Map dynamic rendering method, device and equipment

Abstract

This invention discloses a method, apparatus, and device for dynamic map rendering. The method includes: acquiring environmental perception data; evaluating the scene complexity of the current scene based on the environmental perception data to obtain a scene complexity index; acquiring M-level target map data corresponding to the current scene from a pre-established map database; fusing the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3; and rendering the map based on the scene complexity index and the fused map data, wherein the level of detail in the rendered map is negatively correlated with the magnitude of the scene complexity index. This invention solves the technical problem of existing high-precision maps having data redundancy or insufficient accuracy, resulting in an inability to meet the needs of complex scenes.

Description

Technical Field

[0001] This invention belongs to the field of digital maps, and in particular relates to a method, apparatus and device for dynamic map rendering. Background Technology

[0002] In vehicle-road-cloud integrated systems, digital twin technology is the core support for achieving real-time mapping between the physical and virtual worlds. High-precision maps, as the foundational data for digital twins, directly impact system performance through their accuracy and fusion efficiency. Current high-precision maps employ a single level of precision (e.g., meter-level), which cannot meet the demands of complex scenarios. For example, fine details such as lane lines and manhole covers require centimeter-level precision, while regional road networks only need hundred-meter-level precision. This single level of precision leads to data redundancy or insufficient accuracy, resulting in an inability to meet the needs of complex scenarios. Summary of the Invention

[0003] This invention provides a method, apparatus, and device for dynamic map rendering, which addresses the technical problem that existing high-precision maps suffer from data redundancy or insufficient accuracy, resulting in their inability to meet the needs of complex scenarios.

[0004] In a first aspect, embodiments of the present invention provide a method for dynamic map rendering, comprising: acquiring environmental perception data; evaluating the scene complexity of the current scene based on the environmental perception data to obtain a scene complexity index; acquiring M-level target map data corresponding to the current scene from a pre-established map database; fusing the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3; and rendering the map based on the scene complexity index and the fused map data, wherein the amount of detail in the rendered map is negatively correlated with the magnitude of the scene complexity index.

[0005] In conjunction with the first aspect, in some embodiments, the map database is pre-constructed through the following steps: constructing map data at N accuracy levels, wherein map features in each accuracy level are assigned a unique geocode, and N is an integer greater than or equal to 3; and establishing a hierarchical association between the unique geocodes of map features in the N accuracy levels of map data.

[0006] In conjunction with the first aspect, in some embodiments, the N precision levels of map data include: map data of a first precision level, map data of a second precision level, and map data of a third precision level; the unique geocode of each map feature in the first precision level map data is associated with the unique geocode of multiple map features in the second precision level map data, and the unique geocode of each map feature in the second precision level map data is associated with the unique geocode of multiple map features in the third precision level map data.

[0007] In conjunction with the first aspect, in some embodiments, the step of rendering the map based on the scene complexity index and the fused map data includes: determining the target resolution and target frame rate based on the scene complexity index; allocating video memory space based on the target resolution and target frame rate; and using the video memory space to render the map using the scene complexity index and the fused map data.

[0008] In conjunction with the first aspect, in some embodiments, the step of using the scene complexity index and the fused map data for map rendering includes: if the scene complexity index is greater than a first threshold, rendering a first-level map based on the fused map data; if the scene complexity index is less than or equal to the first threshold and greater than or equal to a second threshold, rendering a second-level map based on the fused map data, wherein the rendering details of the second-level map are less than those of the first-level map; and if the scene complexity index is less than or equal to the second threshold, rendering a third-level map, wherein the rendering details of the third-level map are less than those of the second-level map.

[0009] In conjunction with the first aspect, in some embodiments, acquiring environmental perception data includes: acquiring environmental perception data collected by a sensor cluster of vehicle-road-cloud multi-terminals, the environmental perception data including image data and radar data; evaluating the scene complexity of the current scene based on the environmental perception data, including: detecting edge density and texture complexity of the image data; determining traffic flow density based on the radar data; and fusing the traffic flow density, the edge density, and the texture complexity to obtain the scene complexity index.

[0010] In conjunction with the first aspect, in some embodiments, the fusion of the M-level target map data to obtain fused map data includes: selecting reference map data from the M-level target map data, wherein the target map data at each level other than the reference map data in the M-level target map data are respectively used as map data to be registered; registering the map data to be registered to the coordinate system of the reference map data using an iterative nearest point algorithm based on Halanobis distance, to obtain registered M-level target map data; and fusing the registered M-level target map data using a federated Kalman filter framework to obtain the fused map data.

[0011] In conjunction with the first aspect, in some embodiments, the step of using the iterative nearest neighbor algorithm based on Mahalanobis distance to register the map data to be registered at each level to the coordinate system of the reference map data includes: for each point in each map data to be registered, searching for the nearest neighbor point within a preset radius in the reference map data based on Mahalanobis distance to obtain a set of point pairs; calculating the optimal rigid body transformation for the set of point pairs, and iteratively updating the map data to be registered based on the optimal rigid body transformation; when the error change rate between two adjacent iterations is less than an error threshold or the number of iterations is greater than a number threshold, the update of the map data to be registered is terminated; the federated Kalman filter framework includes a main filter and M sub-filters, and the step of using the federated Kalman filter framework to fuse the registered M-level target map data to obtain the fused map data includes: estimating the spatial state of map feature points in the M-level target map data one-to-one through the M sub-filters to obtain M local estimation results corresponding to the M-level target map data; the main filter receives the M local estimation results and fuses the M local estimation results to obtain the fused map data.

[0012] Secondly, embodiments of the present invention provide a dynamic map rendering apparatus, comprising: a data acquisition unit for acquiring environmental perception data; an evaluation unit for evaluating the scene complexity of the current scene based on the environmental perception data to obtain a scene complexity index; a map acquisition unit for acquiring M-level target map data corresponding to the current scene from a pre-established map database; a fusion unit for fusing the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3; and a rendering unit for rendering a map based on the scene complexity index and the fused map data, wherein the larger the scene complexity index, the higher the accuracy of the rendered map.

[0013] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the map dynamic rendering method described in any embodiment of the first aspect.

[0014] The one or more technical solutions provided in the embodiments of the present invention achieve at least the following technical effects or advantages: This invention, through scene complexity assessment, accurately determines the complexity index of the vehicle's current scene and then selectively selects and fuses M-level target map data. This fusion of multi-level map data can accommodate the accuracy requirements of different scenarios—in complex scenarios, redundant detailed data is unnecessary, retaining only the core navigation and positioning data, reducing data transmission and processing redundancy and improving rendering efficiency; in simple scenarios, more detailed map data can be used to ensure rendering accuracy and avoid insufficient precision. This scene-adaptive data selection, fusion, and rendering fundamentally solves the contradiction between data redundancy and insufficient precision in existing technologies, enabling maps to adapt to scenarios of varying complexity, especially meeting the usage needs of complex scenarios such as highways and congested urban roads.

[0015] Based on the implementation methods provided in the above aspects, the present invention can be further combined to provide more implementation methods. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 A flowchart of a map dynamic rendering method according to some embodiments of the present invention is shown; Figure 2 A schematic diagram of the structure of a map dynamic rendering apparatus according to some embodiments of the present invention is shown; Figure 3 A schematic diagram of the structure of an electronic device according to some embodiments of the present invention is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the embodiments of this specification, the term "multiple" means "two or more", that is, including two or more cases; the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] The map dynamic rendering method provided in this invention can support high-precision map services and real-time decision support in application scenarios such as smart buses, urban logistics, and autonomous passenger vehicles. It is especially suitable for high-precision positioning, environmental perception, and dynamic scene visualization in vehicle-road-cloud integrated systems. Figure 1 A flowchart of a map dynamic rendering method according to some embodiments of the present invention is shown. For example... Figure 1 As shown in the figure, the map dynamic rendering method provided by the embodiment of the present invention includes the following steps S101-S105.

[0021] S101: Acquire environmental perception data.

[0022] In some embodiments, environmental perception data collected by a multi-terminal sensor cluster of vehicle-road-cloud is acquired, and the acquired environmental perception data includes multi-source data. The multi-source data includes image data and radar data. The sensor cluster includes one or more of the following: a traffic flow density sensor (which may be a millimeter-wave radar), a camera, and an inertial measurement unit (IMU) deployed at the vehicle and / or roadside.

[0023] S102: Based on environmental perception data, assess the complexity of the current scene and obtain the scene complexity index.

[0024] In some embodiments, a neural network rendering technology, NVIDIA RTX MegaGeometry, is integrated into the GPU-accelerated rendering engine. This technology supports 8K resolution rendering and monitors scene complexity in real time by analyzing traffic density data collected by a traffic density sensor and image data collected by a camera.

[0025] In some embodiments, assessing the scene complexity of the current scene based on environmental perception data may include: detecting edge density and texture complexity of image data; determining traffic density based on radar data; and fusing traffic density (unit: vehicles / km), edge density, and texture complexity to obtain a scene complexity index.

[0026] Understandably, fusing traffic density, edge density, and texture complexity to obtain the scene complexity index can involve: normalizing traffic density, edge density, and texture complexity; and then, based on the one-to-one weighting coefficients corresponding to traffic density, edge density, and texture complexity, weighting and calculating the normalized traffic density, edge density, and texture complexity to obtain the scene complexity index (SCI): SCI = W1 × Edge Density + W2 × Texture Complexity + W3 × Traffic Density. For example, W1, W2, and W3 correspond to 0.5, 0.3, and 0.2, respectively.

[0027] In other embodiments, edge density and texture complexity are detected in the image data. An image complexity index is determined based on the texture complexity and edge density. Traffic density is determined using a traffic density sensor. The scene complexity index includes two independent data dimensions: image complexity index and traffic density, which do not require weighting or calculation. For example, rendering the first-level map is triggered when the density is greater than 50 vehicles / km or the image complexity index is greater than 0.8; otherwise, rendering the second-level map is triggered.

[0028] S103: Obtain the M-level target map data corresponding to the current scene from the pre-established map database.

[0029] In this embodiment of the invention, the map database includes map data of multiple accuracy levels. Each accuracy level of map data includes multiple map features. Each map feature in each accuracy level of map data is assigned a unique geocode. For example, the unique geocode uses map information coding (GEOCODE). Map features in map data of different accuracy levels are hierarchically associated based on the unique geocode.

[0030] In some embodiments, map data at N accuracy levels may include: map data at a first accuracy level, map data at a second accuracy level, and map data at a third accuracy level. In the map data at the first accuracy level: the unique geocode of each map feature is associated with the unique geocodes of multiple map features in the map data at the second accuracy level. In the map data at the second accuracy level: the unique geocode of each map feature is associated with the unique geocodes of multiple map features in the map data at the third accuracy level.

[0031] In some embodiments, the three precision levels of map data in the map database correspond to: centimeter-level map data, meter-level map data, and hundred-meter-level map data. In hundred-meter-level map data, each map element is a region, and the unique geographic code corresponding to each region is the region code. In meter-level map data, each map element is a road segment, and the unique geographic code corresponding to each road segment is the road segment code. In centimeter-level map data, each map element is a lane code.

[0032] Centimeter-level map data: with an accuracy of ±5cm, including fine map elements such as lane lines and manhole covers. The map database stores the three-dimensional coordinates, material properties (such as asphalt and concrete), and dynamic status (such as manhole cover damage markings) of each map element. It is encoded in the form of area code-road segment code-lane code, such as: A01-B02-C03.

[0033] Meter-level map data: accuracy is ±0.5m, including road segment outlines and traffic signs. The map database stores the road segment topology (such as the number of lanes and speed limits) and the location and type of traffic signs (such as speed limit signs and traffic lights) for each map element, encoded using area code-road segment code, such as: A01-B02.

[0034] 100-meter-level map data: with an accuracy of ±5m, including regional road network topology, encoded using regional codes such as A01, and the map database stores road connectivity, entrance and exit locations, and traffic flow statistics for each map element.

[0035] By establishing hierarchical associations between centimeter-level, meter-level, and hexadecimal-level map data through map information coding (GEOCODE), a hierarchical structure of area code-road segment code-lane code is formed, enabling rapid retrieval of three levels of map data. For example, when querying the road network of area code A01 in hexadecimal-level map data, it automatically associates with the road segment codes of all road segments associated with area code A01 in meter-level map data, and automatically associates with the lane codes associated with the road segment codes of all road segments associated with area code A01 in centimeter-level map data. The map database employs spatial indexing technology (such as R-tree) to solve the problems of single-precision maps and inefficient association in traditional maps. Joint queries of map data at different precision levels can be completed in less than 50ms.

[0036] Of course, in practice, the map database can include map data of more precision levels, such as map data of precision levels one through four.

[0037] In some embodiments, the map database can be stored using a distributed storage architecture, where data is sharded and stored on multiple nodes using a consistent hashing algorithm.

[0038] In some embodiments, the map database can be pre-constructed through the following steps: constructing map data at N precision levels, where each precision level assigns a unique geocode to map features, and N is an integer greater than or equal to 3; establishing a hierarchical association between the unique geocodes of map features at the N precision levels. By hierarchically associating map features of different precision levels with unique geocodes based on a unified coding rule, cross-precision retrieval is facilitated, reducing the excessive time required for cross-precision map retrieval and supporting real-time applications.

[0039] S104: Fuse the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3.

[0040] In some embodiments, M-level target map data can be fused by combining the Iterative Closest Point (ICP) algorithm based on Halanobis distance and the federated Kalman filter framework to obtain fused map data.

[0041] In some embodiments, fusing M-level target map data with an iterative nearest point algorithm based on Halanobis distance and a federated filtering framework to obtain fused map data may include the following steps: selecting a reference map data from the M-level target map data, and using the target map data at each level other than the reference map data as map data to be registered; using the iterative nearest point algorithm based on Halanobis distance to register each level of map data to be registered to the coordinate system of the reference map data to obtain registered M-level target map data; and using a federated Kalman filtering framework to fuse the registered M-level target map data to obtain fused map data.

[0042] In some embodiments, the iterative nearest neighbor algorithm based on Mahalanobis distance is used to register map data at each level to the coordinate system of the reference map data. This may include: for each point in each map data to be registered, searching for the nearest neighbor point within a preset radius in the reference map data based on Mahalanobis distance to obtain a set of point pairs; calculating the optimal rigid body transformation for the set of point pairs, and iteratively updating the map data to be registered based on the optimal rigid body transformation; and ending the iterative update of the map data to be registered when the error change rate between two adjacent iterations is less than the error threshold or the number of iterations is greater than the number threshold.

[0043] Understandably, the highest precision level of the M-level target map data can be used as the baseline map data, while target map data of other precision levels can be used as the maps to be registered. Taking target map data with centimeter, meter, and hundred-meter precision levels as an example, the centimeter precision target map data can be used as the baseline map data, while the meter and hundred-meter precision target map data can be used as the maps to be registered.

[0044] Understandably, searching for the nearest neighbor within a preset radius in the base map data based on the Mahalanobis distance can be done by taking the current point in the map data to be registered as the center and searching for neighboring points within a radius of 0.5m in the base map data.

[0045] The iterative closest point algorithm based on Haranobis distance improves robustness and accuracy against noise by introducing the local covariance matrix of the point cloud and replacing Euclidean distance with Mahalanobis distance. Combined with termination conditions (e.g., error rate of change < 0.1% or number of iterations > 50), the registration error is controlled within ±5cm. Compared to the traditional iterative closest point algorithm based on Euclidean distance, this algorithm improves registration accuracy and reduces computation time.

[0046] In some embodiments, in the step of calculating the optimal rigid body transformation for a set of point pairs, the rigid body transformation ensures that the shape, size, and relative distances between points in the transformed point cloud remain completely unchanged, with only rotation and translation occurring. The optimal rigid body transformation includes finding the optimal rotation matrix R and translation vector t, and using the optimal rotation matrix R and translation vector t to update the map data to be registered, so that the point cloud of the updated map data to be registered overlaps with the point cloud of the reference map data as much as possible.

[0047] In some embodiments, the step of calculating the optimal rigid body transformation for a set of point pairs may include: calculating the centroid of corresponding points in the set of point pairs, and calculating the initial translation vector through the centroid; then, constructing the covariance matrix and solving it using singular value decomposition (SVD) to obtain the rotation matrix and translation vector, so that the error between the point pairs is minimized; finally, repeatedly adjusting the rotation matrix and translation vector until the error rate of change is less than the error threshold or the number of iterations is greater than the number of iterations threshold.

[0048] In some embodiments, a distance threshold (e.g., >10cm) and a normal vector angle threshold (e.g., >15°) can be set. After filtering outliers in the point pair set, the optimal rigid body transformation is calculated on the remaining point pair set after removing outliers, thus achieving the filtering of incorrect matching point pairs.

[0049] In some embodiments, the federated Kalman filter framework includes a main filter and M sub-filters. The federated Kalman filter framework is used to fuse the registered M-level target map data to obtain fused map data. This may include: estimating the spatial state of map feature points in the M-level target map data one-to-one through the M sub-filters to obtain M local estimation results corresponding to the M-level target map data; the main filter receives the M local estimation results and fuses the M local estimation results to obtain fused map data.

[0050] A federated Kalman filter framework is adopted to perform multi-source information fusion on the registered M-level target map data, thereby eliminating redundancy in the multi-source map data and obtaining fused map data with higher accuracy and stronger stability. Specifically, it can include the following steps: M-level target map data are processed in parallel through M sub-filters, one-to-one correspondence. The sub-filters perform spatial state estimation on the map feature points contained in the first-level target map data to obtain the local estimation results corresponding to the target map data of that level. Each local estimation result includes at least the state estimate value and error covariance matrix output by the corresponding sub-filter.

[0051] The main filter receives M local estimation results from each sub-filter and fuses the M local estimation results based on a preset fusion strategy to obtain the globally optimal spatial state estimate as the fusion state estimate. Then, fusion map data is generated based on the fusion state estimate.

[0052] In some embodiments, corresponding to the acquisition of target map data at three accuracy levels, the sub-filter includes three corresponding sub-filters: a first sub-filter, a second sub-filter, and a third sub-filter. The first sub-filter is used to estimate the spatial state of map feature points on the target map data at the first accuracy level, obtaining a local estimation result corresponding to the target map data at the first accuracy level. The second sub-filter is used to estimate the spatial state of map feature points on the target map data at the second accuracy level, obtaining a local estimation result corresponding to the target map data at the second accuracy level. The third sub-filter is used to estimate the spatial state of map feature points on the target map data at the third accuracy level, obtaining a local estimation result corresponding to the target map data at the third accuracy level.

[0053] In some embodiments, when fusing M local estimation results based on a preset fusion strategy, the preset fusion strategy may be a covariance intersection (CI) fusion strategy or a linear weighted fusion strategy.

[0054] In some embodiments, when performing weighted fusion of M local estimation results based on a linear weighted fusion strategy, fixed weights are assigned to each of the M sub-filters, and the M local estimation results output by the M sub-filters are weighted and fused based on the fixed weights corresponding to the M sub-filters to obtain fused map data.

[0055] In some embodiments, when performing weighted fusion based on a covariance cross-validation fusion strategy, the weights are optimized using the covariance cross-validation strategy to obtain optimal weights. These optimal weights are then used to weight and fuse the M local estimation results to obtain fused map data. This achieves dynamic adjustment of the weights in the fused map data, rather than a fixed weight, thus improving the accuracy of the fused map data. For example, the optimal weight is 0.7 for centimeter-level data, 0.2 for meter-level data, and 0.1 for hundred-meter-level data.

[0056] Specifically, each sub-filter outputs a local state estimate x. i The covariance matrix P i (i=1,2,...,M); The goal of the federated Kalman filtering framework is to output a robust fused state estimate x. CI and the fusion covariance matrix P CI And guarantee P CI Positive definite. For M sub-filters (i=1,2,...,M), determine the weight optimization objective. Solve for the optimal weights through the optimization objective. There are two optimization objectives: one is to minimize the trace of the covariance matrix, and the other is to minimize the determinant of the covariance matrix. Iterate through the optimization objective to find the optimal weights. After the iteration, the optimal weights are obtained. Substitute the optimal weights into the calculation to obtain the fused covariance matrix and the fused state estimate. Verify whether the fused covariance matrix is ​​positive definite: calculate its eigenvalues; if all eigenvalues ​​are greater than 0, then it is positive definite. It can also be verified whether the fused state estimate is within a reasonable range, avoiding significant deviations.

[0057] In some embodiments, after obtaining the local state estimates x of the output of each sub-filter i The covariance matrix P i Next, redundant data from the sub-filter outputs is removed based on a data consistency threshold. This process of removing redundant data before fusion reduces the redundancy rate of the fused map data and improves data processing efficiency. This can be understood as calculating the trace of the covariance matrix output of each sub-filter, and filtering out the local state estimates x of each sub-filter whose trace of the covariance matrix is ​​greater than a preset threshold (e.g., 0.01). i The local state estimates of the outputs of each sub-filter whose traces of the covariance matrix are not greater than a preset threshold are weighted and fused based on their corresponding optimal weights to obtain a fused state estimate. The fused covariance matrix is ​​obtained by weighting and fusing the covariance matrices of each sub-filter whose traces of the covariance matrix are not greater than a preset threshold. The calculation of the fused covariance matrix and the fused state estimate based on the optimal weights can be referenced in the following formulas: ; ; in, It is the optimal weight corresponding to the i-th sub-filter. is the covariance matrix of the output of the i-th sub-filter, and N is the number of sub-filters whose trace of the covariance matrix is ​​not greater than a preset threshold. The fused covariance matrix, For fusion state estimation, It is the local state estimate of the output of the i-th sub-filter.

[0058] In some embodiments, in the step of multi-source information fusion of the registered M-level target map data, before processing the M-level target map data in parallel through M sub-filters in a one-to-one correspondence, the M-level target map data can also be time-synchronized (accuracy <10ms) and format-unified. This completes the preprocessing before filtering.

[0059] S105: Render the map based on the scene complexity index and the fused map data. The amount of detail in the rendered map is negatively correlated with the magnitude of the scene complexity index.

[0060] In complex scenarios (such as traffic density > 50 vehicles / km), traditional rendering techniques struggle to balance accuracy and frame rate, often resulting in stuttering (<20fps), impacting the real-time performance of autonomous driving decisions. In some embodiments, the step of rendering a map based on a scene complexity index and fused map data may include: determining the target resolution and target frame rate based on the scene complexity index; allocating video memory space based on the target resolution and target frame rate; and using the video memory space to render the map using the scene complexity index and fused map data.

[0061] Understandably, based on the pre-established correspondence between resolution, frame rate, and scene complexity, a target resolution and target frame rate that match the scene complexity index can be determined. The smaller the scene complexity index, the larger the determined target resolution and target frame rate, and the larger the allocated video memory space.

[0062] For example, when dynamically allocating video memory, 12GB of video memory is allocated for rendering when the scene complexity index is less than a first threshold; 6GB of video memory is allocated when the scene complexity index is not less than the first threshold and not greater than a second threshold; and 3GB of video memory is allocated when the scene complexity index is greater than the second threshold. This achieves automatic allocation of video memory based on resolution and frame rate requirements. The formula for calculating allocated video memory is: Allocated video memory = Resolution × Frame rate × Pixel format × Dynamic coefficient. For example, for 8K (7680×4320), 60fps, RGBA32 format, the allocated video memory is 7680×4320×4×1.5=195MB / frame, for a total allocation of 12GB. Allocating video memory based on resolution and frame rate requirements enables adaptive adjustment of the frame rate from 30-60fps and dynamic adjustment of video memory usage.

[0063] In some embodiments, when a geometric model in the rendered map leaves the view frustum for more than 2 seconds, the occupied video memory space is automatically released, thereby reducing the video memory release latency.

[0064] In some embodiments, the step of rendering a map using a scene complexity index and fused map data may include: if the scene complexity index is greater than a first threshold, rendering a first-level map based on the fused map data; if the scene complexity index is less than or equal to the first threshold and greater than or equal to a second threshold, rendering a second-level map based on the fused map data, wherein the rendering details of the second-level map are less than those of the first-level map, resulting in a lower precision of the second-level map than the first-level map; and if the scene complexity index is less than or equal to the second threshold, rendering a third-level map, wherein the rendering details of the third-level map are less than those of the second-level map, resulting in a lower precision of the third-level map than the second-level map.

[0065] In some embodiments, the Scene Complexity Index (SCI) is in the range of 0 to 1, and the first threshold is less than the second threshold. For example, the first threshold is in the range of 0.4 to 0.6, and the second threshold is in the range of 0.7 to 0.9. Using LOD (Level of Detail) technology, the level of detail of the geometric models of map elements in the rendered map is dynamically adjusted according to the Scene Complexity Index. For example, the first threshold is 0.5, and the second threshold is 0.8. When SCI < 0.5, image quality is prioritized, and a first-level map is rendered based on the fused map data. The first-level map retains all details of the geometric models of all map elements; the first-level map can be an 8K fully detailed map. When SCI ≤ 0.8, a balance between performance and quality is achieved, and a second-level map is rendered based on the fused map data. The second-level map retains some details of the geometric models of some or all map elements. The geometric models in the first-level map can be simplified versions of secondary geometric models of map elements, such as simplified versions of trees and streetlights. When SCI > 0.8, performance is prioritized, and only some details of key map elements are retained in the third-level map rendered from the fused map data. For example, only some details of key map elements such as lane lines and traffic signs are retained.

[0066] In some embodiments, multiple GPUs are interconnected via NVLink to achieve a GPU cluster architecture. This heterogeneous computing across multiple GPUs improves processing efficiency and supports simultaneous rendering of multiple 8K map scenes. This addresses the performance limitations of a single GPU when processing ultra-large-scale maps. Taking two GPUs as an example, the task allocation can be as follows: GPU1 handles map fusion and physics simulation (such as collision detection), while GPU2 focuses on rendering and memory management. A Round-Robin scheduling algorithm dynamically allocates computing tasks: when GPU1's load exceeds 80%, 50% of the computing tasks are automatically migrated to GPU2.

[0067] By employing a GPU cluster architecture with multiple interconnected GPUs to achieve dynamic map rendering in this embodiment of the invention, processing efficiency is improved, supporting the simultaneous rendering of four 8K scenes. When processing a scene containing 1 million models, the frame rate is increased from 25fps with a single GPU to 40fps with dual GPUs, and the response time is shortened to 300ms.

[0068] The embodiments of the present invention improve rendering performance compared to the prior art, increasing the frame rate from 20fps in the traditional solution to an adaptive 30-60fps, and improving response speed.

[0069] The experimental comparison between the map dynamic rendering method in this embodiment and the prior art is shown in Table 1 below: Table 1.

[0070] In some embodiments of the present invention, the sensor cluster used in the sensor cluster employs a silicon nitride ceramic substrate to improve sensor accuracy. A microchannel liquid cooling system is added to the GPU, using deionized water as the coolant with a flow rate of 5 L / min to reduce chip temperature. After continuous operation at 80°C for 24 hours, the temperature stabilized at 65°C without any failures. The sensor accuracy error was reduced from ±8 cm at high temperatures to ±5 cm, and the failure rate was reduced by 70%.

[0071] According to the map dynamic rendering method provided in this embodiment of the invention, the map configuration in a highway scenario can be as follows: 100-meter level map data: regional road network topology (GEOCODE = H01), including entrance / exit locations and traffic flow statistics. 1-meter level map data: road segment outline (GEOCODE = H01-R02), including the number of lanes (4 lanes) and speed limit (120 km / h). 1-centimeter level map data: lane lines (GEOCODE = H01-R02-L03), including lane width and texture.

[0072] The map dynamic rendering method provided in this embodiment of the invention can configure the map in an urban road scene as follows: 100-meter level map data: urban road network (GEOCODE = C01), including road connectivity and traffic flow hotspots. 1-meter level map data: intersections (GEOCODE = C01-I05), including traffic signs (traffic lights, speed limit signs) and pedestrian crossings. 1-centimeter level map data: manhole covers (GEOCODE = C01-I05-M12), including material and status.

[0073] Based on the same inventive concept, embodiments of the present invention provide a map dynamic rendering device, such as... Figure 2 As shown, the dynamic map rendering device includes: a data acquisition unit 201 for acquiring environmental perception data; an evaluation unit 202 for evaluating the scene complexity of the current scene based on the environmental perception data to obtain a scene complexity index; a map acquisition unit 203 for acquiring M-level target map data corresponding to the current scene from a pre-established map database; a fusion unit 204 for fusing the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3; and a rendering unit 205 for rendering the map based on the scene complexity index and the fused map data, wherein the higher the scene complexity index, the more accurate the rendered map.

[0074] In some embodiments, the map dynamic rendering apparatus further includes a construction unit for performing the following steps of the map database: constructing map data at N accuracy levels, wherein map features in each accuracy level of map data are assigned a unique geocode, and N is an integer greater than or equal to 3; and establishing a hierarchical association between the unique geocodes of map features in the N accuracy levels of map data.

[0075] In some embodiments, the N precision levels of map data include: map data of a first precision level, map data of a second precision level, and map data of a third precision level; the unique geocode of each map feature in the map data of the first precision level is associated with the unique geocode of multiple map features in the map data of the second precision level, and the unique geocode of each map feature in the map data of the second precision level is associated with the unique geocode of multiple map features in the map data of the third precision level.

[0076] In some embodiments, the rendering unit 205 is configured to: determine a target resolution and a target frame rate based on the scene complexity index; allocate video memory space based on the target resolution and the target frame rate; and use the video memory space to perform map rendering using the scene complexity index and the fused map data.

[0077] In some embodiments, the rendering unit 205 is configured to: render a first-level map based on the fused map data if the scene complexity index is greater than a first threshold; render a second-level map based on the fused map data if the scene complexity index is less than or equal to the first threshold and greater than or equal to a second threshold, wherein the rendering details of the second-level map are less than those of the first-level map; and render a third-level map if the scene complexity index is less than or equal to the second threshold, wherein the rendering details of the third-level map are less than those of the second-level map.

[0078] In some embodiments, the rendering unit 205 and the data acquisition unit 201 are used to acquire environmental perception data collected by the sensor cluster of the vehicle-road-cloud multi-terminal, wherein the environmental perception data includes image data and radar data. In some embodiments, the evaluation unit 202 is used to: detect edge density and texture complexity of the image data; determine traffic flow density based on the radar data; and fuse the traffic flow density, the edge density, and the texture complexity to obtain the scene complexity index.

[0079] In some embodiments, the fusion unit 204 includes: a registration subunit, configured to select reference map data from M-level target map data, wherein target map data at each level other than the reference map data in the M-level target map data are respectively used as map data to be registered; a fusion subunit, configured to register the map data to be registered to the coordinate system of the reference map data using an iterative nearest point algorithm based on Halanobis distance, to obtain registered M-level target map data; and to fuse the registered M-level target map data using a federated Kalman filter framework to obtain the fused map data.

[0080] In some embodiments, the registration subunit is configured to: for each point in each map data to be registered, search for the nearest neighbor point within a preset radius in the reference map data based on the Mahalanobis distance to obtain a set of point pairs; calculate the optimal rigid body transformation for the set of point pairs, and iteratively update the map data to be registered based on the optimal rigid body transformation; and terminate the update of the map data to be registered when the error change rate between two adjacent iterations is less than an error threshold or the number of iterations is greater than a number threshold.

[0081] In some embodiments, the fusion subunit is configured to: estimate the spatial state of map feature points in the M-level target map data through the M sub-filters in a one-to-one correspondence, to obtain M local estimation results corresponding to the M-level target map data; the main filter receives the M local estimation results and fuses the M local estimation results to obtain the fused map data.

[0082] The map dynamic rendering method provided in this embodiment of the invention is applied to a map dynamic rendering device. More implementation details can be found in the description of the aforementioned map dynamic rendering method embodiments. For the sake of brevity, these details will not be repeated here.

[0083] Based on the same inventive concept, embodiments of the present invention provide an electronic device. Figure 3 Schematic diagrams of electronic devices according to some embodiments of the present invention are shown. For example... Figure 3 As shown, the electronic device includes: a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302. When the processor 302 executes the program, it implements the map dynamic rendering method described in any of the above embodiments.

[0084] Among them, Figure 3In this document, a bus architecture (represented by bus 300) is used. Bus 300 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 302 and memory represented by memory 301. Bus 300 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 303 and transmitter 304. Receiver 303 and transmitter 304 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 301 can be used to store execution data used by processor 302 during operation.

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable code.

[0086] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer instructions. These computer instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer instructions may also be loaded onto a computer or other programmable execution data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0090] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for dynamic map rendering, characterized in that, include: Acquire environmental perception data; The scene complexity index is obtained by evaluating the current scene based on the environmental perception data. Obtain the M-level target map data corresponding to the current scene from a pre-established map database; The M-level target map data is fused to obtain fused map data, where M is an integer greater than or equal to 3; Map rendering is performed based on the scene complexity index and the fused map data, wherein the level of detail in the rendered map is negatively correlated with the magnitude of the scene complexity index.

2. The map dynamic rendering method as described in claim 1, characterized in that, The map database is pre-built through the following steps: Construct map data at N precision levels, where each precision level of map data assigns a unique geocode to the map features, and N is an integer greater than or equal to 3. A hierarchical association is established between the unique geocodes of map features in the N precision levels of map data.

3. The map dynamic rendering method as described in claim 2, characterized in that, The map data of the N precision levels include: map data of the first precision level, map data of the second precision level, and map data of the third precision level. The unique geocode of each map feature in the map data of the first accuracy level is associated with the unique geocode of multiple map features in the map data of the second accuracy level, and the unique geocode of each map feature in the map data of the second accuracy level is associated with the unique geocode of multiple map features in the map data of the third accuracy level.

4. The method as described in claim 1, characterized in that, The step of rendering the map based on the scene complexity index and the fused map data includes: The target resolution and target frame rate are determined based on the scene complexity index. Allocate video memory space according to the target resolution and target frame rate; Using the aforementioned video memory space, map rendering is performed using the aforementioned scene complexity index and the aforementioned fused map data.

5. The method as described in claim 4, characterized in that, The process of rendering a map using the scene complexity index and the fused map data includes: If the scene complexity index is greater than the first threshold, a first-level map is rendered based on the fused map data; If the scene complexity index is less than or equal to the first threshold and greater than or equal to the second threshold, a second-level map is rendered based on the fused map data, and the rendering details of the second-level map are less than those of the first-level map. If the scene complexity index is less than or equal to the second threshold, a third-level map is rendered, and the rendering details of the third-level map are less than those of the second-level map.

6. The method as described in claim 1, characterized in that, The acquisition of environmental perception data includes: acquiring environmental perception data collected by a sensor cluster of vehicle-road-cloud multi-terminals, wherein the environmental perception data includes image data and radar data; Based on the environmental perception data, the current scene complexity is assessed, including: The image data is then analyzed to detect edge density and texture complexity. Traffic density is determined based on the radar data; The scene complexity index is obtained by fusing the traffic density, the edge density, and the texture complexity.

7. The method according to any one of claims 1-6, characterized in that, The process of fusing the M-level target map data to obtain fused map data includes: Reference map data is selected from the M-level target map data, and the target map data at each level other than the reference map data in the M-level target map data are respectively used as the map data to be registered; The iterative nearest point algorithm based on Halanobis distance is used to register the map data to be registered to the coordinate system of the reference map data to obtain the registered M-level target map data; The registered M-level target map data is fused using a federated Kalman filter framework to obtain the fused map data.

8. The method as described in claim 7, characterized in that, The step of using an iterative nearest-point algorithm based on Halanobis distance to register the map data to be registered at each level to the coordinate system of the reference map data includes: For each point in each map data to be registered, the nearest neighbor point within a preset radius is searched in the reference map data based on the Mahalanobis distance to obtain a set of point pairs. Calculate the optimal rigid body transformation for the set of point pairs, and iteratively update the map data to be registered based on the optimal rigid body transformation; When the error change rate between two consecutive iterations is less than the error threshold or the number of iterations is greater than the number threshold, the update of the map data to be registered ends. The federated Kalman filter framework includes a main filter and M sub-filters. The federated Kalman filter framework is used to fuse the registered M-level target map data to obtain the fused map data, which includes: By estimating the spatial state of map feature points of the M-level target map data one-to-one through the M sub-filters, M local estimation results corresponding to the M-level target map data are obtained. The main filter receives the M local estimation results and fuses them to obtain the fused map data.

9. A map dynamic rendering device, characterized in that, include: The data acquisition unit is used to acquire environmental perception data. An evaluation unit is used to evaluate the scene complexity of the current scene based on the environmental perception data, and obtain a scene complexity index. The map acquisition unit is used to acquire M-level target map data corresponding to the current scene from a pre-established map database; The fusion unit is used to fuse the M-level target map data to obtain fused map data, where M is an integer greater than or equal to 3; The rendering unit is used to render a map based on the scene complexity index and the fused map data, wherein the level of detail in the rendered map is negatively correlated with the magnitude of the scene complexity index.

10. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the map dynamic rendering method according to any one of claims 1-8.

Images