used to indicate the influence range of the control point, and the preset weight calculation model is used to characterize that the second distance and the weight of the control point are negatively correlated. According to the weight of each control point and a thin plate spline deformation model, the position of the feature point after updating is determined, wherein the thin plate spline deformation model is a model corresponding to the thin plate spline function. Differential processing is performed based on the position of each feature point after updating and the position of each feature point before updating to obtain data to be updated. The data to be updated is updated based on the point cloud data to obtain updated map data. By determining the control points based on the points with the same name and assigning different weights to each control point, the reduction in registration accuracy caused by the selection of control points can be eliminated. Then, the position of each feature point after updating is differentially processed with the position of each feature point before updating, and then the data to be updated is updated to obtain the updated map data, thereby avoiding the need to process all the original map data, reducing the workload of updating the map, and improving the efficiency of updating.
[0147] Figure 6 This is a schematic diagram of the structure of the map updating device provided by the present invention. Figure 6 As shown, the device 60 includes: a processing module 601, a same-name point generating module 602 and an updating module 603.
[0148] The processing module 601 is used to perform vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data;
[0149] A homonymous point generating module 602 is used to obtain a plurality of homonymous points according to original map data and point cloud vector data, wherein the original map data is vector data;
[0150] The updating module 603 is used to update the original map data according to the points with the same name to obtain updated map data.
[0151] Optionally, the same-name point generating module 602 is specifically used for:
[0152] Acquire feature point data corresponding to the point cloud vector data in the original map data; wherein the feature point data and the point cloud vector data correspond to the same object;
[0153] Obtaining a first distance between each vector point corresponding to the point cloud vector data and a feature point corresponding to the feature point data;
[0154] According to each first distance, a plurality of points with the same name are obtained.
[0155] Optionally, the same-name point generating module 602 is specifically used for:
[0156] For each vector point, determining the minimum first distance corresponding to the vector point;
[0157] If the minimum first distance is within the preset distance range, the feature point corresponding to the minimum first distance is used as the same-name point of the vector point.
[0158] Optionally, the same-name point generating module 602 is specifically used for:
[0159] Acquire the classification type to which the point cloud vector data belongs, wherein the classification type includes at least one of the following: point data class, line data class, and surface data class;
[0160] According to the classification type to which the point cloud vector data belongs, feature point data corresponding to the point cloud vector data is obtained in the original map data.
[0161] Optionally, the same-name point generating module 602 is specifically used for:
[0162] If the classification type to which the point cloud vector data belongs is the point data class, determining at least one point data corresponding to the point cloud vector data in the original map data as feature point data;
[0163] If the classification to which the point cloud vector data belongs is the linear data class, determining at least one shape point of linear data corresponding to the point cloud vector data in the original map data as feature point data;
[0164] If the point cloud vector data belongs to the category of planar data, the centroid and corner point of at least one planar data corresponding to the point cloud vector data in the original map data are determined as feature point data.
[0165] Optionally, the processing module 601 is specifically configured to:
[0166] Classify the point cloud data according to the preset classification type to obtain the classified point cloud data, where the classified point cloud data contains semantic information;
[0167] The classified point cloud data is vectorized to obtain point cloud vector data.
[0168] Optionally, the updating module 603 is specifically used for:
[0169] For each feature point in the original map data, a preset number of points with the same name around each feature point are selected as control points;
[0170] Determine the weight of each control point according to the second distance between each control point;
[0171] Determine the position of the feature point after updating according to the weight of each control point and the thin plate spline deformation model, wherein the thin plate spline deformation model is a model corresponding to the thin plate spline function;
[0172] According to the updated position of each feature point, updated map data is obtained.
[0173] Optionally, the updating module 603 is specifically used for:
[0174] For each control point, obtain the second distance between the control point and the remaining control points;
[0175] The weight of each control point is determined according to a preset weight calculation model and each second distance, wherein the weight is used to indicate the influence range of the control point, and the preset weight calculation model is used to characterize that the second distance and the weight of the control point are negatively correlated.
[0176] Optionally, the updating module 603 is specifically used for:
[0177] Perform differential processing on the position of each feature point after updating and the position of each feature point before updating to obtain the data to be updated;
[0178] The data to be updated is updated according to the point cloud data to obtain updated map data.
[0179] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.
[0180] Figure 7 The hardware structure diagram of the cloud platform provided by the present invention is as follows: Figure 7 As shown, the cloud platform 70 of this embodiment includes: a processor 701 and a memory 702;
[0181] Memory 702, used to store computer-executable instructions;
[0182] The processor 701 is used to execute the computer-executable instructions stored in the memory to implement the various steps performed by the map updating method in the above embodiment. For details, please refer to the relevant description in the above method embodiment.
[0183] Optionally, the memory 702 may be independent or integrated with the processor 701 .
[0184] When the memory 702 is independently configured, the cloud platform further includes a bus 703 for connecting the memory 702 and the processor 701 .
[0185] An embodiment of the present invention further provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the map updating method executed by the cloud platform is implemented.
[0186] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.
[0187] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to perform some steps of the method described in each embodiment of the present application.
[0188] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.
[0189] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.
[0190] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.
[0191] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.
[0192] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.
[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A map updating method, characterized in that: include: Performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data; Obtaining a plurality of points with the same name according to the original map data and the point cloud vector data, wherein the original map data is vector data; updating the original map data according to the points with the same name to obtain updated map data; The updating of the original map data according to the points of the same name to obtain updated map data includes: For each feature point in the original map data, a preset number of points with the same name around each feature point are selected as control points; Determining a weight of each of the control points according to a second distance between each of the control points; Determining the position of the feature point after updating according to the weight of each control point and the thin plate spline deformation model, wherein the thin plate spline deformation model is a model corresponding to the thin plate spline function; The updated map data is obtained according to the updated position of each feature point.
2. The method according to claim 1, characterized in that The method of obtaining a plurality of points with the same name based on the original map data and the point cloud vector data includes: Acquire feature point data corresponding to the point cloud vector data in the original map data; wherein the feature point data and the point cloud vector data correspond to the same object; Acquire a first distance between each vector point in the point cloud vector data and a feature point in the feature point data; A plurality of points with the same name are obtained according to each of the first distances.
3. The method according to claim 2, characterized in that The obtaining of a plurality of points of the same name according to each of the first distances comprises: For each vector point, determining a minimum first distance corresponding to the vector point; If the minimum first distance is within a preset distance range, the feature point corresponding to the minimum first distance is used as the same-name point of the vector point.
4. The method according to claim 2, characterized in that: The step of acquiring feature point data corresponding to the point cloud vector data in the original map data includes: Acquire the classification type to which the point cloud vector data belongs, wherein the classification type includes at least one of the following: a point data class, a line data class, and a surface data class; According to the classification type to which the point cloud vector data belongs, feature point data corresponding to the point cloud vector data is acquired in the original map data.
5. The method according to claim 4, characterized in that The acquiring, in the original map data, feature point data corresponding to the point cloud vector data according to the classification type to which the point cloud vector data belongs, comprises: If the classification type to which the point cloud vector data belongs is a point data class, determining at least one point data corresponding to the point cloud vector data in the original map data as feature point data; If the point cloud vector data belongs to a linear data category, determining at least one shape point of linear data corresponding to the point cloud vector data in the original map data as feature point data; If the point cloud vector data is classified as a planar data class, a centroid and a corner point of at least one planar data corresponding to the point cloud vector data in the original map data are determined as feature point data.
6. The method according to claim 1, characterized in that The vectorization processing of the collected point cloud data to obtain point cloud vector data includes: Classify the point cloud data according to the preset classification type to obtain the classified point cloud data, wherein the classified point cloud data contains semantic information; Vectorization is performed on the classified point cloud data to obtain point cloud vector data.
7. The method according to claim 1, characterized in that Determining the weight of each control point according to the distance between each control point includes: For each control point, obtaining a second distance between the control point and the remaining control points; The weight of each control point is determined according to a preset weight calculation model and each second distance, wherein the weight is used to indicate the influence range of the control point, and the preset weight calculation model is used to characterize that the second distance and the weight of the control point are negatively correlated.
8. The method according to claim 1, characterized in that The step of obtaining updated map data according to the updated positions of the feature points comprises: Performing differential processing on the position of each feature point after being updated and the position of each feature point before being updated to obtain data to be updated; The data to be updated is updated according to the point cloud data to obtain updated map data.
9. A map updating device, characterized in that: include: A processing module, used for performing vectorization processing on the collected point cloud data to obtain point cloud vector data, wherein the point cloud vector data is vector data corresponding to the point cloud data; A homonymous point generation module, used to obtain a plurality of homonymous points according to original map data and the point cloud vector data, wherein the original map data is vector data; An updating module, used for updating the original map data according to the same-name points to obtain updated map data; The updating module is specifically used to select a preset number of points with the same name around each feature point in the original map data as control points; determine the weight of each control point according to the second distance between each control point; determine the position of the feature point after updating according to the weight of each control point and a thin plate spline deformation model, wherein the thin plate spline deformation model is a model corresponding to the thin plate spline function; and obtain updated map data according to the updated position of each feature point.
10. A cloud platform, characterized in that: include: Memory, used to store programs; A processor, configured to execute the program stored in the memory; when the program is executed, the processor is configured to execute the method according to any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that: The method comprises instructions which, when executed on a computer, cause the computer to perform the method as claimed in any one of claims 1 to 8.
Citation Information
Patent Citations
Threshold-free automatic robust matching method for multi-source residence surface elements
CN104574519A
Map difference extraction method, device and terminal
CN105787873A
Airborne laser point cloud data and vehicle laser point cloud data fusion method
CN106056563A
Road vector data integrating and updating method
CN106203491A
Method and device for acquiring homonymy points of images
CN106529607A
