Grid map updating method

Abstract

The embodiment of the specification provides a grid map updating method, wherein the grid map updating method comprises: acquiring a reference grid map containing a target object and a concerned object; determining a target region corresponding to the concerned object in the reference grid map, and constructing a target distribution heat map corresponding to the reference grid map according to attribute information of the target region; determining position information of the target object in the reference grid map, and constructing an object route map corresponding to the reference grid map according to the position information; constructing a grid distribution curve based on the target distribution heat map and the object route map, and updating the reference grid map into a target grid map according to the grid distribution curve.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of autonomous driving technology, and in particular to a grid map update method. Background Technology

[0002] With the booming development of the automotive industry, the demand for autonomous driving is growing among users. Furthermore, the need for autonomous driving technology in transportation, logistics, and unmanned delivery scenarios is becoming increasingly strong. In existing technologies, to achieve autonomous driving and improve driving safety, point cloud rasterization is typically used to map the collected point cloud data onto a raster map, which is then used for autonomous driving. Most 3D point cloud rasterization processes employ uniform rasterization based on the Cartesian system. While uniform rasterization can support autonomous driving, considering the complexity of driving scenarios, uniformly dividing a large space into grids presents a trade-off between resolution and computational resources. Reducing resolution to maintain real-time performance can save computational resources but leads to lower accuracy; conversely, increasing resolution to maintain performance significantly increases computational resource consumption. Therefore, an effective solution to address these issues is urgently needed. Summary of the Invention

[0003] In view of this, embodiments of this specification provide a raster map updating method. One or more embodiments of this specification also relate to a raster map updating apparatus, a model training method, a model training apparatus, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0004] According to a first aspect of the embodiments of this specification, a raster map updating method is provided, comprising:

[0005] Obtain a baseline raster map containing the target object and the objects of interest;

[0006] In the baseline raster map, the target region corresponding to the object of interest is determined, and a target distribution heatmap corresponding to the baseline raster map is constructed based on the attribute information of the target region.

[0007] The location information of the target object is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information;

[0008] Based on the target distribution heatmap and the object roadmap, a raster distribution curve is constructed, and the baseline raster map is updated to the target raster map according to the raster distribution curve.

[0009] According to a second aspect of the embodiments of this specification, another grid map updating method is provided, applied to a vehicle control terminal, including:

[0010] Obtain a baseline grid map containing the target vehicle and the vehicle of interest;

[0011] In the reference grid map, the target area corresponding to the vehicle of interest is determined, and a target distribution heatmap corresponding to the reference grid map is constructed based on the attribute information of the target area.

[0012] The location information of the target vehicle is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information;

[0013] A grid distribution curve is constructed based on the target distribution heatmap and the object roadmap, and the reference grid map is updated to a target grid map according to the grid distribution curve, wherein the target grid map is used for the autonomous driving of the target vehicle.

[0014] According to a third aspect of the embodiments of this specification, a model training method is provided, comprising:

[0015] Obtain a baseline raster image containing the target object and the object of interest, and input the baseline raster image into the initial raster partitioning model for processing to obtain predicted raster size information;

[0016] In the baseline raster map, the target region corresponding to the object of interest is determined, and a target distribution heatmap corresponding to the baseline raster map is constructed based on the attribute information of the target region.

[0017] The location information of the target object is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information;

[0018] A grid distribution curve is constructed based on the target distribution heatmap and the object route map, and the reference grid size information is determined based on the grid distribution curve;

[0019] The loss value is calculated based on the predicted grid size information and the baseline grid size information, and the parameters of the initial grid partitioning model are tuned using the loss value until a target grid partitioning model that meets the training stopping condition is obtained.

[0020] According to a fourth aspect of the embodiments of this specification, a raster map updating apparatus is provided, comprising:

[0021] The acquisition module is configured to acquire a baseline raster map containing the target object and the objects of interest.

[0022] The first construction module is configured to determine the target region corresponding to the object of interest in the baseline raster map, and construct a target distribution heatmap corresponding to the baseline raster map based on the attribute information of the target region.

[0023] The second construction module is configured to determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0024] The update module is configured to construct a raster distribution curve based on the target distribution heatmap and the object roadmap, and update the reference raster map to the target raster map according to the raster distribution curve.

[0025] According to a fifth aspect of the embodiments of this specification, another grid map updating device is provided, applied to a vehicle control terminal, comprising:

[0026] The raster acquisition module is configured to acquire a baseline raster image containing the target vehicle and the vehicle of interest;

[0027] A heatmap construction module is configured to determine the target area corresponding to the vehicle of interest in the reference grid map, and construct a target distribution heatmap corresponding to the reference grid map based on the attribute information of the target area.

[0028] A route map construction module is configured to determine the location information of the target vehicle in the reference grid map and construct an object route map corresponding to the reference grid map based on the location information;

[0029] The grid map update module is configured to construct a grid distribution curve based on the target distribution heatmap and the object roadmap, and update the reference grid map to the target grid map according to the grid distribution curve, wherein the target grid map is used for the autonomous driving of the target vehicle.

[0030] According to a sixth aspect of the embodiments of this specification, a model training apparatus is provided, comprising:

[0031] The model processing module is configured to acquire a baseline raster image containing the target object and the object of interest, and input the baseline raster image into the initial raster partitioning model for processing to obtain predicted raster size information;

[0032] The heat map construction module is configured to determine the target region corresponding to the object of interest in the baseline grid map, and construct a target heat map corresponding to the baseline grid map based on the attribute information of the target region.

[0033] The object route map construction module is configured to determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0034] The size information determination module is configured to construct a grid distribution curve based on the target distribution heatmap and the object route map, and determine the reference grid size information according to the grid distribution curve;

[0035] The model parameter tuning module is configured to calculate a loss value based on the predicted grid size information and the baseline grid size information, and use the loss value to tune the initial grid partitioning model until a target grid partitioning model that meets the training stopping condition is obtained.

[0036] According to a seventh aspect of the embodiments of this specification, a computing device is provided, comprising:

[0037] Memory and processor;

[0038] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the above-described raster update method or model training method.

[0039] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the above-described raster update method or model training method.

[0040] According to a ninth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described raster update method or model training method.

[0041] The raster update method provided in this embodiment aims to concentrate limited resources on the areas of interest and improve target detection performance without increasing processing time. It first acquires a baseline raster containing the target object and the corresponding area of ​​interest. Then, it determines the target region corresponding to the object of interest within the baseline raster and constructs a target distribution heatmap based on the attribute information of the target region. This heatmap reflects the areas of interest within the baseline raster. Simultaneously, it determines the location information of the target object within the baseline raster and constructs an object route map based on this location information. This route map reflects the route areas of interest within the baseline raster. Considering that the areas affecting the target object's movement are the areas of interest and route areas, the target distribution heatmap and object route map can be used as the basis for dividing the baseline raster. During raster updates, denser gratings can be added to the areas of interest and route areas, while sparser gratings can be added to other areas. In other words, a raster distribution curve is constructed based on the target distribution heatmap and object route map, reflecting the attention situation in each direction. This allows the baseline raster to be updated to the target raster. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of a raster map update method provided in one embodiment of this specification;

[0043] Figure 2 This is a flowchart illustrating a raster map update method provided in one embodiment of this specification;

[0044] Figure 3 This is a schematic diagram of a raster map in a raster map update method provided in one embodiment of this specification;

[0045] Figure 4 This is a schematic diagram of a raster map in another raster map update method provided in one embodiment of this specification;

[0046] Figure 5 This is a flowchart illustrating the processing procedure of a raster map update method provided in one embodiment of this specification.

[0047] Figure 6 This is a flowchart of another raster map update method provided in one embodiment of this specification;

[0048] Figure 7 This is a flowchart illustrating a model training method provided in one embodiment of this specification;

[0049] Figure 8 This is a schematic diagram of a grid map updating device provided in one embodiment of this specification;

[0050] Figure 9 This is a schematic diagram of another raster map updating device provided in one embodiment of this specification;

[0051] Figure 10 This is a schematic diagram of the structure of a model training device provided in one embodiment of this specification;

[0052] Figure 11 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0053] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0054] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0055] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0056] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0057] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0058] Point cloud: A collection of three-dimensional point data of the actual road surface collected by a three-dimensional laser scanning instrument.

[0059] Raster: In image editing, a raster is a pixel, meaning the conversion of instructions in an image into pixels.

[0060] This specification provides a raster map update method, and also relates to a raster map update apparatus, a model training method, a model training apparatus, a computing device, and a computer-readable storage medium, which are described in detail in the following embodiments.

[0061] See Figure 1 The schematic diagram illustrates the raster map update method provided in this embodiment. To concentrate limited resources on areas of interest and improve target detection performance without increasing processing time, a baseline raster map containing the target object and the corresponding area of ​​interest is first obtained. Then, the target area corresponding to the object of interest is determined within the baseline raster map. A target distribution heatmap corresponding to the baseline raster map is constructed based on the attribute information of the target area, thus representing the area of ​​interest in the baseline raster map through the heatmap. Simultaneously, the location information of the target object is determined within the baseline raster map, and an object route map corresponding to the baseline raster map is constructed based on the location information, thus representing the route area of ​​interest in the baseline raster map. Considering that the areas affecting the target object's movement are the areas of interest and route areas, the target distribution heatmap and object route map can be used as the basis for dividing the baseline raster map. This allows for the addition of denser grates to the areas of interest and route areas during raster updates, while sparser grates can be added to other areas. In other words, a raster distribution curve is constructed based on the target distribution heatmap and object route map, reflecting the attention situation in each direction, thereby updating the baseline raster map to the target raster map. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy.

[0062] See Figure 2 , Figure 2 A flowchart of a raster map update method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0063] Step S202: Obtain a baseline raster map containing the target object and the object of interest.

[0064] The grid map update method described in this embodiment can be applied to any autonomous driving scenario, such as autonomous driving of private cars, autonomous driving of public transportation vehicles, and autonomous driving of unmanned delivery vehicles. This embodiment uses an unmanned delivery vehicle as an example to illustrate the grid map update method. For descriptions of other scenarios, please refer to the same or corresponding descriptions in this embodiment. This embodiment will not elaborate further here.

[0065] Specifically, the target object refers to the moving object in the current autonomous driving scenario, such as unmanned delivery vehicles, cars, and transport vehicles. Correspondingly, the object of interest refers to the object of interest in the current autonomous driving scenario, such as pedestrians, roadblocks, vehicles, and green belts. Furthermore, the current scenario requires raster re-division processing of the baseline raster map containing both the target object and the object of interest. This re-division ensures that the re-divided raster map uses dense graticules for the areas of interest and sparse graticules for other areas, presenting the final raster map to be used. This reduces computational resources while ensuring computational accuracy during the application phase. Correspondingly, the baseline raster map refers to the raster map generated for the target object at the current moment before any raster updates. The baseline raster map has uniformly sized graticules, meaning it is a uniformly sized two-dimensional image. Subsequent dynamic raster division of this baseline raster map yields a non-uniformly distributed target raster map, enabling the use of the raster map derived from mapped point cloud data to drive the target object's movement in the autonomous driving scenario.

[0066] Based on this, in order to reduce the consumption of computing resources while improving the calculation accuracy of the region of interest, a reference grid map containing the target object and the object of interest can be obtained first. The reference object and the object of interest can be located in the reference grid map. Subsequently, the grid can be updated according to the position and attributes of the reference object and the object of interest, so that the grid distribution in the region of interest is denser and the grid distribution in the non-region of interest is sparser, which makes it more practical when driving the target object.

[0067] Furthermore, when obtaining the baseline raster map containing the baseline object and the object of interest, since the baseline raster map is the raster map that needs to be re-divided subsequently, and the re-division is based on the object of interest and the baseline object, it is necessary to combine the geographical location information of the two objects to determine the baseline raster map; in this embodiment, the specific implementation method is as follows:

[0068] A target virtual map is loaded based on the geographic location information of the target object, and an initial raster map is constructed based on the target virtual map; the initial raster map is updated based on the geographic location information of the target object and the geographic location information of the object of interest, and a reference raster map containing the target object and the object of interest is generated based on the update result; wherein the raster map contains rasters with the same raster size.

[0069] Specifically, geographic location information refers to the current coordinates of an object on the virtual map, used to locate the object's position on the map and construct a baseline raster map based on the target virtual map containing the object's location. The target virtual map refers to the map used by the target object during its movement, and this map has a mapping relationship with the target object's driving area. Correspondingly, the initial raster map refers to a two-dimensional map constructed using the target virtual map without raster division, and it does not include the target object or the object of interest.

[0070] Therefore, to construct a target raster map with dense raster coverage in the region of interest and sparse raster coverage in other regions, a virtual target map is first loaded based on the geographic location information of the target objects. An initial raster map, excluding the target objects and objects of interest, can then be constructed from this virtual map. To further refine this initial raster map and create a new raster map capable of raster re-division, the objects of interest and the target objects need to be added to the initial raster map, providing a basis for the re-division. At this point, the initial raster map can be updated based on the geographic location information of the target objects and the objects of interest, generating a baseline raster map containing both the target objects and the objects of interest. The baseline raster map contains raster cells of equal size for subsequent reconstruction.

[0071] For example, an unmanned delivery vehicle A, used for express delivery, is traveling in the rightmost lane of road A. During its journey, a virtual map is loaded based on vehicle A's current location information to construct an initial grid map. Then, by combining vehicle A's current location information with the location information of obstacles and pedestrians it needs to monitor, a baseline grid map is constructed. The constructed baseline grid map is as follows: Figure 3 As shown in (a), the grid size is uniform and includes the unmanned delivery vehicle A, roadblocks and pedestrians, which facilitates subsequent grid re-division for roadblocks and pedestrians, for use when driving the unmanned delivery vehicle A.

[0072] In summary, by constructing a baseline raster map with the same grid size and containing both the target object and the object of interest, it is convenient to use the objects of interest and the baseline object in the baseline raster map to construct elements for re-dividing the raster map. This allows for raster re-division based on actual driving needs, thereby meeting the computational requirements in real-world scenarios.

[0073] Step S204: Determine the target region corresponding to the object of interest in the reference grid map, and construct a target distribution heatmap corresponding to the reference grid map based on the attribute information of the target region.

[0074] Specifically, after obtaining the baseline raster map containing the object of interest and the target object, further considering that the re-division of the raster relies on the content corresponding to the region of interest of the target object—for example, if the target object encounters a road obstacle or pedestrian while driving, and needs to make an avoidance decision—the calculation is completed by mapping the point cloud in the raster map to the raster position. If each raster is the same size, the computing resources will be evenly distributed across each raster, potentially resulting in insufficient computing resources for the raster corresponding to the point cloud mapping position. Therefore, to improve the utilization of computing resources and the calculation accuracy of the region of interest, dense raster division can be used for the region of interest associated with the raster map, while sparse raster division can be used for other regions. Based on this, to achieve the above-mentioned raster division objective, the region of interest corresponding to the object of interest can be determined in the baseline raster map first, so as to construct a target distribution heatmap of the corresponding baseline raster map based on the region of interest. This allows the target distribution heatmap to be used as the basis for subsequent raster division, thereby making the raster division corresponding to the region of interest in the raster map denser.

[0075] Specifically, the target region refers to the bounding box in the baseline raster image that selects the object of interest. This bounding box can be any polygonal shape, as long as it can select the region of interest. This bounding box is used to represent the position of the object of interest in the raster image, allowing for denser raster division of this region during raster updates. Correspondingly, the attribute information refers to the attribute description information corresponding to the target region, including but not limited to the size, coordinates, and / or radius of the target region. The target distribution heatmap is a heatmap constructed based on the attribute information of the target region, which maps to the baseline raster image. This target distribution heatmap contains multiple heat regions, each corresponding to a unique heat value. The heat value indicates whether the heat region is associated with the object of interest. Using this heatmap as the raster distribution allows for denser raster division of heat regions with high heat values ​​during raster updates.

[0076] Furthermore, when constructing the target distribution heatmap, considering that the target distribution heatmap will be used as a raster distribution to update the baseline raster map, so that the updated raster map shows the effect of dense raster in the area of ​​interest and sparse in other areas; therefore, when constructing the target distribution heatmap, the area heat value can be calculated by combining the attribute information of the target area, so as to realize the use of heat value to represent the location of the area of ​​interest, and thus be used in the raster division stage; in this embodiment, the specific implementation method is as follows:

[0077] Determine the attribute information of the target region and construct an initial distribution heatmap corresponding to the baseline raster map; add regional heat values ​​to the heat regions corresponding to the target region in the initial distribution heatmap according to the attribute information, and add background heat values ​​to the background regions other than the heat regions; generate the target distribution heatmap based on the addition results of the regional heat values ​​and the background heat values.

[0078] Specifically, the initial distribution heatmap refers to a heatmap constructed based on a baseline raster map. This heatmap has a mapping relationship with the baseline raster map and contains multiple regions. The heat value corresponding to each region needs to be calculated based on the attribute information of the target region to represent the location of the target region, which is then mapped onto the raster map for raster division. Correspondingly, the heatmap region refers to the region in the initial distribution heatmap that corresponds to the target region; and the region heat value refers to the heat value added to the heat region. This heat value is added based on the attribute information of the target region and is used to characterize the target region corresponding to the heat region. Since the target region contains the object of interest, during the usage phase, the heat value can be combined to identify the heat region in the heatmap to improve the density of the raster corresponding to that region during raster division.

[0079] The background region specifically refers to other heat areas in the initial heatmap that do not correspond to the target region. Because these areas do not contain any objects of interest, they still need to be rasterized, albeit with a relatively sparser raster division. Therefore, a background heat value can be added to these areas. If no heat value is added to these areas (i.e., the heat value is set to 0 by default), the subsequently constructed raster distribution curve will not be smooth enough, resulting in an overly sparse or even non-existent raster map after re-division, thus affecting the usability of the raster map. Therefore, adding a preset background heat value to the background region ensures that the corresponding raster division in these areas can also be achieved without being too sparse or dense. The background heat value can be understood as a bias value for the heat value, biasing the raster density and making the raster division more reasonable. In practical applications, the background heat value can be set according to actual needs, as long as it does not exceed the regional heat value; this embodiment does not impose any limitations.

[0080] Therefore, to ensure that grid division is completed according to the region corresponding to the object of interest, and that the region corresponding to the object of interest is densely divided while other regions are sparsely divided and still usable, we can first determine the attribute information of the target region and construct an initial distribution heatmap corresponding to the baseline grid map. Then, we can add regional heat values ​​to the heatmap corresponding to the target region according to the attribute information to reflect the position of the object of interest in the heatmap. Simultaneously, to avoid the default heat values ​​of other regions affecting the accuracy of grid division (i.e., too sparse or too dense), we can add background heat values ​​to the background regions other than the heat regions. This ensures that each region has a corresponding and suitable heat value, enabling the generation of the target distribution heatmap based on the added regional and background heat values, facilitating its subsequent use as a grid distribution.

[0081] In practical applications, when constructing a target distribution heatmap, in order to ensure that the heat value of each region is assigned more reasonably and can be used as a grid distribution, the size of the target box corresponding to the object of interest in the baseline grid map can be read, and the Gaussian kernel radius can be calculated based on the width and height of the size. The radius is used as the heat value to create a target distribution heatmap with values ​​distributed between 0 and 1. The heat value in the target distribution heatmap can be mapped to the grid density. Therefore, it can be used as a grid distribution for subsequent grid updates, which can make the grid division of the region corresponding to the object of interest in the final target grid map more dense. At the same time, in order to avoid unreasonable grid division in other regions, such as the grid being too sparse or having no grid, it is also necessary to increase the heat value of the background region to avoid the background region not being assigned a grid. In addition, when calculating the grid distribution, it can be combined with the distribution of the target box corresponding to the object of interest for mapping, as shown in the following formula (1):

[0082] D v =max(D o +η,1); (1)

[0083] Among them, D v Indicates grid distribution; D o This represents the distribution of the target bounding boxes; η is a value between 0 and 1, i.e., a smoothing factor, which is calculated based on the heat values ​​in the heatmap. The larger η is, the smoother the final grid distribution, resulting in a denser grid during grid updates, used to represent the position of the object of interest in the grid map.

[0084] Furthermore, since the target object is in motion, the raster image calculated for each frame may be different, and the position of the object of interest will also change in each frame. In order to determine the target region corresponding to the object of interest in the reference raster image, the target region of the object of interest in the current frame can be predicted by combining the raster image of the previous frame. In this embodiment, the specific implementation method is as follows:

[0085] Obtain a historical raster image and determine the historical target region corresponding to the object of interest in the historical raster image; based on the region information of the historical target region, add the target region for the object of interest in the baseline raster image.

[0086] Specifically, the historical raster map refers to the raster map used in the calculation of the previous frame. Correspondingly, the historical target region is the target region corresponding to the object of interest in the historical raster map. The corresponding region information refers to the region information corresponding to the historical target region, including but not limited to region size, location, and other information.

[0087] Based on this, in order to accurately locate the target area of ​​the object of interest, the target area of ​​the current frame can be predicted by combining the historical target areas of historical frames. That is, first, the historical raster map is obtained, and the historical target area corresponding to the object of interest is determined in the historical raster map. At this time, based on the area information of the historical target area, the position of the target area corresponding to the object of interest in the current frame is predicted in the reference raster map, and the predicted target area is added to the raster map corresponding to the current frame, which can facilitate the subsequent construction of target distribution heatmap.

[0088] Continuing with the previous example, after obtaining... Figure 3 After obtaining the baseline grid map shown in (a), an initial distribution heatmap of the corresponding baseline grid map can be constructed based on the target boxes corresponding to obstacles and pedestrians in the baseline grid map. Subsequently, the Gaussian kernel radius can be calculated based on the dimensions of the target boxes of obstacles and pedestrians, and the heat value of the corresponding heat region in the initial distribution heatmap can be determined based on the calculation result. Simultaneously, a preset background heat value is added to the background area excluding obstacles and pedestrians, thus obtaining a target distribution heatmap corresponding to the baseline grid map with heat values ​​distributed between 0 and 1. Based on this, the grid distribution can be calculated, enabling the identification of obstacles and pedestrians in densely distributed grid areas, facilitating its use when driving the unmanned delivery vehicle A.

[0089] In summary, by combining the position of the object of interest in the baseline raster map to locate the target area, and then constructing a heatmap based on this, the area corresponding to the target area can be represented in the heatmap. Based on this, subsequent raster distribution calculations can make the raster distribution corresponding to the object of interest denser and other background areas sparser, thus making the updated target raster map more consistent with the road conditions in the real scene.

[0090] Step S206: Determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information.

[0091] Specifically, while constructing the target distribution heatmap as described above, it's important to consider that the area of ​​interest in the target object's current road conditions includes not only the object itself but also, and more importantly, the road distribution, such as motor vehicle lanes, non-motor vehicle lanes, zebra crossings, and yellow / white solid lines. This road distribution requires a corresponding dense grid to ensure that the updated target grid map can drive the target object safely and in accordance with traffic rules. Therefore, we can also combine the target object's location information to construct an object route map corresponding to the baseline grid map. This object route map reflects the road conditions, allowing for a denser grid distribution in the relevant road areas during grid re-division.

[0092] Specifically, location information refers to the positioning information of the target object during actual driving. After determining the location information, it allows for the determination of associated road information within the current driving scenario, enabling the construction of an object route map. Correspondingly, the object route map specifically refers to the route map (RoadMap) constructed after loading road information based on the location information. This route map reflects the road conditions corresponding to the current location and allows for the highlighting of specific road areas. This enables the grid division of these areas into denser grids and sparser grids for other areas.

[0093] Furthermore, when constructing the object route map, considering that it will be used as a grid distribution to update the baseline grid map, so that the updated grid map reflects the effect of dense grids in the road areas of interest and sparse grids in other areas, the object route map can be constructed in conjunction with the route interest values. In this embodiment, the specific implementation method is as follows:

[0094] Obtain the location information corresponding to the target object, and load the object association information according to the location information; construct an initial route map corresponding to the baseline raster map, and determine the associated regions in the initial route map based on the object association information; add route attention values ​​for the associated regions in the initial route map according to the object association information, and add non-route attention values ​​for non-associated regions other than the associated regions; generate the object route map based on the addition results of the route attention values ​​and the non-route attention values.

[0095] Specifically, object association information refers to the information of road objects that affect the travel of the target object in the road area associated with the target object on the map. This includes, but is not limited to, the road object's identification information, size information, and location information, such as pedestrian crossings and length in meters. Correspondingly, the initial route map refers to the route map corresponding to the base grid map without route attention values ​​added. Correspondingly, associated areas refer to the areas in the initial route map associated with road objects. Correspondingly, unassociated areas refer to the areas in the initial route map not associated with road objects. Correspondingly, route attention values ​​refer to the set values ​​added for associated areas, used to characterize the road objects that need attention during travel. Non-route attention values ​​refer to the set values ​​added for unassociated areas, used to ensure that although unassociated areas do not affect travel, attention values ​​still need to be set to ensure that these areas have grids of default size when influencing grid division, avoiding problems such as no grids or sparse grids.

[0096] Based on this, to ensure that grid division can be completed according to the associated regions corresponding to the road objects of interest in the road, and that the grid division in the associated regions is dense while the grid division in other regions is sparse and still usable, the location information of the target objects can be obtained first from the baseline grid map. At this time, the object association information can be loaded according to the location information. At the same time, an initial route map corresponding to the baseline grid map is constructed. Then, the associated regions are determined in the initial route map according to the object association information. On this basis, route attention values ​​are added to the associated regions in the initial route map according to the object association information, and non-route attention values ​​are added to the non-associated regions. This achieves the addition of route attention values ​​to the associated regions to promote grid density, and the addition of non-route attention values ​​to the non-associated regions to ensure the existence of grids but not too densely. After that, the object route map can be generated according to the addition results of route attention values ​​and non-route attention values, which can be used as a grid distribution in the future.

[0097] In practical applications, when constructing an object route map, in order to ensure that the attention value assigned to each area is more reasonable and can be used as a grid distribution, the location of the current target object can be located, and road information can be obtained from the map based on the location information, namely, the drivable area, pedestrian crossing, tidal lane, bus lane, etc. of the road; then, the attention value can be set to 1 for the areas of interest in the autonomous driving scenario, such as setting 1 for the drivable area and pedestrian crossing area, while setting 0 for other areas, such as green belts and blind paths. Based on this, attention values ​​of related and unrelated areas can be added to the initial route map to form an object route map, which is convenient for determining the grid distribution based on the attention value of each area in the object route map. In addition, when calculating the grid distribution, the distribution of related and unrelated areas can be combined for mapping, as shown in the following formula (2):

[0098] D v =max(D r +η,1); (2)

[0099] Among them, D v Indicates grid distribution; D r This represents the distribution of associated regions; η is a value between 0 and 1, i.e., a smoothing factor, which is calculated based on the attention value in the road map; the larger η is, the smoother the final grid distribution, and thus the grid is denser when updating the grid, which is used to reflect the position of road objects in the grid map.

[0100] Continuing with the previous example, after obtaining... Figure 3 After obtaining the baseline grid map shown in (a), the location of the unmanned delivery vehicle A on the map can be determined first. Then, the road information corresponding to the current location can be loaded, and an initial route map corresponding to the baseline grid map can be constructed. Afterward, the drivable area corresponding to the right lane can be determined in the initial route map based on the road information. The attention value for this area is then set to g1, and the attention values ​​for other areas are set to g2. A route map can be generated based on these settings. The grid distribution can then be calculated based on this, allowing densely distributed grid areas to correspond to drivable areas, thus facilitating the driving of the unmanned delivery vehicle A.

[0101] In summary, by combining object association information to determine the regions of interest and non-interest in the initial roadmap, and setting different attention values ​​for different regions, the distinction between regions of interest and non-interest can be reflected in the final constructed object roadmap, thereby affecting the density of the raster distribution. This allows the final updated target raster map to achieve a balance between computing resources and computational accuracy.

[0102] Step S208: Construct a grid distribution curve based on the target distribution heatmap and the object roadmap, and update the reference grid map to the target grid map according to the grid distribution curve.

[0103] Specifically, after constructing the corresponding baseline grid map, object route map, and target distribution heatmap, further considering that the object route map and target distribution heatmap are built based on the area of ​​interest and location information, and can reflect the content of interest during the target object's driving process, such as pedestrians and crosswalks, in order to allocate more computing resources to the content of interest, a denser grid can be divided for this content, while other areas can be divided into sparser grids. Therefore, a grid distribution curve can be constructed based on the target distribution map and object route map to reflect the distribution of the area of ​​interest and road areas under the current driving scenario. This distribution can then be used as the grid distribution to update the grids in the baseline grid map, thereby updating the baseline grid map into the target grid map for convenient use in autonomous driving scenarios.

[0104] Specifically, the grid distribution curve refers to the curve representing the corresponding grid distribution constructed based on the target distribution heatmap and the object roadmap. This curve reflects the probability of the heatmap region, background region and / or associated region, and non-associated region in the grid distribution direction within the target distribution heatmap. The probability reflects the grid distribution density, allowing for grid updates to be performed according to this curve, resulting in a denser grid distribution in the areas of interest and a sparser distribution in other areas. Correspondingly, the target grid map refers to the grid map obtained after updating the baseline grid map. The grid distribution in this map will be distributed in a manner where the areas of interest are densely distributed, and other areas are sparsely distributed.

[0105] Furthermore, when constructing grid distribution curves using the target distribution heatmap and object roadmap, and updating the baseline grid map, considering that the target distribution heatmap corresponds to the object of interest, while the object roadmap corresponds to road information, two different curves can be constructed. Based on these curves, the grid size can be calculated to update the grid map. In this embodiment, the specific implementation is as follows:

[0106] A target raster distribution curve is constructed based on the target distribution heatmap, and a route raster distribution curve is constructed based on the object route map; raster size information is calculated based on the target raster distribution curve and the route raster distribution curve; the raster size contained in the reference raster map is adjusted according to the raster size information, and a target raster map is generated based on the adjustment result.

[0107] Specifically, the target grid distribution curve refers to the grid distribution curve constructed based on the target distribution map. It is used to reflect the probability of heatmap areas and background areas in the grid distribution direction in the target distribution heatmap, and the grid distribution density is reflected through probability. Similarly, the route grid distribution curve refers to the grid distribution curve constructed based on the object route map. It is used to reflect the probability of associated and unassociated areas in the object route map in the grid distribution direction, and the grid distribution density is reflected through probability. Correspondingly, the grid size information refers to the grid size calculated by combining the target grid distribution curve and the route grid distribution curve. During the calculation process, the size of each grid can be calculated sequentially by combining the two grid distribution curves. Therefore, after all grid sizes have been calculated, the calculated grid size information can be used to adjust the grid size.

[0108] When calculating grid size information based on the target grid distribution curve and the route grid distribution curve, considering that the same grid may have probability values ​​on both curves, that is, the object of interest and the road object may overlap, the size of a grid can be calculated based on the two probability values. This can be done by taking the average value or by selecting the maximum value. Alternatively, the two grid sizes can be calculated separately, and then the average value or the maximum value can be taken as the grid size. This embodiment does not impose any limitations on this.

[0109] Based on this, when updating the baseline raster map, the target raster distribution curve can be constructed first based on the target distribution heatmap, and the route raster distribution curve can be constructed based on the object route map; the raster size information can be calculated by combining the target raster distribution curve and the route raster distribution curve; the raster size of each updated raster can be obtained, and then the raster size contained in the baseline raster map can be adjusted according to the raster size information to generate the target raster map based on the adjustment results.

[0110] For example, after obtaining the heatmap and roadmap, the distribution curves of the target boxes corresponding to the obstacles in the heatmap can be constructed in the x and y directions based on the heat values ​​in the heatmap. Figure 3 As shown in (b), p1 in the curve represents the probability of the roadblock being located at each position in the x and y directions. To ensure the grid distribution follows the probability distribution of the roadblocks, i.e., the grid distribution and roadblock distribution are consistent, the grid distribution can be calculated based on the curve. At this point, the size of the corresponding roadblock's grid can be calculated using the curve; that is, grid 11 has a size of s11*s11, grid 12 has a size of s12*s12, and so on, with grid 1n having a size of s1n*s1n. Similarly, based on the attention values ​​in the route map, the distribution curves of the drivable road area in the x and y directions in the route map are constructed, as shown below. Figure 3As shown in (b), p2 in the curve represents the probability of the drivable road area being located at each position in the x and y directions. To ensure the grid distribution follows the probability distribution of the drivable road area, i.e., the grid distribution and the drivable road area distribution are consistent, the grid distribution can be calculated based on the curve. The size of the corresponding grid for the drivable road area can then be calculated using the curve; that is, grid 21 has a size of s21*s21, grid 22 has a size of s22*s22, and so on, with grid 2n having a size of s2n*s2n. After obtaining the size of each grid, the baseline grid map can be updated based on the grid size, and the updated result can be obtained as shown in the figure. Figure 4 The target grid map shown has a denser grid distribution corresponding to roadblocks and drivable roads, while the grid distribution in other areas is sparser.

[0111] In summary, by combining the target grid distribution curve and the route grid distribution curve to calculate grid size information, it is possible to achieve grid distribution based on the influence of the region of interest, thereby ensuring that the grid distribution corresponding to the region of interest is denser and the grid distribution in other regions is sparser, thus achieving the goal of balancing computing resources and accuracy.

[0112] Furthermore, when updating the reference raster map to the target raster map using the obtained raster distribution curve, the raster scale in two directions is actually determined. This allows for the raster to be re-divided on the reference raster map according to the scale, thereby obtaining the target raster map. In this embodiment, the specific implementation method is as follows:

[0113] Construct horizontal distribution parameters for the associated horizontal dimension and vertical distribution parameters for the associated vertical dimension based on the grid distribution curve; generate horizontal grid scales based on the horizontal distribution parameters and vertical grid scales based on the vertical distribution parameters; divide the reference grid map into grids according to the horizontal grid scales and the vertical grid scales to obtain the target grid map.

[0114] Specifically, the horizontal distribution parameter refers to the raster distribution parameters associated with the horizontal dimension, used to reflect the raster size ratio in the horizontal dimension; correspondingly, the vertical distribution parameter refers to the raster distribution parameters associated with the vertical dimension, used to reflect the raster size in the vertical dimension. Correspondingly, the horizontal and vertical raster scales refer to the scales used when dividing the raster, enabling the re-division of the raster according to the scales, ensuring efficient raster division.

[0115] Based on this, when dividing the raster according to the raster distribution curve, we can first construct the horizontal distribution parameters of the horizontal dimension and the vertical distribution parameters of the vertical dimension based on the raster distribution curve; then we can generate the horizontal raster scale based on the horizontal distribution parameters and the vertical raster scale based on the vertical distribution parameters; after obtaining the raster scale, we can perform raster division processing on the reference raster map according to the horizontal raster scale and the vertical raster scale, and thus obtain the target raster map based on the division results.

[0116] In practical applications, after obtaining the grid distribution curve, the grid distribution is actually mapped along both the horizontal and vertical directions. That is, when calculating the horizontal scale, the vertical distribution parameters are compressed to obtain a one-dimensional vector associated with the horizontal dimension for calculating the horizontal grid scale; similarly, when calculating the vertical scale, the horizontal distribution parameters are compressed to obtain a one-dimensional vector associated with the vertical dimension for calculating the vertical grid scale.

[0117] Specifically, a discrete distribution can first be fitted based on the grid distribution curve, then high-density interpolation can be performed to return it to the discrete state. The area under the curve can be calculated using numerical methods such as the trapz function. Finally, a scale is found such that the area between every two scales approaches 1 / w of the vertical plane, where w is the total number of grid cells. This yields a grid scale that meets the usage requirements for grid division, resulting in the target grid map. In practical implementation, when calculating the grid scale, a neural network can be used to predict the grid distribution to approximate the above calculation process, thereby improving computational efficiency.

[0118] In summary, by using the raster distribution curve to calculate the distribution parameters in two dimensions and using this as a basis to determine the raster scale, the efficiency of raster calculation can be effectively improved, thereby ensuring the accuracy of the target raster map update.

[0119] Furthermore, after obtaining the target grid map, considering that it is used to drive the target object, the point cloud data needs to be mapped to the grid map before it can be used for autonomous driving calculations. However, because the grid is densely distributed according to the location of the objects of interest and road objects, and sparse in other areas, the grid coordinates may change, making it impossible to map the point cloud data to the corresponding grid in the grid map. To avoid this problem, a coordinate mapping relationship can be established first, and the mapping can be completed based on this relationship. In this embodiment, the specific implementation method is as follows:

[0120] Point cloud data is collected for the target object, and a coordinate mapping relationship is constructed between the point cloud data and the grids in the target raster image; the point cloud data is mapped to the target raster image based on the coordinate mapping relationship, wherein the target raster image that maps the point cloud data is used for the target object to perform object adjustment tasks.

[0121] Specifically, point cloud data refers to a set of 3D point data of the actual road surface collected by a 3D laser scanning instrument. These 3D points include, but are not limited to, 3D points corresponding to any object on the road, such as vehicles, people, trees, and road obstacles. Correspondingly, the coordinate mapping relationship refers to the mapping relationship between the point cloud coordinates in the point cloud data and the raster coordinates of the grid in the target raster image, and each raster coordinate can correspond to one or more point cloud coordinates. Correspondingly, the object adjustment task refers to the task of driving the target object to adjust its route, including but not limited to driving the target object to maintain a constant speed, accelerate, change lanes, and brake.

[0122] Therefore, because the grids in the target raster map change, the coordinates of each grid cell change. If a uniform raster map is still used for point cloud rasterization, the point cloud data will not be able to map to the actual location, thus affecting the driving safety of the target object. Therefore, to ensure that the point cloud data can still be accurately mapped to the corresponding grids, after acquiring the point cloud data, the grid coordinates of the non-uniform grids in the raster map need to be repositioned. Then, a coordinate mapping relationship needs to be established between the point cloud data and the grids with repositioned coordinates. This allows the point cloud data to be accurately mapped to the target raster map for subsequent processing based on the coordinate mapping relationship. Afterward, based on the coordinate mapping relationship, the point cloud data can be mapped to the target raster map, resulting in a target raster map that maps the point cloud data, enabling safe driving of the target object. The target raster map mapping the point cloud data will then drive the target object to perform object adjustment tasks.

[0123] For example, after the 3D laser scanner configured on unmanned delivery vehicle A collects point cloud data, it can match the point cloud coordinates with the grid coordinates. This matching process determines whether point cloud coordinate x1 matches with grid coordinate b1; point cloud coordinate x2 matches with grid coordinate b1, and so on, until all matching relationships are established. Once all matching relationships are determined, the point cloud coordinates can be used to map them onto the target grid map, enabling unmanned delivery vehicle A to drive safely and normally. For instance, if a road obstacle appears ahead, unmanned delivery vehicle A can be driven to change lanes to the left.

[0124] In summary, by establishing coordinate mapping relationships to map point cloud data onto the target raster map, it can be ensured that the mapping result is the same as the positional relationship of the real scene, thereby ensuring the safe and stable movement of the target object.

[0125] The raster update method provided in this embodiment aims to concentrate limited resources on the areas of interest and improve target detection performance without increasing processing time. It first acquires a baseline raster containing the target object and the corresponding area of ​​interest. Then, it determines the target region corresponding to the object of interest within the baseline raster and constructs a target distribution heatmap based on the attribute information of the target region. This heatmap reflects the areas of interest within the baseline raster. Simultaneously, it determines the location information of the target object within the baseline raster and constructs an object route map based on this location information. This route map reflects the route areas of interest within the baseline raster. Considering that the areas affecting the target object's movement are the areas of interest and route areas, the target distribution heatmap and object route map can be used as the basis for dividing the baseline raster. During raster updates, denser gratings can be added to the areas of interest and route areas, while sparser gratings can be added to other areas. In other words, a raster distribution curve is constructed based on the target distribution heatmap and object route map, reflecting the attention situation in each direction. This allows the baseline raster to be updated to the target raster. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy.

[0126] The following is in conjunction with the appendix Figure 5 Taking the application of the grid map update method provided in this specification in an autonomous driving scenario as an example, the grid map update method will be further explained. Figure 5 The present specification illustrates a flowchart of a raster map update method according to an embodiment, which includes the following steps.

[0127] Step S502: Load the target virtual map based on the geographical location information of the target vehicle, and construct an initial raster map based on the target virtual map;

[0128] Step S504: Based on the geographic location information of the target vehicle and the geographic location information of the vehicle of interest, the initial grid map is updated, and a baseline grid map containing the target vehicle and the vehicle of interest is generated according to the update result.

[0129] Step S506: Determine the target area corresponding to the vehicle of interest in the reference grid map, determine the attribute information of the target area, and construct an initial distribution heatmap corresponding to the reference grid map.

[0130] Step S508: Add regional heat values ​​to the heat regions corresponding to the target region in the initial distribution heatmap according to the attribute information, and add background heat values ​​to the background regions other than the heat regions.

[0131] Step S510: Generate a target distribution heatmap based on the added regional heat value and background heat value.

[0132] Step S512: Obtain the location information corresponding to the target vehicle, and load the vehicle association information based on the location information.

[0133] Step S514: Construct an initial route map corresponding to the baseline grid map, and determine the associated areas in the initial route map based on vehicle association information.

[0134] Step S516: Add route attention values ​​to associated areas in the initial route map according to vehicle association information, and add non-route attention values ​​to non-associated areas other than associated areas.

[0135] Step S518: Generate a vehicle route map based on the addition results of route attention values ​​and non-route attention values.

[0136] Step S520: Construct a target grid distribution curve based on the target distribution heatmap, and construct a route grid distribution curve based on the vehicle route map.

[0137] Step S522: Calculate grid size information based on the target grid distribution curve and the route grid distribution curve.

[0138] Step S524: Adjust the grid size contained in the reference grid map according to the grid size information, and generate the target grid map based on the adjustment result.

[0139] Subsequently, point cloud data can be collected for the target vehicle, and a coordinate mapping relationship between the point cloud data and the grids in the target raster map can be constructed. Based on the coordinate mapping relationship, the point cloud data is mapped to the target raster map, where the target raster map that maps the point cloud data is used for the target vehicle to perform vehicle adjustment tasks.

[0140] The raster update method provided in this embodiment aims to concentrate limited resources on the areas of interest and improve target detection performance without increasing processing time. It first acquires a baseline raster containing the target vehicle and the corresponding areas of interest. Then, it determines the target area corresponding to the vehicle of interest within the baseline raster and constructs a target distribution heatmap based on the attribute information of the target area. This heatmap reflects the areas of interest within the baseline raster. Simultaneously, it determines the location information of the target vehicle within the baseline raster and constructs a vehicle route map corresponding to the baseline raster based on the location information. This route map reflects the route areas of interest within the baseline raster. Considering that the areas affecting the target vehicle's movement are the areas of interest and route areas, the target distribution heatmap and vehicle route map can be used as the basis for dividing the baseline raster. This allows for the addition of denser grates to the areas of interest and route areas, while sparser grates can be added to other areas during raster updates. In other words, a raster distribution curve is constructed based on the target distribution heatmap and vehicle route map, reflecting the attention situation in each direction. This curve is then used to update the baseline raster to the target raster. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy.

[0141] Corresponding to the above method embodiments, this specification also provides another embodiment of a raster map update method. Figure 6 A schematic diagram of another raster map update method provided in one embodiment of this specification is shown. Figure 6 As shown, this method is applied to the vehicle control terminal and includes:

[0142] Step S602: Obtain a baseline grid map containing the target vehicle and the vehicle of interest;

[0143] Step S604: Determine the target area corresponding to the vehicle of interest in the reference grid map, and construct a target distribution heatmap corresponding to the reference grid map based on the attribute information of the target area.

[0144] Step S606: Determine the location information of the target vehicle in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0145] Step S608: Construct a grid distribution curve based on the target distribution heatmap and the object route map, and update the reference grid map to the target grid map according to the grid distribution curve, wherein the target grid map is used for the autonomous driving of the target vehicle.

[0146] The other raster map update method provided in this embodiment corresponds to the raster map update method in the above embodiments. The same or similar descriptions can be found in the corresponding content of the above embodiments, and will not be elaborated further in this embodiment. The vehicle control terminal is the terminal device that controls the target vehicle to drive.

[0147] In summary, to concentrate limited resources on areas of interest and improve target detection performance without increasing processing time, we can first obtain a baseline raster map containing the target vehicle and the corresponding areas of interest. Then, we determine the target region corresponding to the vehicle of interest in the baseline raster map, and construct a target distribution heatmap corresponding to the baseline raster map based on the attribute information of the target region. This heatmap reflects the areas of interest in the baseline raster map. Simultaneously, we determine the location information of the target vehicle in the baseline raster map, and construct a vehicle route map corresponding to the baseline raster map based on the location information. This route map reflects the route areas of interest in the baseline raster map. Considering that the areas affecting the target vehicle's movement are the areas of interest and route areas, the target distribution heatmap and vehicle route map can be used as the basis for dividing the baseline raster map. This allows for the addition of denser grates to the areas of interest and route areas, and sparser grates to other areas during raster updates. In other words, we construct a raster distribution curve based on the target distribution heatmap and vehicle route map, and use the curve to reflect the attention situation in each direction. This allows us to update the baseline raster map to the target raster map. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy.

[0148] Corresponding to the above method embodiments, this specification also provides a model training method embodiment. Figure 7 A schematic diagram of a model training method provided in one embodiment of this specification is shown. Figure 7 As shown, the method includes:

[0149] Step S702: Obtain a reference grid map containing the target object and the object of interest, and input the reference grid map into the initial grid division model for processing to obtain the predicted grid size information;

[0150] Step S704: Determine the target region corresponding to the object of interest in the reference grid map, and construct a target distribution heatmap corresponding to the reference grid map based on the attribute information of the target region;

[0151] Step S706: Determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0152] Step S708: Construct a grid distribution curve based on the target distribution heatmap and the object route map, and determine the reference grid size information according to the grid distribution curve;

[0153] Step S710: Calculate the loss value based on the predicted grid size information and the reference grid size information, and use the loss value to tune the parameters of the initial grid partitioning model until a target grid partitioning model that meets the training stopping condition is obtained.

[0154] Specifically, the initial grid partitioning model refers to a model capable of partitioning grids according to the content presented in the baseline grid image. The model's input is the baseline grid image, and its output is grid size information, which corresponds to the density of the region of interest and the characteristics of coefficients in other regions. Correspondingly, the predicted grid size information refers to the prediction results obtained by the initial grid partitioning model after processing the samples during the training phase. Correspondingly, the baseline grid size information refers to the label corresponding to the sample, i.e., the true grid size information. The calculation of this size information is the same as or similarly described in the above embodiments, and will not be elaborated further in this embodiment. Correspondingly, the training stopping condition refers to the conditions for stopping the training of the initial grid partitioning model, including but not limited to iteration count conditions, loss value comparison conditions, and validation set verification conditions.

[0155] Therefore, to enable rapid grid segmentation during the application phase, a grid segmentation model can be pre-trained. During training, samples—specifically, a baseline grid image containing the object of interest and the target object—are first acquired and input into the initial grid segmentation model to be trained, thereby obtaining predicted grid size information. Simultaneously, the baseline grid size information is calculated using the grid segmentation method described in the above embodiment as labels. The labels and model prediction results are then used to calculate the loss value, and the model is tuned using the loss value. The tuned model is then assessed to determine if it meets the training stopping condition. If not, new samples are selected, and the above process is repeated until a target grid segmentation model that meets the training stopping condition is obtained. In the application phase, the trained model can be directly used to obtain the segmented target grid image for driving the object.

[0156] Corresponding to the above method embodiments, this specification also provides embodiments of a raster map updating device. Figure 8 A schematic diagram of a raster map updating apparatus according to one embodiment of this specification is shown. Figure 8 As shown, the device includes:

[0157] Acquisition module 802 is configured to acquire a baseline raster map containing the target object and the object of interest;

[0158] The first construction module 804 is configured to determine the target region corresponding to the object of interest in the reference grid map, and construct a target distribution heatmap corresponding to the reference grid map based on the attribute information of the target region.

[0159] The second construction module 806 is configured to determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0160] The update module 808 is configured to construct a raster distribution curve based on the target distribution heatmap and the object roadmap, and update the reference raster map to the target raster map according to the raster distribution curve.

[0161] In an optional embodiment, the first building module 804 is further configured to:

[0162] Determine the attribute information of the target region and construct an initial distribution heatmap corresponding to the baseline raster map; add regional heat values ​​to the heat regions corresponding to the target region in the initial distribution heatmap according to the attribute information, and add background heat values ​​to the background regions other than the heat regions; generate the target distribution heatmap based on the addition results of the regional heat values ​​and the background heat values.

[0163] In an optional embodiment, the second building module 806 is further configured to:

[0164] Obtain the location information corresponding to the target object, and load the object association information according to the location information; construct an initial route map corresponding to the baseline raster map, and determine the associated regions in the initial route map based on the object association information; add route attention values ​​for the associated regions in the initial route map according to the object association information, and add non-route attention values ​​for non-associated regions other than the associated regions; generate the object route map based on the addition results of the route attention values ​​and the non-route attention values.

[0165] In an optional embodiment, the update module 808 is further configured to:

[0166] A target raster distribution curve is constructed based on the target distribution heatmap, and a route raster distribution curve is constructed based on the object route map; raster size information is calculated based on the target raster distribution curve and the route raster distribution curve; the raster size contained in the reference raster map is adjusted according to the raster size information, and a target raster map is generated based on the adjustment result.

[0167] In an optional embodiment, the update module 808 is further configured to:

[0168] Construct horizontal distribution parameters for the associated horizontal dimension and vertical distribution parameters for the associated vertical dimension based on the grid distribution curve; generate horizontal grid scales based on the horizontal distribution parameters and vertical grid scales based on the vertical distribution parameters; divide the reference grid map into grids according to the horizontal grid scales and the vertical grid scales to obtain the target grid map.

[0169] In an optional embodiment, the apparatus further includes:

[0170] The module is configured to acquire a historical raster image and determine the historical target region corresponding to the object of interest in the historical raster image; and add the target region to the reference raster image for the object of interest based on the region information of the historical target region.

[0171] In an optional embodiment, the acquisition module 802 is further configured to:

[0172] A target virtual map is loaded based on the geographic location information of the target object, and an initial raster map is constructed based on the target virtual map; the initial raster map is updated based on the geographic location information of the target object and the geographic location information of the object of interest, and a reference raster map containing the target object and the object of interest is generated based on the update result; wherein the raster map contains rasters with the same raster size.

[0173] In an optional embodiment, the apparatus further includes:

[0174] The rasterization module is configured to collect point cloud data for the target object and construct a coordinate mapping relationship between the point cloud data and the grids in the target raster image; based on the coordinate mapping relationship, the point cloud data is mapped to the target raster image, wherein the target raster image that maps the point cloud data is used by the target object to perform object adjustment tasks.

[0175] The raster map updating device provided in this embodiment aims to concentrate limited resources on the areas of interest and improve target detection performance without increasing processing time. It first acquires a baseline raster map containing the target object and the corresponding area of ​​interest. Then, it determines the target region corresponding to the object of interest within the baseline raster map. Based on the attribute information of the target region, it constructs a target distribution heatmap corresponding to the baseline raster map, thus representing the area of ​​interest in the baseline raster map through the heatmap. Simultaneously, it determines the location information of the target object within the baseline raster map, and constructs an object route map corresponding to the baseline raster map based on the location information, thus representing the route area of ​​interest in the baseline raster map through the route map. Considering that the areas affecting the target object's movement are the areas of interest and route areas, the target distribution heatmap and object route map can be used as the basis for dividing the baseline raster map. This allows for the addition of denser grates to the areas of interest and route areas during raster updates, while sparser grates can be added to other areas. In other words, a raster distribution curve is constructed based on the target distribution heatmap and object route map, reflecting the attention situation in each direction, thereby updating the baseline raster map to the target raster map. When performing grid division, the system can select the content of interest for dense grid division based on the current driving situation, while other areas are divided into sparse grids, thereby achieving dynamic gridding and balancing computing resources and detection accuracy.

[0176] The above is a schematic scheme of a raster map updating device according to this embodiment. It should be noted that the technical solution of this raster map updating device and the technical solution of the above-described raster map updating method belong to the same concept. For details not described in detail in the technical solution of the raster map updating device, please refer to the description of the technical solution of the above-described raster map updating method.

[0177] Corresponding to the above method embodiments, this specification also provides another embodiment of a raster map updating device. Figure 9 A schematic diagram of another raster map updating apparatus provided in one embodiment of this specification is shown. Figure 9 As shown, this device is used in the vehicle control system and includes:

[0178] The raster acquisition module 902 is configured to acquire a reference raster image containing the target vehicle and the vehicle of interest;

[0179] The heat map construction module 904 is configured to determine the target area corresponding to the vehicle of interest in the reference grid map, and construct a target distribution heat map corresponding to the reference grid map based on the attribute information of the target area.

[0180] The route map construction module 906 is configured to determine the location information of the target vehicle in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0181] The grid map update module 908 is configured to construct a grid distribution curve based on the target distribution heatmap and the object roadmap, and update the reference grid map to a target grid map according to the grid distribution curve, wherein the target grid map is used for the autonomous driving of the target vehicle.

[0182] The above is an illustrative scheme of another raster map updating device according to this embodiment. It should be noted that the technical solution of this raster map updating device and the technical solution of the above-described raster map updating method belong to the same concept. For details not described in detail in the technical solution of the raster map updating device, please refer to the description of the technical solution of the above-described raster map updating method.

[0183] Corresponding to the above method embodiments, this specification also provides embodiments of a model training device. Figure 10 A schematic diagram of a model training apparatus provided in one embodiment of this specification is shown. Figure 10 As shown, the device includes:

[0184] The model processing module 1002 is configured to acquire a reference raster image containing the target object and the object of interest, and input the reference raster image into the initial raster partitioning model for processing to obtain predicted raster size information;

[0185] The heat map construction module 1004 is configured to determine the target region corresponding to the object of interest in the baseline grid map, and construct a target heat map corresponding to the baseline grid map based on the attribute information of the target region.

[0186] The object route map construction module 1006 is configured to determine the location information of the target object in the reference grid map, and construct an object route map corresponding to the reference grid map based on the location information;

[0187] The size information determination module 1008 is configured to construct a grid distribution curve based on the target distribution heatmap and the object route map, and determine the reference grid size information according to the grid distribution curve;

[0188] The model parameter tuning module 1010 is configured to calculate a loss value based on the predicted grid size information and the reference grid size information, and to use the loss value to tune the initial grid partitioning model until a target grid partitioning model that meets the training stopping condition is obtained.

[0189] The above is an illustrative scheme of a model training device according to this embodiment. It should be noted that the technical solution of this model training device and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the model training device, please refer to the description of the technical solution of the model training method described above.

[0190] It should be noted that the algorithm modules (functional modules) mentioned in the above embodiments will differ depending on the type of autonomous vehicle. For example, different algorithm modules will be involved for logistics vehicles, public service vehicles, medical service vehicles, and terminal service vehicles. The following examples illustrate the algorithm modules for these four types of autonomous vehicles:

[0191] Logistics vehicles refer to vehicles used in logistics scenarios, such as logistics vehicles with automatic sorting functions, logistics vehicles with refrigeration and insulation functions, and logistics vehicles with measurement functions. These logistics vehicles involve different algorithm modules.

[0192] For example, logistics vehicles can be equipped with automated sorting devices that can automatically retrieve, transport, sort, and store goods after the vehicle arrives at its destination. This involves an algorithm module for goods sorting, which mainly implements the logical control of goods retrieval, handling, sorting, and storage.

[0193] For example, in cold chain logistics scenarios, logistics vehicles can also be equipped with refrigeration and insulation devices. These devices can refrigerate or keep warm transported fruits, vegetables, aquatic products, frozen foods, and other perishable foods, maintaining them at a suitable temperature environment and solving the problem of long-distance transportation of perishable foods. This involves an algorithm module for refrigeration and insulation control. This module is mainly used to dynamically and adaptively calculate the appropriate temperature for refrigeration or insulation based on information such as the nature of the food (or item), its perishability, transportation time, current season, and climate. Based on this appropriate temperature, the refrigeration and insulation device is automatically adjusted. This eliminates the need for manual temperature adjustments by transport personnel when transporting different foods or items, freeing them from tedious temperature control and improving the efficiency of refrigerated and insulated transportation.

[0194] For example, in most logistics scenarios, charges are based on package volume and / or weight. However, the number of logistics packages is enormous, and relying solely on couriers to measure package volume and / or weight is highly inefficient and labor-intensive. Therefore, some logistics vehicles are equipped with measuring devices that automatically measure the volume and / or weight of logistics packages and calculate the cost. This involves an algorithm module for logistics package measurement. This module primarily identifies the type of logistics package, determines the measurement method (e.g., volume measurement, weight measurement, or a combination of both), and completes the volume and / or weight measurement based on the determined method, as well as calculating the cost based on the measurement results.

[0195] Public service vehicles refer to vehicles that provide certain public services, such as fire trucks, de-icing trucks, water trucks, snowplows, garbage trucks, and traffic control vehicles. These public service vehicles involve different algorithm modules.

[0196] For example, the main task of an autonomous fire truck is to carry out reasonable fire extinguishing tasks at the fire scene. This involves an algorithm module for fire extinguishing tasks. This algorithm module needs to realize at least the logic of fire condition identification, fire extinguishing plan planning, and automatic control of fire extinguishing devices.

[0197] For example, the main task of a de-icing truck is to remove ice and snow from the road surface. This involves a de-icing algorithm module, which at least needs to identify the ice and snow conditions on the road surface, formulate a de-icing plan based on the ice and snow conditions, such as which road sections need de-icing, which road sections do not need de-icing, whether to use salting, the amount of salt to be applied, and the logic for automatic control of the de-icing device after determining the de-icing plan.

[0198] Among them, medical service vehicles refer to autonomous vehicles capable of providing one or more medical services. These vehicles can provide medical services such as disinfection, temperature measurement, medication dispensing, and isolation. This involves algorithm modules that provide various self-service medical services. These algorithm modules mainly realize the identification of disinfection needs and the control of disinfection devices so that the disinfection devices can disinfect patients; or the identification of patient positions and the control of temperature measuring devices to automatically place them close to the patient's forehead or other positions to measure the patient's temperature; or the identification of symptoms, the provision of prescriptions based on the diagnosis results, and the identification of medicines / medicine containers, as well as the control of the medication dispensing robot to make it grab medicines for patients according to the prescription, etc.

[0199] Among them, terminal service vehicles refer to self-service autonomous vehicles that can replace some terminal equipment to provide users with certain convenient services. For example, these vehicles can provide users with services such as printing, attendance, scanning, unlocking, payment, and retail.

[0200] For example, in some application scenarios, users often need to go to a specific location to print or scan documents, which is time-consuming and laborious. Therefore, a terminal service vehicle has emerged that can provide printing / scanning services to users. These service vehicles can interconnect with user terminal devices. Users issue print commands through their terminal devices, and the service vehicle responds by automatically printing the required documents and delivering them to the user's location. Users do not need to queue at the printer, greatly improving printing efficiency. Alternatively, it can respond to scanning commands issued by users through their terminal devices, move to the user's location, and the user places the document to be scanned on the service vehicle's scanning tool to complete the scan, eliminating the need to queue at the printer / scanner and saving time and effort. This involves an algorithm module that provides printing / scanning services. This algorithm module needs to at least recognize the interconnection with the user's terminal device, the response to print / scan commands, the user's location, and movement control.

[0201] For example, with the development of new retail projects, more and more e-commerce companies are using vending machines to deliver goods to office buildings and public areas. However, these vending machines are placed in fixed locations and cannot be moved. Users need to go to the vending machine to purchase the goods they need, which is inconvenient. Therefore, self-driving vehicles providing retail services have emerged. These vehicles can carry goods automatically and provide corresponding self-service shopping apps or shopping portals. Users can place orders with the self-driving vehicles using their mobile phones or other devices through the apps or shopping portals. The order includes the name and quantity of the goods to be purchased and the user's location. After receiving the order request, the vehicle can determine whether the remaining goods are available and whether the quantity is sufficient. If the goods are available and the quantity is sufficient, the vehicle can automatically move to the user's location and provide the goods, further improving the convenience of shopping, saving users time, and allowing them to focus on more important things. This involves the algorithm modules for providing retail services. These algorithm modules mainly implement logic such as responding to user order requests, order processing, product information maintenance, user location positioning, and payment management.

[0202] Figure 11 A structural block diagram of a computing device 1100 according to one embodiment of this specification is shown. The components of the computing device 1100 include, but are not limited to, a memory 1110 and a processor 1120. The processor 1120 is connected to the memory 1110 via a bus 1130, and a database 1150 is used to store data.

[0203] The computing device 1100 also includes an access device 1140, which enables the computing device 1100 to communicate via one or more networks 1160. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1140 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0204] In one embodiment of this specification, the aforementioned components of the computing device 1100 and Figure 11 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 11 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0205] The computing device 1100 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1100 can also be a mobile or stationary server.

[0206] The processor 1120 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described raster update method or model training method.

[0207] The above is a schematic representation of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the technical solutions of the aforementioned raster map update method or model training method. Details not described in detail in the technical solution of the computing device can be found in the descriptions of the technical solutions of the aforementioned raster map update method or model training method.

[0208] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described raster update method or model training method.

[0209] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the above-described raster map update method or model training method. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-described raster map update method or model training method.

[0210] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described raster update method or model training method.

[0211] The above is an illustrative scheme of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the above-described raster map update method or model training method. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the above-described raster map update method or model training method.

[0212] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0213] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0214] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0215] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0216] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A raster map update method, comprising: Obtain a baseline raster map containing the target object and the objects of interest; In the baseline raster map, the target region corresponding to the object of interest is determined, and a target distribution heatmap corresponding to the baseline raster map is constructed based on the attribute information of the target region. The location information of the target object is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information; A grid distribution curve is constructed based on the target distribution heatmap and the object roadmap, and the reference grid map is updated to the target grid map according to the grid distribution curve. The grid distribution curve is used to reflect the probability of each region in the target distribution heatmap in the grid distribution direction.

2. The raster map update method according to claim 1, wherein constructing a target distribution heatmap corresponding to the baseline raster map based on the attribute information of the target region includes: Determine the attribute information of the target region and construct an initial distribution heatmap corresponding to the baseline raster map; According to the attribute information, add a regional heat value to the heat region corresponding to the target region in the initial distribution heat map, and add a background heat value to the background region other than the heat region; The target distribution heatmap is generated based on the addition of the regional heat value and the background heat value.

3. The raster map update method according to claim 1, wherein determining the location information of the target object in the reference raster map and constructing an object route map corresponding to the reference raster map based on the location information includes: Obtain the location information corresponding to the target object, and load the object association information based on the location information; Construct an initial route map corresponding to the baseline raster map, and determine the associated regions in the initial route map based on the object association information; According to the object association information, add route attention values ​​for the associated areas in the initial route map, and add non-route attention values ​​for the non-associated areas other than the associated areas; Based on the addition of the route attention value and the non-route attention value, the object route map is generated.

4. The raster map updating method according to claim 1, wherein constructing a raster distribution curve based on the target distribution heatmap and the object roadmap, and updating the reference raster map to the target raster map according to the raster distribution curve, comprises: Construct a target grid distribution curve based on the target distribution heatmap, and construct a route grid distribution curve based on the object route map; Calculate grid size information based on the target grid distribution curve and the route grid distribution curve; The grid sizes contained in the reference grid map are adjusted according to the grid size information, and the target grid map is generated based on the adjustment results.

5. The raster map updating method according to claim 1, wherein updating the reference raster map to the target raster map according to the raster distribution curve comprises: Based on the grid distribution curve, construct the horizontal distribution parameters of the associated horizontal dimension and the vertical distribution parameters of the associated vertical dimension; A horizontal grid scale is generated based on the horizontal distribution parameters, and a vertical grid scale is generated based on the vertical distribution parameters; The reference raster image is divided into raster sections according to the horizontal and vertical raster scales to obtain the target raster image.

6. The raster map update method according to claim 1, further comprising, before the step of determining the target region corresponding to the object of interest in the reference raster map, the method includes: Obtain a historical raster image, and determine the historical target area corresponding to the object of interest in the historical raster image; Based on the regional information of the historical target area, the target area is added to the baseline raster map for the object of interest.

7. The raster map update method according to any one of claims 1-6, wherein obtaining the baseline raster map containing the target object and the object of interest comprises: Load a target virtual map based on the geographic location information of the target object, and construct an initial raster map based on the target virtual map; Based on the geographic location information of the target object and the geographic location information of the object of interest, the initial raster map is updated, and a baseline raster map containing the target object and the object of interest is generated according to the update result; The grids in the reference grid map have the same grid size.

8. The raster map updating method according to any one of claims 1-6, wherein after the step of updating the reference raster map to the target raster map according to the raster distribution curve is performed, the method further includes: Point cloud data is collected for the target object, and a coordinate mapping relationship is constructed between the point cloud data and the grids in the target raster image; The point cloud data is mapped to the target raster image based on the coordinate mapping relationship, wherein the target raster image that maps the point cloud data is used by the target object to perform object adjustment tasks.

9. A raster graph update method, applied to a vehicle control terminal, comprising: Obtain a baseline grid map containing the target vehicle and the vehicle of interest; In the reference grid map, the target area corresponding to the vehicle of interest is determined, and a target distribution heatmap corresponding to the reference grid map is constructed based on the attribute information of the target area. The location information of the target vehicle is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information; A grid distribution curve is constructed based on the target distribution heatmap and the object route map, and the reference grid map is updated to the target grid map according to the grid distribution curve. The target grid map is used for the autonomous driving of the target vehicle, and the grid distribution curve is used to reflect the probability of each region in the target distribution heatmap in the grid distribution direction.

10. A model training method, comprising: Obtain a baseline raster image containing the target object and the object of interest, and input the baseline raster image into the initial raster partitioning model for processing to obtain predicted raster size information; In the baseline raster map, the target region corresponding to the object of interest is determined, and a target distribution heatmap corresponding to the baseline raster map is constructed based on the attribute information of the target region. The location information of the target object is determined in the reference grid map, and an object route map corresponding to the reference grid map is constructed based on the location information; A grid distribution curve is constructed based on the target distribution heatmap and the object route map, and the reference grid size information is determined according to the grid distribution curve. The grid distribution curve is used to reflect the probability of each region in the target distribution heatmap in the grid distribution direction. The loss value is calculated based on the predicted grid size information and the baseline grid size information, and the parameters of the initial grid partitioning model are tuned using the loss value until a target grid partitioning model that meets the training stopping condition is obtained.

11. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 10.

12. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

Images