Terrain data generation method, device and system and data acquisition equipment

Abstract

The present application discloses a terrain data generation method, device, system and data acquisition equipment, which belong to the field of map making. The point cloud is projected into a pre-constructed grid network, and the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network are obtained. The target grid probability of the second-level grid is obtained based on the distance from the second-level grid to the ground plane, and the current frame semantic data of the first-level grid is obtained based on the target grid probability of the second-level grid. Finally, the current frame semantic data and the historical semantic data of the first-level grid are fused to obtain the terrain data of the target area. The present application scheme processes the point cloud through grids of different scales in the grid network to ensure that the semantic data of each frame of the final first-level grid is accurate, so that the terrain data obtained after multi-frame fusion is accurate, and since only the point cloud in the grid needs to be processed, no operations such as point cloud segmentation are required, and the terrain data of the target area can be quickly obtained.

Description

Technical Field

[0001] The present application relates to the field of map making technology, and in particular, to a terrain data generation method, device and system and data acquisition equipment. Background Art

[0002] The current production of high-precision maps relies on terrain data (DEM, digital elevation model), which is required to draw key information such as roads and intersections. There are some problems with terrain data production: the terrain data is noisy, the terrain accuracy is not high enough, which affects the mapping process; the entire production pipeline process: including data uploading, data analysis, point cloud mapping, point cloud segmentation, terrain production, etc., takes too long. Especially for scenarios that need to quickly support the passage of unmanned vehicles, long-term map updates are not allowed, which is a pain point for current map updates. Summary of the invention

[0003] In order to overcome the deficiencies of the related art, the present application provides a terrain data generation method, device and system and data acquisition equipment to solve the problem that the existing terrain data generation has inaccurate terrain and is time-consuming.

[0004] The technical solution adopted by this application to solve its technical problem is: In a first aspect, a method for generating terrain data is provided, comprising: Acquire point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and acquire the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids; Based on the distance from the secondary grid to the ground plane, updating the initial grid probability of the secondary grid to obtain the target grid probability of the secondary grid; Based on the target grid probability of the secondary grid, the initial grid probability of the primary grid in the secondary grid is updated to obtain the current frame semantic data of the primary grid; The current frame semantic data and historical semantic data of the first-level grid are fused to obtain the terrain data of the target area.

[0005] Furthermore, it also includes: Based on the initial grid probability of the secondary grid, determining the first and second-level grids of the obstacle type, and clustering the first and second-level grids to obtain obstacle targets; Calculate the preset parameters in the obstacle target, and determine the type of the obstacle target based on the preset parameters. The types of the obstacle target include dynamic obstacles and static obstacles. The preset parameters include: probability density, mean number of hits, size and area of ​​the obstacle target. Among them, the probability density is the ratio of the sum of the fusion probability values ​​of the target first-level grid to the total number of the target first-level grids, and the mean number of hits is the ratio of the total number of the target first-level grids hit in the preset number of frames to the total number of the target first-level grids. The target first-level grid is the first-level grid included in the obstacle target in the current frame. The target first-level grid being hit indicates that the target first-level grid is an obstacle in one frame of data. The fusion probability value = initial grid probability + * historical grid probability, the historical grid probability is obtained according to the historical semantic data, The number of frames where the target first-level grid type remains unchanged; If the type of the obstacle target is determined to be a dynamic obstacle based on the preset parameters, the point cloud within the first-level grid of the target is deleted.

[0006] Further, judging the type of the obstacle target based on the preset parameters includes: When the preset parameters simultaneously meet the following preset conditions, the type of the obstacle target is determined to be a dynamic obstacle; when the preset parameters cannot simultaneously meet the following preset conditions, the type of the obstacle target is determined to be a static obstacle, and the preset conditions are: The interval where the probability density is located is an obstacle; The average number of hits is less than a preset average; The size and area of ​​the obstacle target are within a preset size and area range.

[0007] Further, the updating of the initial grid probability of the secondary grid based on the distance from the secondary grid to the ground plane to obtain the target grid probability of the secondary grid includes: If the distance from the center point of the secondary grid to the ground plane is greater than the first preset distance, the target grid probability of the secondary grid is updated to be a first preset coefficient, where the first preset coefficient is a probability value when the type of the secondary grid is an obstacle; If the distance from the center point of the secondary grid to the ground plane is less than or equal to the first preset distance, the target grid probability of the secondary grid is updated to a second preset coefficient, and the second preset coefficient is the probability value when the type of the secondary grid is ground.

[0008] Furthermore, it also includes: Perform plane fitting on the secondary grid to obtain a fitting plane, and calculate the normal vector angle and elevation difference between any two fitting planes; Clustering the secondary grids whose normal vector angle is less than the angle threshold and whose elevation difference is less than the elevation difference threshold; Calculate the regional mean probability of the clustered area, determine the type of the cluster area according to the regional mean probability, and use the cluster area whose type is ground as the second secondary grid, the regional mean probability is the ratio of the sum of the preset grid probabilities of all secondary grids in the area to the number of all secondary grids in the area, the preset grid probability is one of the initial grid probability or the historical grid probability, and the historical grid probability is obtained according to the historical semantic data; A ground plane is obtained by performing plane fitting on all second diode grids in a three-level grid, wherein the three-level grid is composed of m adjacent second-level grids.

[0009] Furthermore, the target grid probability based on the secondary grid updates the initial grid probability of the primary grid in the secondary grid to obtain the current frame semantic data of the primary grid, including: Based on the target grid probability of the dipole grid, the third secondary grid of the obstacle type is determined; Calculate the target distance from the center point of the first-level grid within the third second-level grid to the ground plane; Based on the target distance, the initial grid probability of the first-level grid is updated to obtain the target grid probability of the first-level grid; The current frame semantic data of the first-level grid is obtained according to the target grid probability of the first-level grid.

[0010] Further, the updating of the initial grid probability of the first-level grid based on the target distance to obtain the target grid probability of the first-level grid includes: If the target distance is greater than the second preset distance, the target grid probability of the first-level grid is updated to be a third preset coefficient, and the third preset coefficient is a probability value when the type of the first-level grid is an obstacle; If the target distance is less than or equal to the second preset distance, the target grid probability of the first-level grid is updated to a fourth preset coefficient, and the fourth preset coefficient is a probability value when the type of the first-level grid is ground.

[0011] In a second aspect, a terrain data generating device is provided, comprising: A point cloud data projection module is used to obtain point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and obtain the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids; A secondary grid updating module, used to update the initial grid probability of the secondary grid based on the distance from the secondary grid to the ground plane, and obtain the target grid probability of the secondary grid; A primary grid updating module, used for updating the initial grid probability of the primary grid within the secondary grid based on the target grid probability of the secondary grid, and obtaining the current frame semantic data of the primary grid; The semantic data fusion module is used to fuse the current frame semantic data and historical semantic data of the first-level grid to obtain the terrain data of the target area.

[0012] In a third aspect, a computer-readable storage medium is provided, on which a computer program or instruction is stored, and when the computer program or instruction is executed by a processor, the steps of the terrain data generation method provided by the technical solution of the first aspect are implemented.

[0013] In a fourth aspect, a terrain data generation system is provided, comprising: at least one processor and at least one memory; The memory stores executable instructions of the processor; The processor is configured to execute the terrain data generating method provided by the technical solution of the first aspect.

[0014] In a fifth aspect, a data acquisition device is provided, which applies the terrain data generation method provided by the technical solution of the first aspect.

[0015] Beneficial effects: The technical solution of the present application provides a terrain data generation method, device, system and data acquisition equipment. After obtaining the point cloud data of the target area, the point cloud of the current frame is projected into a pre-constructed grid network, and the initial semantic data and historical semantic data of the current frame of the first-level grid and the second-level grid in the grid network are obtained. The initial grid probability of the second-level grid is updated based on the distance from the second-level grid to the ground plane to obtain the target grid probability of the second-level grid, and the initial grid probability of the first-level grid is updated based on the target grid probability of the second-level grid to obtain the current frame semantic data of the first-level grid. Finally, the current frame semantic data and historical semantic data of the first-level grid are fused to obtain the terrain data of the target area. The solution of the present application processes the point cloud through grids of different scales in the grid network to ensure that the semantic data of each frame of the final first-level grid is accurate, so that the terrain data obtained after multi-frame fusion is accurate, and since only the point cloud in the grid needs to be processed, no operations such as point cloud segmentation are required, and the terrain data of the target area can be quickly obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 is a flow chart of a method for generating terrain data provided by an embodiment of the present application; Figure 2 is a flow chart of another method for generating terrain data provided by an embodiment of the present application; Figure 3 is a schematic diagram of a dynamic obstacle recognition result provided by an embodiment of the present application; Figure 4 It is a multi-frame dense terrain schematic diagram provided in an embodiment of the present application; Figure 5 It is a schematic diagram of the structure of a terrain data generating device provided in an embodiment of the present application; Figure 6 It is a schematic diagram of the structure of a terrain data generation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application is described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other implementation methods obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0019] The existing terrain generation method is that the data acquisition device collects point cloud data, uploads it to the cloud through the network, and then performs data analysis, point cloud mapping, point cloud segmentation and terrain production in the cloud, which takes too long and cannot meet the requirements for scenes that need to quickly generate terrain data.

[0020] To solve the above problems, refer to Figure 1 , the embodiment of the present application provides a method for generating terrain data, comprising: S11: Acquire point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and acquire the current frame initial semantic data and historical semantic data of the primary grid and the secondary grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the secondary grid is composed of n adjacent primary grids, where n is a positive integer; It can be understood that the target area is an area where terrain data needs to be generated.

[0021] In addition, the secondary grid is composed of n adjacent primary grids. For example, the primary grid is an area of ​​0.2*0.2, and the secondary grid is a grid of 2*2, so the secondary grid is composed of 10*10 primary grids. In order to avoid repeated calculations, in practice, the initial semantic data of the current frame of the secondary grid is obtained by averaging the initial semantic data of the current frame of all the primary grids that constitute the secondary grid. The historical semantic data of the secondary grid is obtained by averaging the historical semantic data of all the primary grids that constitute the secondary grid.

[0022] In one embodiment, after acquiring the point cloud data of the current frame, semantic point cloud data is obtained by segmentation through a pre-trained deep learning model or a traditional detection model, and the semantic point cloud data is projected onto a grid network to obtain initial semantic data of the current frame of the first-level grid. Exemplarily, the semantic point cloud data includes the semantic probability of each point, and the initial grid probability of the current frame of the first-level grid is the average of the semantic probabilities of all points in the first-level grid.

[0023] In another embodiment, after acquiring the point cloud data of the current frame, the point cloud data is directly projected onto the grid network, and then a pre-trained deep learning model or a pre-trained detection model is used to segment and obtain the initial semantic data of the current frame of the first-level grid.

[0024] It should be noted that the initial grid probability, target grid probability, historical grid probability, etc. in this application are all numerical values, and the grid type is obtained through the corresponding interval. For example, a probability less than 0 is the ground, and a probability greater than or equal to 0 is an obstacle.

[0025] It can be understood that the first-level grid and the second-level grid are grids of different scales in the grid network. The scale of the first-level grid is small, so it can accurately represent the terrain of the target area. However, since it is small, although the initial semantic data gives the initial grid probability, in fact, the initial grid probability cannot accurately indicate whether it represents the ground or an obstacle, and there may be a large error. Therefore, a larger-scale second-level grid is used, and according to its initial grid probability, it can be preliminarily judged whether the second-level grid belongs to the ground or an obstacle. Of course, there will still be a certain error in the preliminary judgment, so this embodiment further updates the initial grid probability of the second-level grid.

[0026] S12: Based on the distance from the secondary grid to the ground plane, the initial grid probability of the secondary grid is updated to obtain the target grid probability of the secondary grid; wherein the ground plane is a plane representing the ground obtained by fitting.

[0027] Exemplarily, if the distance from the center point of the secondary grid to the ground plane is greater than a first preset distance, the target grid probability of the secondary grid is updated to a first preset coefficient, where the first preset coefficient is a probability value when the type of the secondary grid is an obstacle; If the distance from the center point of the secondary grid to the ground plane is less than or equal to the first preset distance, the target grid probability of the secondary grid is updated to a second preset coefficient, and the second preset coefficient is the probability value when the type of the secondary grid is ground.

[0028] That is, when the distance from the center point of the secondary grid to the ground plane is greater than the first preset distance, it means that the distance between the secondary grid and the ground is far, and its type is determined to be an obstacle. When it is less than or equal to the first preset distance, it means that the distance between the secondary grid and the ground is close, and its type is determined to be ground. The type of the secondary grid is more accurately determined by the fitted ground plane, and its initial grid probability is updated based on the type.

[0029] The steps for obtaining the ground plane are as follows: Perform plane fitting on the secondary grid to obtain a fitting plane, and calculate the normal vector angle and elevation difference between any two fitting planes; cluster the secondary grids whose normal vector angle is less than the angle threshold and the elevation difference is less than the elevation difference threshold; determine that the secondary grids belong to the same class when the vector angle is less than the angle threshold and the elevation difference is less than the elevation difference threshold.

[0030] Calculate the regional mean probability of the clustered area, determine the type of the cluster area according to the regional mean probability, and use the cluster area whose type is ground as the second secondary grid, wherein the regional mean probability is the ratio of the sum of the preset grid probabilities of all secondary grids in the area to the number of all secondary grids in the area, and the preset grid probability is one of the initial grid probability or the historical grid probability, and the historical grid probability is obtained according to the historical semantic data; that is, in the ground plane fitting process of this embodiment, only the data of the current frame can be used, or the data of the historical frame can be used. Even in some embodiments, the ground plane obtained after fitting multiple frames of data is used as a fixed ground plane, and when needed, the ground plane is directly called. That is, the first a frames at the beginning of the acquisition need to fit the ground plane, and the ground plane obtained according to the data of the first a frames is directly used after the a frame, wherein a is a preset positive integer.

[0031] The ground plane is obtained by plane fitting all the second-level grids in the third-level grid, and the third-level grid is composed of m adjacent second-level grids. Because in practice, although the type of the second-level grid is ground, there may be a certain difference in the normal vectors and elevations of different second-level grids, and it is impossible to use a plane fitted by a second-level grid as the ground plane. Therefore, fitting is chosen in a third-level grid with a larger scale, and in order to remove obstacle interference during fitting, only the second-level grid of the ground type is plane fitted.

[0032] S13: based on the target grid probability of the secondary grid, updating the initial grid probability of the primary grid in the secondary grid, and obtaining the current frame semantic data of the primary grid; Preferably, based on the target grid probability of the dipole grid, the third and second level grids of the obstacle type are determined; the target distance from the center point of the first level grid within the third and second level grids to the ground plane is calculated; based on the target distance, the initial grid probability of the first level grid is updated to obtain the target grid probability of the first level grid; and according to the target grid probability of the first level grid, the current frame semantic data of the first level grid is obtained.

[0033] Wherein, updating the initial grid probability of the first-level grid based on the target distance to obtain the target grid probability of the first-level grid includes: If the target distance is greater than the second preset distance, the target grid probability of the first-level grid is updated to be a third preset coefficient, and the third preset coefficient is a probability value when the type of the first-level grid is an obstacle; If the target distance is less than or equal to the second preset distance, the target grid probability of the first-level grid is updated to a fourth preset coefficient, and the fourth preset coefficient is a probability value when the type of the first-level grid is ground.

[0034] That is, in the embodiment of the present application, the third secondary grid whose current type is an obstacle is determined according to the target grid probability of the secondary grid. Although the secondary grid type is an obstacle at this time, there may be a part connected to the ground inside it, which is judged as an obstacle. Therefore, it is necessary to distinguish the data in the third and second grids more accurately, so the distance from the first grid in the third and second grids to the ground plane is calculated. According to the distance, its type is determined to be an obstacle or the ground, so that the probability of obtaining the first grid is more accurate.

[0035] In addition, for the second level grid whose type is ground, the type of the first level grid inside it can be directly judged as ground, because even if there are some first level grids far away from the ground plane, since they are surrounded by ground, they are actually either convex or concave on the ground, or some small obstacles, which will not affect the driving of the vehicle. This processing can greatly reduce the amount of data processing and speed up the processing while ensuring that the map can meet the needs of vehicles.

[0036] Of course, in some embodiments, in order to obtain a more accurate map, the initial grid probabilities of the first-level grids in the second-level grid may also be updated based on the distance to the ground plane.

[0037] In the embodiment of the present application, the semantic data includes elevation in addition to grid probability, and the elevation only needs to be obtained based on the coordinates of all point clouds in the grid.

[0038] S14: The current frame semantic data and the historical semantic data of the first-level grid are integrated to obtain the terrain data of the target area.

[0039] Exemplarily, the first-level grid probability=the target grid probability+the historical grid probability. The historical grid probability is obtained according to the historical semantic data.

[0040] Level 1 grid elevation = current grid elevation * current height coefficient + historical grid elevation * historical height coefficient. The current grid elevation is the grid elevation obtained in the current frame, and the historical grid elevation is the grid elevation obtained from historical semantic data. The current height coefficient and historical height coefficient are set according to actual needs, for example, both are 0.5.

[0041] Then, DEM (digital elevation model) data is generated based on the fused raster data, because DEM data is essentially a kind of raster data with a grid size. DEM data of the corresponding size is obtained by scaling the first-level grid.

[0042] The terrain data generation method provided in the embodiment of the present application processes the point cloud through grids of different scales in the grid network to ensure the accuracy of semantic data of each frame of the final first-level grid, so that the terrain data obtained after fusion of multiple frames is accurate. Moreover, since only the point cloud in the grid needs to be processed, no operations such as point cloud segmentation are required, and the terrain data of the target area can be quickly obtained.

[0043] However, the terrain data obtained in the above embodiment and the terrain data obtained by traditional point cloud segmentation have the same problem, that is, the dynamic obstacle recognition accuracy is low, resulting in a large noise in the terrain data. Figure 2 As shown, the embodiment of the present application provides another method for generating terrain data, including: S21: Acquire point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and acquire the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids.

[0044] S22: Obtain obstacle targets based on the initial grid probability of the secondary grid, and delete point clouds of obstacle targets whose type is dynamic obstacles.

[0045] Preferably, based on the initial grid probability of the secondary grid, the first secondary grid of obstacle type is determined, and the first secondary grid is clustered to obtain obstacle targets; exemplary clustering methods include but are not limited to: area generation clustering and distance clustering.

[0046] Calculate the preset parameters in the obstacle target, and determine the type of the obstacle target based on the preset parameters. The types of the obstacle target include dynamic obstacles and static obstacles. The preset parameters include: probability density, mean number of hits, size and area of ​​the obstacle target. Among them, the probability density is the ratio of the sum of the fusion probability values ​​of the target first-level grid to the total number of the target first-level grids, and the mean number of hits is the ratio of the total number of the target first-level grids hit in the preset number of frames to the total number of the target first-level grids. The target first-level grid is the first-level grid included in the obstacle target in the current frame. The target first-level grid being hit indicates that the target first-level grid is an obstacle in one frame of data. The fusion probability value = initial grid probability + * historical grid probability, the historical grid probability is obtained according to the historical semantic data, is the number of frames in which the target first-level grid type remains unchanged; if the obstacle target type is determined to be a dynamic obstacle based on the preset parameters, the point cloud in the target first-level grid is deleted. The preset forgetting rate is designed to make the grid type unchanged for multiple consecutive frames tend to an unknown state, and the value is less than 1.

[0047] It should be noted that a single frame of data cannot determine whether an obstacle is static or dynamic, so multiple frames of data are needed to make a judgment. When the obstacle is a dynamic obstacle, the current frame grid probability of a certain target first-level grid will indicate an obstacle type, while the grid probability of the previous frame indicates the type of the ground. If it is directly fused, its accurate probability cannot be obtained (the probability obtained by directly adding may still be the ground). Therefore, in order to prevent the grid probability of the first grid of the target that does not change for a long time from affecting the judgment, a forgetting rate is set. Since the forgetting rate is less than 1, when the number of frames with a continuous unchanged type (i.e. ), the larger it is, the closer it is to 0, and the smaller the impact on the fusion probability value. When the type of the previous frame is ground and the type of the next frame becomes an obstacle, i is reset to 1.

[0048] Wherein, judging the type of the obstacle target based on the preset parameters includes: When the preset parameters simultaneously satisfy the following preset conditions, the type of the obstacle target is determined to be a dynamic obstacle; when the preset parameters cannot simultaneously satisfy the following preset conditions, the type of the obstacle target is determined to be a static obstacle, and the preset conditions are: The interval where the probability density is located is an obstacle; The average number of hits is less than a preset average; The size and area of ​​the obstacle target are within a preset size and area range.

[0049] It is understandable that when the mean number of hits is less than the preset mean, it means that in the preset number of frames, there are multiple frames of data where the target first-level grid is not an obstacle, while in theory, the target first-level grid of a static obstacle is an obstacle in each frame of data. The size and area of ​​the obstacle target are determined because both the probability density and the mean number of hits require multiple frames of data to be judged. Therefore, when there are only one or a few frames of data, there is a possibility that a static obstacle may be mistaken for a dynamic obstacle. Therefore, the approximate range of the size and area of ​​the obstacle target is set. When using the map, only the behavior of dynamic obstacles of these sizes and areas is concerned.

[0050] In addition, after deleting the point cloud of the obstacle target whose type is dynamic obstacle, the subsequent processing can reduce the amount of calculation.

[0051] S23: Based on the distance from the secondary grid to the ground plane, the initial grid probability of the secondary grid is updated to obtain the target grid probability of the secondary grid.

[0052] S24: Based on the target grid probability of the secondary grid, the initial grid probability of the primary grid in the secondary grid is updated to obtain the current frame semantic data of the primary grid.

[0053] S25: The current frame semantic data and the historical semantic data of the first-level grid are integrated to obtain the terrain data of the target area.

[0054] It is understandable that steps S21, S23, S24 and S25 are Figure 1 The steps of the illustrated embodiment are the same and will not be described again.

[0055] like Figure 3 As shown in the figure, a mining area collection task is given, where the red box where the purple point is located is the recognition result of the dynamic vehicle. The closed area surrounded by the orange line is the recognition result of the static obstacle.

[0056] Figure 4 Schematic diagram of dense terrain obtained from multi-frame data at a certain time, where the red dense grid is the ground and the orange periphery is the static retaining wall area.

[0057] The terrain data generation method provided in the embodiment of the present application can not only quickly and accurately obtain the terrain data of the target area, but also effectively distinguish between dynamic obstacles and static obstacles, thereby further improving the accuracy of the terrain data.

[0058] Based on the same inventive concept, Figure 5 As shown, the present application provides a terrain data generating device 50, comprising: The point cloud data projection module 51 is used to obtain the point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and obtain the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids; A secondary grid updating module 52, configured to update the initial grid probability of the secondary grid based on the distance from the secondary grid to the ground plane, and obtain a target grid probability of the secondary grid; The first-level grid updating module 53 is used to update the initial grid probability of the first-level grid in the second-level grid based on the target grid probability of the second-level grid, and obtain the current frame semantic data of the first-level grid; The semantic data fusion module 54 is used to fuse the current frame semantic data and the historical semantic data of the first-level grid to obtain the terrain data of the target area.

[0059] The terrain data generating device provided in the embodiment of the present application, by applying the terrain data generating method provided in any of the above embodiments, can quickly and accurately obtain the terrain data of the target area.

[0060] Based on the same inventive concept, the present application provides a computer-readable storage medium having a computer program or instruction stored thereon, which, when executed by a processor, implements the steps of the terrain data generating method provided in any of the above embodiments.

[0061] Based on the same inventive concept, Figure 6 As shown, the present application provides a terrain data generation system 60, comprising: at least one processor 61 and at least one memory 62; The memory 62 stores executable instructions of the processor 61; The processor 61 is configured to execute the terrain data generating method provided in any of the above embodiments.

[0062] Based on the same inventive concept, the present application provides a data acquisition device, applying the terrain data generation method provided by any of the above embodiments.

[0063] The data collection equipment is a collection vehicle or a drone. The collection vehicle can be an unmanned collection vehicle or a manned collection vehicle. The collection vehicle can be a vehicle specially set up for data collection, or it can be a vehicle (such as a mining truck, excavator, command vehicle light) that performs other tasks (such as excavation, loading, soil discharge, etc.) in the target area.

[0064] The data acquisition device provided in the embodiment of the present application, after acquiring the point cloud data of the target area, projects the point cloud of the current frame into a pre-constructed grid network, and acquires the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, updates the initial grid probability of the second-level grid based on the distance from the second-level grid to the ground plane, obtains the target grid probability of the second-level grid, and updates the initial grid probability of the first-level grid based on the target grid probability of the second-level grid to obtain the current frame semantic data of the first-level grid. Finally, the current frame semantic data and historical semantic data of the first-level grid are fused to obtain the terrain data of the target area. The solution of the present application processes the point cloud through grids of different scales in the grid network to ensure that each frame of semantic data of the final first-level grid is accurate, so that the terrain data obtained after multi-frame fusion is accurate. Compared with the traditional processing method, it can be processed inside the acquisition device without uploading to the cloud, and since only the point cloud in the grid needs to be processed, no operations such as point cloud segmentation are required, and the terrain data of the target area can be quickly obtained.

[0065] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0066] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0067] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0068] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0069] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0070] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0071] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0072] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0073] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for generating terrain data, characterized in that: include: Acquire point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and acquire the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids; Based on the distance from the secondary grid to the ground plane, updating the initial grid probability of the secondary grid to obtain the target grid probability of the secondary grid; Based on the target grid probability of the secondary grid, the initial grid probability of the primary grid in the secondary grid is updated to obtain the current frame semantic data of the primary grid; The current frame semantic data and historical semantic data of the first-level grid are fused to obtain the terrain data of the target area.

2. The method according to claim 1, characterized in that Also includes: Based on the initial grid probability of the secondary grid, determining the first and second-level grids of the obstacle type, and clustering the first and second-level grids to obtain obstacle targets; Calculating preset parameters in the obstacle target, and determining the type of the obstacle target based on the preset parameters, where the type of the obstacle target includes a dynamic obstacle and a static obstacle; The preset parameters include: probability density, mean number of hits, size and area of ​​obstacle target; wherein, probability density is the ratio of the sum of fusion probability values ​​of target first-level grids to the total number of target first-level grids, and mean number of hits is the ratio of the total number of target first-level grids hit in a preset number of frames to the total number of target first-level grids; the target first-level grid is the first-level grid included in the obstacle target in the current frame, and the target first-level grid being hit indicates that the target first-level grid is an obstacle in a frame of data, and fusion probability value = initial grid probability + * historical grid probability, the historical grid probability is obtained according to the historical semantic data, The number of frames where the target first-level grid type remains unchanged; If the type of the obstacle target is determined to be a dynamic obstacle based on the preset parameters, the point cloud within the first-level grid of the target is deleted.

3. The method according to claim 2, characterized in that The determining the type of the obstacle target based on the preset parameters includes: When the preset parameters simultaneously satisfy the following preset conditions, the type of the obstacle target is determined to be a dynamic obstacle; when the preset parameters cannot simultaneously satisfy the following preset conditions, the type of the obstacle target is determined to be a static obstacle, and the preset conditions are: The interval where the probability density is located is an obstacle; The average number of hits is less than a preset average; The size and area of ​​the obstacle target are within a preset size and area range.

4. The method according to claim 1, characterized in that: The updating of the initial grid probability of the secondary grid based on the distance from the secondary grid to the ground plane to obtain the target grid probability of the secondary grid includes: If the distance from the center point of the secondary grid to the ground plane is greater than the first preset distance, the target grid probability of the secondary grid is updated to be a first preset coefficient, where the first preset coefficient is a probability value when the type of the secondary grid is an obstacle; If the distance from the center point of the secondary grid to the ground plane is less than or equal to the first preset distance, the target grid probability of the secondary grid is updated to a second preset coefficient, and the second preset coefficient is the probability value when the type of the secondary grid is ground.

5. The method according to claim 1, characterized in that Also includes: Perform plane fitting on the secondary grid to obtain a fitting plane, and calculate the normal vector angle and elevation difference between any two fitting planes; Clustering the secondary grids whose normal vector angle is less than the angle threshold and whose elevation difference is less than the elevation difference threshold; Calculate the regional mean probability of the clustered area, determine the type of the cluster area according to the regional mean probability, and use the cluster area whose type is ground as the second secondary grid, the regional mean probability is the ratio of the sum of the preset grid probabilities of all secondary grids in the area to the number of all secondary grids in the area, the preset grid probability is one of the initial grid probability or the historical grid probability, and the historical grid probability is obtained according to the historical semantic data; A ground plane is obtained by performing plane fitting on all second diode grids in a three-level grid, wherein the three-level grid is composed of m adjacent second-level grids.

6. The method according to claim 1, characterized in that The method of updating the initial grid probability of the first-level grid in the second-level grid based on the target grid probability of the second-level grid to obtain the current frame semantic data of the first-level grid includes: Based on the target grid probability of the dipole grid, the third secondary grid of the obstacle type is determined; Calculate the target distance from the center point of the first-level grid within the third second-level grid to the ground plane; Based on the target distance, the initial grid probability of the first-level grid is updated to obtain the target grid probability of the first-level grid; The current frame semantic data of the first-level grid is obtained according to the target grid probability of the first-level grid.

7. The method according to claim 1, characterized in that The updating of the initial grid probability of the first-level grid based on the target distance to obtain the target grid probability of the first-level grid includes: If the target distance is greater than the second preset distance, the target grid probability of the first-level grid is updated to be a third preset coefficient, and the third preset coefficient is a probability value when the type of the first-level grid is an obstacle; If the target distance is less than or equal to the second preset distance, the target grid probability of the first-level grid is updated to a fourth preset coefficient, and the fourth preset coefficient is a probability value when the type of the first-level grid is ground.

8. A terrain data generating device, characterized in that: include: A point cloud data projection module is used to obtain point cloud data of the target area, project the point cloud of the current frame into a pre-constructed grid network, and obtain the current frame initial semantic data and historical semantic data of the first-level grid and the second-level grid in the grid network, wherein the current frame initial semantic data includes the initial grid probability obtained from the current frame point cloud, and the historical semantic data includes the semantic data obtained from all historical frame point clouds, and the second-level grid is composed of n adjacent first-level grids; A secondary grid updating module, used to update the initial grid probability of the secondary grid based on the distance from the secondary grid to the ground plane, and obtain the target grid probability of the secondary grid; A primary grid updating module, used for updating the initial grid probability of the primary grid within the secondary grid based on the target grid probability of the secondary grid, and obtaining the current frame semantic data of the primary grid; The semantic data fusion module is used to fuse the current frame semantic data and historical semantic data of the first-level grid to obtain the terrain data of the target area.

9. A terrain data generation system, characterized in that: include: at least one processor and at least one memory; The memory stores executable instructions of the processor; The processor is configured to execute the method according to any one of claims 1 to 7.

10. A data acquisition device, characterized in that: The method according to any one of claims 1 to 7 is used.

Images