A map data collection swarm perception participant recruitment method and related system

Abstract

The application provides a map data collection crowd-sensing participant recruitment method and a related system, and relates to the technical field of intelligent transportation. The application combines a vehicle dynamic sensing utility evaluation mechanism and a vehicle historical task reputation reuse mechanism under an edge computing framework, recruits intelligent vehicles with sufficient sensing utility intensity and good reputation for collaborative sensing mapping according to actual map defect areas existing in a target map construction area responsible by each edge node on demand, so as to effectively enhance the timeliness, accuracy and system robustness of map data collection operations in a complex dynamic traffic environment, improve the overall timeliness, overall efficiency and map quality of crowd-sensing collaborative mapping operations, and further alleviate problems such as local area sensing redundancy, local area sensing deficiency and serious mapping communication delay existing in existing crowd-sensing collaborative mapping schemes.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and more specifically, to a method and system for recruiting participants in map data collection and crowdsensing. Background Technology

[0002] With the rapid development of intelligent transportation and autonomous driving technologies, high-precision maps have become a core infrastructure supporting autonomous driving systems in achieving environmental perception, precise positioning, and path planning. Traditional high-precision map construction methods mainly rely on high-cost sensor equipment (such as LiDAR and industrial-grade inertial navigation systems) mounted on professional surveying vehicles for centralized mapping. However, this centralized mapping model is inherently costly, has a long deployment cycle, and a low map update frequency, making it difficult to cope with dynamic scenarios such as urban road construction and temporary traffic control. It can no longer meet the multiple requirements of intelligent driving for maps that are "highly timely, highly covered, and highly accurate."

[0003] To reduce mapping costs and improve map update timeliness, the industry has begun exploring online mapping technologies based on sensors from mass-produced vehicles. These technologies utilize onboard sensors such as cameras and millimeter-wave radars in mass-produced vehicles to achieve real-time extraction of road features and incremental map updates. However, single-vehicle perception is often limited by factors such as sensor deployment range, obstacle obstruction, and environmental complexity, resulting in problems such as limited field of view and blind spots. Therefore, it is necessary to introduce crowdsourcing concepts to achieve collaborative perception among multiple vehicles, thereby realizing a highly covered, redundant, and fault-tolerant collective intelligent map construction effect.

[0004] Currently, crowdsourcing-based collaborative perception mapping solutions mainly rely on centralized cloud scheduling mechanisms. This requires a cloud server to coordinate the selection of vehicles participating in map data perception tasks. Each participant uploads and aggregates their traffic information perception data to the cloud server, which then updates the high-precision map based on this massive amount of feedback data. This approach suffers from communication delays, competition for computing resources, and low efficiency in real-time perception task allocation when dealing with large-scale vehicle collaboration, severely impacting the timeliness of collaborative mapping. Furthermore, existing collaborative perception mapping solutions primarily rely on static traffic flow distribution characteristics (e.g., vehicle location density, vehicle GPS trajectory quality) for participant selection, failing to consider the dynamic perception performance of individual vehicles at different task locations, times, and environments. This easily leads to perception redundancy in some areas and perception gaps in others, severely affecting the efficiency and quality of collaborative mapping. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method for recruiting participants in the collaborative perception of map data collection and a map building system. Under the edge computing framework, this system combines a vehicle dynamic perception utility evaluation mechanism (which is used to evaluate the dynamic perception performance of intelligent vehicles in the corresponding edge node mapping area) and a vehicle historical task reputation reuse mechanism. For map defect areas existing in the edge node mapping area, suitable intelligent vehicles are recruited as needed for collaborative perception mapping, thereby effectively improving the overall timeliness, efficiency and map quality of the collaborative perception mapping operation, and alleviating the problems of local area perception redundancy, local area perception deficiency and serious mapping communication delay in existing collaborative perception mapping schemes.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0007] In a first aspect, this application provides a method for recruiting participants in a map data acquisition crowdsensing system, applied to each edge node in a map building system, wherein each edge node is solely responsible for a target map building area. The recruitment method includes:

[0008] Obtain vehicle perception performance information and vehicle motion status information periodically uploaded by each intelligent vehicle within the target map construction area, and query the historical task reputation data of each intelligent vehicle.

[0009] Based on the vehicle perception performance information and vehicle motion state information of each intelligent vehicle, the perception utility score data of each intelligent vehicle in the target map construction area is calculated.

[0010] Based on the actual map integrity matrix of the target map construction area, determine the actual map defect area in the target map construction area that corresponds to the current map perception task;

[0011] For each intelligent vehicle, based on the actual map completeness matrix, as well as the intelligent vehicle's historical mission reputation data and perception utility score data, the perception adaptation state tensor of the intelligent vehicle in the target map construction area is calculated.

[0012] The task participant recruitment model is invoked to perform vehicle recruitment evaluation processing based on the perception adaptation state tensor of each intelligent vehicle and the actual map defect area, so as to obtain the target participating vehicles that match the current map perception task.

[0013] In an optional implementation, the recruitment method further includes:

[0014] Obtain the actual perception images uploaded by each of the target participating vehicles during the current map perception task execution, and extract BEV (Bird's Eye View) features from the actual perception images of each of the target participating vehicles to obtain the BEV perception feature tensor of each of the target participating vehicles.

[0015] Obtain a standard map image of the target map construction area, and extract multi-scale spatial features from the standard map image to obtain the multi-scale spatial feature tensor of the standard map image;

[0016] For each target participating vehicle, the multi-scale spatial feature tensor and the BEV perception feature tensor of the target participating vehicle are aligned and stitched together by feature dimensions, and the fused feature tensor obtained by stitching is subjected to semantic segmentation and decoding processing to obtain the predicted semantic segmentation image of the target participating vehicle in the target map construction area.

[0017] A non-maximum suppression mechanism is used to perform grid alignment and fusion on the predicted semantic segmentation images of all the target participating vehicles to obtain a local semantic map for the actual map defect area.

[0018] The map completeness matrix of the target map construction area is updated based on the local semantic map.

[0019] In an optional implementation, if the map building system includes edge nodes that share the same vehicle's semantic segmentation image, perception data upload status, and task participation reputation value for any map perception task, then the recruitment method further includes:

[0020] For each target vehicle, the actual map construction quality index value of the target vehicle in the current map perception task is calculated based on the local semantic map, the predicted semantic segmentation image of the target vehicle, and the historical task semantic segmentation image of the target vehicle.

[0021] Based on the perception data upload status of the target vehicle in the current map perception task and the historical task data upload status of the target vehicle, calculate the actual data upload stability index value of the target vehicle in the current map perception task.

[0022] Based on the actual map construction quality index and actual data upload stability index of the target vehicle, as well as the historical map construction quality index and historical data upload stability index of the target vehicle, the initial reputation value of the target vehicle in the current map perception task is calculated using the entropy weight method.

[0023] Detect whether the vehicle recruitment probability of the target participating vehicle in the target map construction area exceeds a preset probability threshold;

[0024] If the vehicle recruitment probability is detected to exceed the preset probability threshold, the initial reputation value is corrected by using the upper confidence bound algorithm based on the task participation status of the target participating vehicle in the target map construction area to obtain the task participation reputation value of the target participating vehicle in the current map perception task; otherwise, the initial reputation value is directly used as the task participation reputation value of the target participating vehicle in the current map perception task.

[0025] Secondly, this application provides a map building system, which includes a cloud server and multiple edge nodes, wherein the multiple edge nodes are respectively connected to the cloud server to realize data sharing between the multiple edge nodes through the cloud server, and each edge node is solely responsible for realizing the map building function of a target map building area.

[0026] Each edge node stores a computer program, and by running the computer program, the map data collection crowd sensing participant recruitment method described in any of the foregoing embodiments can be implemented.

[0027] In this case, the beneficial effects of the embodiments of this application may include the following:

[0028] This application targets edge nodes in a map building system implemented using an edge computing framework. It calculates the perception utility score of each intelligent vehicle within the target map building area by acquiring periodically uploaded vehicle perception performance and motion state information from intelligent vehicles located within that edge node's responsibility. Then, based on the actual map completeness matrix of the target map building area, it determines the actual map defect areas corresponding to the current map perception task. Next, based on the aforementioned actual map completeness matrix, along with the historical task reputation data and perception utility score of each intelligent vehicle, it calculates the perception adaptation state tensor of each intelligent vehicle within the target map building area. Finally, it invokes a task participant recruitment model based on the perception adaptation state of each intelligent vehicle. Tensor and the aforementioned actual map defect areas are used to evaluate vehicle recruitment, resulting in target vehicles that match the current map perception task. This allows for the recruitment of suitable intelligent vehicles for collaborative perception mapping within the edge computing framework, combining a vehicle dynamic perception utility evaluation mechanism and a vehicle historical task reputation reuse mechanism. This addresses actual map defect areas within the edge node mapping area (i.e., the target map construction area responsible for the corresponding edge node). This effectively enhances the timeliness, accuracy, and robustness of map data collection operations in complex dynamic traffic environments, improves the overall timeliness, efficiency, and map quality of collaborative perception mapping operations, and further alleviates problems such as local area perception redundancy, local area perception deficiency, and severe mapping communication delays in existing collaborative perception mapping schemes.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the system composition of the map building system provided in the embodiments of this application;

[0032] Figure 2 One of the flowcharts illustrating the map data collection and crowd-sensing participant recruitment method provided in this application embodiment;

[0033] Figure 3 for Figure 2 A flowchart illustrating the sub-steps included in step S220;

[0034] Figure 4 for Figure 2 A flowchart illustrating the sub-steps included in step S240;

[0035] Figure 5 A second schematic flowchart illustrating the map data collection and crowd-sensing participant recruitment method provided in this application embodiment;

[0036] Figure 6 This is the third flowchart illustrating the map data collection and crowdsensing participant recruitment method provided in this application embodiment.

[0037] Icons: 10 - Map building system; 11 - Edge node; 12 - Cloud server. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0039] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0040] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0041] In the description of this application, it should be understood that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are used only for the convenience of describing this application and simplifying the description, and are not intended to indicate or imply that the equipment or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0042] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0043] Furthermore, it is understood in the description of this application that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Those skilled in the art will understand the specific meaning of the above terms in this application based on the specific circumstances.

[0044] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0045] Please refer to Figure 1 , Figure 1 This is a schematic diagram of the system composition of the map building system 10 provided in this application embodiment. In this application embodiment, the map building system 10 is implemented using an edge computing framework. The map building system 10 may include a cloud server 12 and multiple edge nodes 11, wherein each edge node 11 is solely responsible for implementing the map building function of a target map building area. By utilizing edge computing technology, which has the advantages of localization, proximity, and low latency, the heavy computing tasks (which involve map perception building operations, perception task creation operations, perception task allocation operations, etc.) at the cloud server in the prior art are offloaded to each edge node 11, thereby reducing the burden of data transmission and storage and improving the timeliness of collaborative perception mapping.

[0046] In this embodiment, the multiple edge nodes 11 are respectively connected to the cloud server 12 to realize data sharing among the multiple edge nodes 11 through the cloud server 12. This allows each edge node 11 to know the task execution status of one or more intelligent vehicles at other edge nodes 11 through the cloud server 12 (including the upload status of the perception data of the corresponding intelligent vehicle for the map perception task it participates in (e.g., whether the intelligent vehicle has successfully uploaded the traffic environment perception data of the corresponding map perception task to the task publishing edge node) and the task participation reputation value, as well as the semantic segmentation image constructed based on the successfully uploaded traffic environment perception data). Based on the known task execution status, the edge node 11 selects suitable vehicles for perception tasks in map defect areas from among the intelligent vehicles within its target map construction area. This improves the timeliness, accuracy, and robustness of map data collection in the corresponding map defect areas, thereby improving the overall timeliness, efficiency, and map quality of the collaborative perception mapping operation and alleviating the problems of local area perception redundancy, local area perception deficiency, and severe mapping communication delay in existing collaborative perception mapping schemes.

[0047] In this embodiment, each edge node 11 of the map building system 10 pre-stores a specific computer program related to the recruitment function of intelligent sensing participants for map data acquisition. By running and executing the specific computer program, the system combines a vehicle dynamic perception utility evaluation mechanism and a vehicle historical task reputation reuse mechanism within the edge computing framework. For map defect areas that actually exist within its target map building area, the system recruits suitable intelligent vehicles (i.e., intelligent vehicles with sufficient perception utility and good reputation) as needed to conduct collaborative perception mapping. This effectively improves the overall timeliness, efficiency, and map quality of the collaborative perception mapping operation, and alleviates the problems of local area perception redundancy, local area perception deficiency, and severe mapping communication delay in existing collaborative perception mapping schemes.

[0048] Understandable, Figure 1 The block diagram shown is only a schematic diagram of one component of the map building system 10. The map building system 10 may also include components such as... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented using hardware, software, or a combination thereof.

[0049] In this application, to ensure that each edge node 11 in the map building system 10 can independently recruit intelligent vehicles with sufficient perception utility and good reputation to conduct collaborative perception mapping for actual map defect areas in the target map building area it is responsible for, thereby effectively enhancing the timeliness, accuracy, and system robustness of map data collection operations in complex dynamic traffic environments, and improving the overall timeliness, efficiency, and map quality of collaborative perception mapping operations, this application embodiment achieves the aforementioned objective by providing a method for recruiting participants in collaborative perception mapping for map data collection. The method for recruiting participants in collaborative perception mapping for map data collection provided in this application will be described in detail below.

[0050] Please refer to Figure 2 , Figure 2 This is one of the flowcharts illustrating the map data collection and crowdsensing participant recruitment method provided in this application embodiment. In this application embodiment, Figure 2 The map data collection and crowdsensing participant recruitment method shown may include steps S210 to S250.

[0051] Step S210: Obtain the vehicle perception performance information and vehicle motion status information periodically uploaded by each intelligent vehicle within the target map construction area, and query the historical task reputation data of each intelligent vehicle.

[0052] In this embodiment, for any edge node 11, when a smart vehicle is within the target map construction area under the responsibility of that edge node 11, the smart vehicle periodically sends its own vehicle status information (i.e., vehicle perception performance information and vehicle motion status information) to the edge node 11 so that the corresponding edge node 11 can confirm the smart vehicle's dynamic perception performance of the traffic environment within the target map construction area. The vehicle perception performance information of a single smart vehicle includes its effective perception length range and maximum perception field of view, while the vehicle motion status information includes the smart vehicle's actual position, driving direction angle, and size within the corresponding target map construction area.

[0053] When any edge node 11 receives vehicle status information uploaded by a smart vehicle, it will query the smart vehicle's historical task reputation data (including the task participation reputation value of each of the smart vehicle's historical perception tasks (not limited to map perception tasks published by the edge node 11, but also map perception tasks published by other edge nodes 11) based on the multi-node data sharing mechanism) in order to improve the reliability and accuracy of subsequent vehicle recruitment operations by utilizing the smart vehicle's historical task reputation data.

[0054] Step S220: Calculate the perception utility score data of each intelligent vehicle in the target map construction area based on the vehicle perception performance information and vehicle motion state information of each intelligent vehicle.

[0055] In this embodiment, when an edge node 11 receives vehicle status information uploaded by any intelligent vehicle, it will evaluate the intelligent vehicle's dynamic perception performance in the real traffic environment within the target map construction area based on the vehicle perception performance information and vehicle motion status information of the intelligent vehicle in the target map construction area based on the vehicle dynamic perception utility evaluation mechanism, and obtain the corresponding perception utility score data.

[0056] Alternatively, please refer to Figure 3 , Figure 3 yes Figure 2 A flowchart illustrating the sub-steps included in step S220. In this embodiment, the target map construction area corresponding to any edge node 11 can be constructed according to... OK The column grid layout is rasterized to divide the target map construction area into regions. In the target map construction area, the perception utility score data of any intelligent vehicle includes the actual perception utility score of the intelligent vehicle for each grid area in the target map construction area. The above step S220 may include sub-steps S221 to S223 to effectively evaluate the actual perception contribution of any intelligent vehicle to different grid areas in the corresponding target map construction area (i.e., the dynamic perception performance of the traffic environment).

[0057] Sub-step S221: For each intelligent vehicle, based on the actual vehicle position and size of other intelligent vehicles, perform perception field occlusion detection on the intelligent vehicle within the target map construction area to obtain the occlusion mask function of the intelligent vehicle for different grid areas within the target map construction area.

[0058] In this embodiment, the first in the target map construction area Line number The grid area of ​​the column in the first The perception occlusion constraint that a smart vehicle must satisfy when its full field of vision is occluded by other smart vehicles is expressed by the following formula:

[0059] and / or ;

[0060] in, Used to indicate the first The actual location of each intelligent vehicle. Used to indicate the first Line number The grid position of the column's grid area; , , and They are used to represent the number excluding the first. The position of the vertex of the fitted rectangle for any intelligent vehicle other than the three intelligent vehicles in the bird's-eye view. and They lie on the same diagonal of the fitted rectangle. and They lie on the diagonal of the same fitted rectangle.

[0061] Based on this, the first [unclear] within the target map construction area The intelligent vehicle is targeting the first Line number The occlusion mask function for the grid area of ​​a column is represented by the following function:

[0062] ;

[0063] in, Used to represent the first in the target map construction area Line number The grid area of ​​the column; Used to represent the first A smart vehicle targets the grid area. The occlusion mask function can be used to describe the first... A smart vehicle, under the geometric occlusion of other smart vehicles, accesses a grid area. The visibility of the occlusion mask function expression, where "0" in the expression represents the corresponding grid area. In the This can be seen in intelligent vehicles, where "1" is used to represent the corresponding grid area. In the The intelligent vehicle is not visible.

[0064] Sub-step S222: Based on the effective perception length range, maximum perception field of view, actual vehicle position, and vehicle driving direction angle of the intelligent vehicle, perform perception performance attenuation detection on the intelligent vehicle within the target map construction area to obtain the perception attenuation function of the intelligent vehicle for different grid areas.

[0065] In this embodiment, the first [item] within the target map construction area For an intelligent vehicle, its theoretical perception grid area set in the target map construction area It can be represented as " ",in Used to indicate the first The vehicle's driving direction angle for an intelligent vehicle. Used to indicate the first The driving direction of an intelligent vehicle and the grid area The absolute value of the angle difference between them Used to represent the first The actual vehicle location of each intelligent vehicle is located within the grid area. The straight-line distance Used to represent the first The lower limit of the effective sensing length range for an intelligent vehicle. Used to represent the first The upper limit of the effective sensing length range for an intelligent vehicle. Used to represent the first The maximum field of view for a smart vehicle.

[0066] Based on this, considering that the sensing quality of the vehicle-mounted sensor changes with sensing distance and / or sensing direction, the first [unclear] within the target map construction area The intelligent vehicle is targeting the first Line number The perceptual attenuation function for the grid area of ​​a column is expressed by the following function:

[0067] ;

[0068] in, Used to indicate the first A smart vehicle targets the grid area. The perceived decay function, Used to indicate the first A smart vehicle targets the grid area. The optimal perceived quality value, Used to represent the attenuation coefficient Used to represent the natural exponential function.

[0069] Sub-step S223: Based on the mapping importance coefficient and actual grid area of ​​each grid area within the target map construction area, as well as the occlusion masking function and perception attenuation function of the intelligent vehicle for different grid areas, calculate the actual perception utility score of the intelligent vehicle for each grid area.

[0070] In this embodiment, considering that different grid regions within the same target map construction area have varying importance during the mapping process (for example, grid regions involving lanes are more important than those involving sidewalks), each grid region within the target map construction area is assigned a mapping importance coefficient. Based on this, the [missing information - likely a specific grid region within the target map construction area]... The intelligent vehicle is targeting the first Line number Column grid area The actual perceived utility score is calculated using the following formula:

[0071] ;

[0072] in, Used to indicate the first A smart vehicle targets the grid area. The actual perceived utility score, Used to indicate the first A smart vehicle targets the grid area. The perceived decay function, Used to represent the first A smart vehicle targets the grid area. The occlusion mask function, Used to represent a grid area The mapping importance coefficient at the target map construction area. Used to represent a grid area The actual grid area in the target map construction area. The higher the actual perception utility score of a single intelligent vehicle for a given grid area, the higher the degree of its real perception contribution to that grid area, the better its dynamic perception of the traffic environment in that grid area, and the more suitable it is to be a perception participant in that grid area.

[0073] Therefore, by executing the above sub-steps S221 to S223, this application can effectively evaluate the degree of real perception contribution (i.e., dynamic perception performance of traffic environment) of any intelligent vehicle to different grid areas within the corresponding target map construction area.

[0074] Step S230: Based on the actual map integrity matrix of the target map construction area, determine the actual map defect area in the target map construction area that corresponds to the current map perception task.

[0075] In this embodiment, for any target map construction area of ​​edge node 11, the concept of a map integrity matrix (whose matrix element layout is consistent with the grid layout of the corresponding target map construction area) can be introduced to describe the map integrity of each grid area within the corresponding target map construction area at the end of a certain map perception task. This map integrity matrix can be represented as " " Used to indicate the map perception task release sequence number at the corresponding edge node 11. Used to represent a grid area In the Raster semantic map at the end of a map perception task True values ​​of raster standard maps The intersection-union ratio (i.e., the completeness of the map) between them.

[0076] Based on this, when an edge node 11 needs to issue the current map perception task (its perception task issuance sequence number can be ""), When represented by "", the corresponding actual map completeness matrix is ​​the previous historical map perception task of the edge node 11 in the current map perception task (its perception task release sequence number can be represented by ""). (This is represented as) the map integrity matrix at the end of the execution. The edge node 11 can filter the target raster regions in the actual map integrity matrix whose map integrity is less than a preset integrity threshold, and use them as a defect raster region in the actual map defect region of the target map construction region under the current map perception task, thereby obtaining the defect raster region set of the actual map defect region.

[0077] Step S230: For each intelligent vehicle, calculate the perception adaptation state tensor of the intelligent vehicle in the target map construction area based on the actual map integrity matrix, the intelligent vehicle's historical task reputation data and perception utility score data.

[0078] In this embodiment, for any intelligent vehicle located within the target map construction area of ​​a certain edge node 11, the actual map completeness matrix of the edge node 11 when issuing the current map perception task, the current perception utility score data of the intelligent vehicle, and the historical task reputation data of the intelligent vehicle before the current map perception task can be used to organically combine the vehicle dynamic perception utility evaluation mechanism and the vehicle historical task reputation reuse mechanism based on the joint coding principle of multimodal features. This allows the calculated perception adaptation state tensor to accurately describe the perception adaptation feature status of the intelligent vehicle within the target map construction area in multiple dimensions.

[0079] Alternatively, please refer to Figure 4 , Figure 4 yes Figure 2 A flowchart illustrating the sub-steps included in step S240. In this embodiment, step S240 may include sub-steps S241 to S244 to accurately describe the perception adaptation features of the intelligent vehicle within the target map construction area from multiple dimensions.

[0080] Sub-step S241: Based on the occlusion masking function, perception attenuation function, actual perception utility score and actual relative distance of the intelligent vehicle in the target map construction area for different grid regions, calculate the spatial perception intensity embedding vector of the intelligent vehicle in the target map construction area.

[0081] In this embodiment, spatial distribution image construction processing can be performed on the actual perception utility score, occlusion mask function, and perception attenuation function of the intelligent vehicle for different grid regions within the target map construction area. This yields: a grid perception utility distribution image (each pixel in this image corresponds to a single grid region within the target map construction area, and the actual content of each pixel is represented by the actual perception utility score of the corresponding grid region), a grid occlusion distribution image (each pixel in this image corresponds to a single grid region within the target map construction area, and the actual content of each pixel is represented by the occlusion mask function value of the corresponding grid region), and a grid perception attenuation distribution image (each pixel in this image corresponds to a single grid region within the target map construction area, and the actual content of each pixel is represented by the occlusion mask function value of the corresponding grid region). Each location corresponds to a single grid region within the target map construction area. The actual content of each pixel location in this image is represented by the perception attenuation function value of the corresponding grid region. Then, distance-normalized images are constructed by performing distance-normalized image processing on the actual relative distances from the intelligent vehicle to different grid regions. This yields a relative spacing distribution image of the intelligent vehicle within the target map construction area (each pixel location in this image corresponds to a single grid region within the target map construction area. The actual content of each pixel location in this image is represented by the distance-normalized value of the actual relative distance of the corresponding grid region). Finally, by stitching together the grid perception utility distribution image, grid occlusion distribution image, grid perception attenuation distribution image, and relative spacing distribution image corresponding to the intelligent vehicle, a corresponding target spatial perception intensity map is obtained. At this point, a preset lightweight convolutional neural network (which is an embedding coding network composed of multiple convolutional layers, pooling layers and activation functions) can be invoked to extract high-dimensional embedding features from the target spatial perception intensity map, thereby obtaining the corresponding high-dimensional embedding feature tensor. The extracted high-dimensional embedding feature tensor can then be flattened to obtain the spatial perception intensity embedding vector of the intelligent vehicle within the target map construction area.

[0082] Sub-step S242 calls the gated loop unit to perform hidden state temporal encoding on the task participation reputation values ​​of all historical perception tasks involved in the intelligent vehicle, so as to obtain the time-aware reputation embedding vector of the intelligent vehicle in the target map construction area.

[0083] In this embodiment, the historical task reputation sequence of the intelligent vehicle can be obtained by constructing a sequence of task participation reputation values ​​of all historical perception tasks involved in the intelligent vehicle according to the task execution time sequence. Then, the historical task reputation sequence is input into a gated loop unit for hidden state encoding to obtain a hidden state sequence that is consistent with the task execution time sequence of the historical task reputation sequence. Finally, the last hidden state tensor in the hidden state sequence is selected as the time perception reputation embedding vector of the intelligent vehicle.

[0084] Sub-step S243: Based on the positional coupling relationship between the intelligent vehicle and the target map construction area, a standard self-attention mechanism is used to learn the spatial attention of the defect area based on the actual map integrity matrix, so as to obtain the spatial attention vector of the intelligent vehicle in the target map construction area.

[0085] In this embodiment, the actual vehicle position of the intelligent vehicle can be coupled and associated with the target map construction area through location encoding to obtain the positional coupling relationship between the intelligent vehicle and the target map construction area. This positional coupling relationship can then be represented by the following functional expression:

[0086] ;

[0087] in, , It is an integer. Used to indicate the first The actual location of each intelligent vehicle. Used to indicate the first The location encoding vector of each intelligent vehicle within the target map construction area.

[0088] Sub-step S244 involves concatenating the spatial perception intensity embedding vector, temporal perception reputation embedding vector, and defect area spatial attention vector corresponding to the intelligent vehicle to obtain the perception adaptation state tensor of the intelligent vehicle in the target map construction area.

[0089] Therefore, by executing the above sub-steps S241 to S244, this application can accurately describe the perception and adaptation characteristics of a single intelligent vehicle within the corresponding target map construction area in multiple dimensions.

[0090] Step S250: The task participant recruitment model is invoked to perform vehicle recruitment evaluation processing based on the perception adaptation state tensor of each intelligent vehicle and the actual map defect area, so as to obtain the target participating vehicles that match the current map perception task.

[0091] In this embodiment, the task participant recruitment model is implemented using a dual-deep Q-network architecture with an introduced noisy network. This involves a main network with an introduced Gaussian noise network and a target network. The main network of the task participant recruitment model can be invoked, and it utilizes a neural network policy function carrying Gaussian noise. The initial Q-value prediction for each intelligent vehicle can be expressed as " ",in Used to represent neural network policy functions Network parameters, Used to indicate the first A smart vehicle is performing a perception task on the current map. Pre-execution perception and adaptation state tensor Used to indicate the first A smart vehicle is performing a perception task on the current map. The scheduling behavior during the execution process (which may be to recruit the intelligent vehicle to participate in the current map perception task, or not to recruit the intelligent vehicle to participate in the current map perception task). Used to indicate the first The initial Q-value prediction result of an intelligent vehicle driven according to scheduling behavior. Used to represent the introduced noise vector, This is used to represent the mathematical expectation function. In this process, the neural network policy function can employ a confrontational architecture to separate the state value function and the advantage function, and then perform weighted fusion calculations at the output layer to determine the performance of each intelligent vehicle. Action effectiveness in a state .

[0092] Then, the target network of the task participant recruitment model is invoked using the frozen target network parameters. Perform target Q-value prediction to obtain the performance of each intelligent vehicle in the current map perception task. The target Q-value prediction result is given below. The target Q-value prediction of the target network can be expressed as " ",in Used to indicate the first The target Q-value prediction result of an intelligent vehicle under the current map perception task. Used to represent discount factors, The Q-value prediction function used to represent the target network. Used to indicate the first A smart vehicle is performing a perception task on the current map. The perception adaptation state tensor after execution. Used to represent the first A smart vehicle Used to represent the first The task of intelligent vehicles with current map perception. The corresponding instant reward value.

[0093] During this process, the first [item] within the target map construction area The instantaneous reward value of each intelligent vehicle at the task participant recruitment model corresponding to the current map perception task is calculated using the following formula:

[0094] ;

[0095] in, Used to represent the first A smart vehicle Used to represent the first The task of intelligent vehicles with current map perception. The corresponding instant reward value, , and They are respectively , and Their respective weighting coefficients Used to represent the first A smart vehicle performs a task of perceiving the current map. The perceived coverage of the corresponding actual map defect area. Used to indicate the actual map defect area in the [number]th Perceptual redundancy factors at the location of each intelligent vehicle. Used to represent the first A smart vehicle is performing a perception task on the current map. The task participation reputation value of the most recent historical perception task (i.e., the aforementioned first-most recent task participation reputation value) (The reputation value of the last task in the historical mission reputation sequence of an intelligent vehicle). A set of defect raster regions used to represent the actual map defect area. Used to represent the first The theoretical perception grid area set of an intelligent vehicle in the target map construction area. Used to represent the first in the target map construction area Line number The grid area of ​​the column, Used to represent the theoretically covered sensing grid area The total number of vehicles, Functions used to calculate the number of elements in a set.

[0096] After the task participant recruitment model is invoked to determine the target Q-value prediction results of each intelligent vehicle under the current map perception task, the target Q-value prediction results of each intelligent vehicle are sorted in descending order, and a preset number of the top-ranked vehicles are selected (which can be done using " Each intelligent vehicle (represented by "") serves as a target participating vehicle for the current map perception task issued by the edge node 11. This allows for the recruitment of intelligent vehicles with sufficient perception utility and good reputation to collaboratively perceive and map areas within the area of ​​actual map defects within the responsibility of the edge node 11. This effectively enhances the timeliness, accuracy, and system robustness of map data collection operations in complex and dynamic traffic environments, improves the overall timeliness, efficiency, and map quality of collaborative perception and mapping operations, and alleviates problems such as localized perception redundancy, localized perception deficiencies, and severe mapping communication delays in existing collaborative perception and mapping schemes.

[0097] Therefore, by executing the above steps S210 to S250, this application can ensure that each edge node 11 in the map building system 10 can independently recruit intelligent vehicles with sufficient perception utility and good reputation to conduct collaborative perception mapping for the actual map defect areas existing in the target map building area it is responsible for. This effectively enhances the timeliness, accuracy and system robustness of map data collection operations in complex dynamic traffic environments, and improves the overall timeliness, overall efficiency and map quality of the collaborative perception mapping operation.

[0098] Alternatively, please refer to Figure 5 , Figure 5 This is the second flowchart illustrating the map data collection and crowdsensing participant recruitment method provided in this application embodiment. In this application embodiment, [the method is related to...]. Figure 2 Compared to the map data collection and crowdsensing participant recruitment methods shown, Figure 5 The map data acquisition crowd perception participant recruitment method shown may also include steps S310 to S350, to drive each intelligent vehicle recruited for the current map perception task to perform collaborative perception and mapping within the corresponding target map construction area, thereby realizing local map update operations for the actual map defect area that matches the current map perception task.

[0099] Step S310: Obtain the actual perception images uploaded by each of the target participating vehicles during the current map perception task execution, and extract BEV features from the actual perception images of each of the target participating vehicles to obtain the BEV perception feature tensors of each of the target participating vehicles.

[0100] In this embodiment, when an edge node 11 determines all the target vehicles participating in the current map perception task, it sends a traffic environment collection instruction to each of the target vehicles, so that each target vehicle can use its own on-board sensors (such as a front-facing camera) to collect the road image in front of the vehicle (i.e., the actual perception image) and upload it to the edge node 11 along with the timestamp and image collection location information. When the edge node 11 receives an actual perception image uploaded by any target participating vehicle, it calls a lightweight backbone network (e.g., a ResNet-18 network) to extract the POV view feature map of the actual perception image from the point of view. Then, it converts the POV view feature map into a BEV view feature map through an inverse perspective transformation operation, obtaining multiple initial reference points in the BEV view feature map. Subsequently, it uses a standard self-attention mechanism and residual and normalization layers to transform the BEV view feature map into a BEV adaptation feature map suitable for input into the cross-attention module. Then, by inputting the BEV adaptation feature map, the POV view feature map, and the multiple initial reference points into the cross-attention module, the cross-attention module performs BEV feature extraction, thereby obtaining the BEV perception feature tensor of the corresponding target participating vehicle.

[0101] Step S320: Obtain the standard map image of the target map construction area, and extract multi-scale spatial features from the standard map image to obtain the multi-scale spatial feature tensor of the standard map image.

[0102] In this embodiment, each edge node 11 can read the standard map data of the target map construction area it is responsible for from the cloud server 12 or the local database, parse it into a two-dimensional image (i.e., the standard map image mentioned above), and then perform grayscale normalization and noise filtering on it. Subsequently, the Canny edge detection operator is used to extract key edge contours to obtain the corresponding binary contour image. Then, the binary contour image is input into a pre-trained deep convolutional network (which consists of several convolutional layers, batch normalization layers, ReLU activation units and max pooling layers) for multi-scale spatial feature extraction, thereby obtaining the multi-scale spatial feature tensor of the standard map image.

[0103] Step S330: For each target participating vehicle, the multi-scale spatial feature tensor and the BEV perception feature tensor of the target participating vehicle are aligned and stitched together according to feature dimensions, and the fused feature tensor obtained by stitching is subjected to semantic segmentation and decoding processing to obtain the predicted semantic segmentation image of the target participating vehicle in the target map construction area.

[0104] In this embodiment, considering that the spatial dimensions of the multi-scale spatial feature tensor and the BEV perception feature tensor may be inconsistent, these two feature tensors can be input into the same fully connected layer. This fully connected layer performs feature compression and channel standardization, resulting in a multi-scale spatial feature tensor and a BEV perception feature tensor with aligned feature dimensions. Channel-level concatenation of these two aligned feature tensors yields a corresponding fused feature tensor. Then, two two-dimensional convolutional layers are introduced to perform feature interaction and channel compression on the fused feature tensor, resulting in a fused output feature tensor. This fused output feature tensor is then decoded into a predicted semantic segmentation image from the BEV perspective using a semantic segmentation decoding network with convolutional layers and an upsampling structure. The aforementioned semantic segmentation decoding network can employ a binary cross-entropy loss function for self-supervised learning during network training.

[0105] Step S340: A non-maximum suppression mechanism is used to perform raster alignment and fusion on the predicted semantic segmentation images of all participating vehicles to obtain a local semantic map for the defective areas of the actual map.

[0106] In this embodiment, after obtaining the predicted semantic segmentation images of multiple target participating vehicles for the current map perception task, the multiple predicted semantic segmentation images can be merged based on the grid alignment method. At the same time, a non-maximum suppression mechanism is introduced. Among all the predicted semantic segmentation images, only the predicted grid semantic segmentation image with the highest confidence is retained as the final grid semantic map of the corresponding grid region, thereby obtaining the local semantic map of the actual map defect area under the current map perception task.

[0107] Step S350: Update the map completeness of the actual map completeness matrix of the target map construction area based on the local semantic map.

[0108] In this embodiment, after obtaining the local semantic map of the actual map defect area, the map integrity update operation of the actual map integrity matrix can be completed by matching the local semantic map to the overall regional map of the corresponding target map construction area and according to the matching result of the overall regional map of the target map construction area.

[0109] Therefore, by executing the above steps S310 to S350, this application can drive the various intelligent vehicles recruited for the current map perception task to perform collaborative perception and mapping within the corresponding target map construction area, thereby realizing the local map update operation for the actual map defect area that matches the current map perception task.

[0110] Alternatively, please refer to Figure 6 , Figure 6This is the third flowchart illustrating the map data acquisition and crowdsensing participant recruitment method provided in this application embodiment. In this application embodiment, the map building system 10 includes edge nodes 11 that share the same semantic segmentation image (i.e., the predicted semantic segmentation image of a map sensing task determined in step S330), sensing data upload status, and determined task participation reputation value of the intelligent vehicle in any map sensing task (which can be the current map sensing task or a historical map sensing task published by an edge node 11, or a historical map sensing task published by other edge nodes 11) for any map sensing task. Figure 5 Compared to the map data collection and crowdsensing participant recruitment methods shown, Figure 6 The map data collection crowd perception participant recruitment method shown may also include steps S410 to S460 to accurately assess the task participation reputation value of any target participating vehicle at the end of the current map perception task execution.

[0111] Step S410: For each target participating vehicle, calculate the actual map construction quality index value of the target participating vehicle in the current map perception task based on the local semantic map, the predicted semantic segmentation image of the target participating vehicle, and the historical task semantic segmentation image of the target participating vehicle.

[0112] In this embodiment, based on the concept of a map integrity matrix, the concept of a map integrity increment matrix for a single participating vehicle under the current map perception task can be introduced to describe the contribution of that participating vehicle to the mapping quality during the local semantic map construction process. Here, the current map perception task... The recruited Complete incremental matrix of maps for each target vehicle It can be represented as " ",in Used to represent a grid area In the The target vehicles are performing the current map perception task. The increase in map completeness during the process, Used to indicate the first The target vehicles participate in the current map perception task. Predictive semantic segmentation of the image with raster regions The corresponding raster semantic segmentation image, Used to indicate the first The target vehicles participate in the historical semantic segmentation image of the previous historical map perception task under the current map perception task, and the raster region. The corresponding raster semantic segmentation image.

[0113] Based on this, the current map perception task The recruited The actual map construction quality index value of the target vehicles It can be calculated using the following formula:

[0114] ;

[0115] in, Used to represent a local semantic map for the current map perception task targeting areas with defects in the actual map. Used to denote the supremum function.

[0116] Step S420: Based on the perception data upload status of the target vehicle in the current map perception task and the historical task data upload status of the target vehicle, calculate the actual data upload stability index value of the target vehicle in the current map perception task.

[0117] In this embodiment, the actual data upload stability index value of the target vehicle in the current map perception task can be obtained by statistically analyzing the total number of all map perception tasks (including the current map perception task and numerous historical map perception tasks before the current map perception task) that the target vehicle has participated in, as well as the total number of times the target vehicle has successfully uploaded perception data for the map perception tasks it has participated in. Then, the ratio of the total number of successful data uploads to the total number of perception tasks is calculated.

[0118] Step S430: Based on the actual map construction quality index value and actual data upload stability index value of the target participating vehicle, as well as the historical map construction quality index value and historical data upload stability index value of the target participating vehicle, calculate the initial reputation value of the target participating vehicle in the current map perception task using the entropy weight method.

[0119] In this embodiment, the historical map construction quality index value of a single target vehicle includes the map construction quality index value corresponding to all historical map perception tasks in which the target vehicle participated before the current map perception task, and the historical data upload stability index value of a single target vehicle includes the data upload stability index value corresponding to all historical map perception tasks in which the target vehicle participated before the current map perception task. At this point, according to the perception task participation sequence and task execution evaluation index type (i.e., map building quality index and data upload stability index) of the target participating vehicle, the actual map building quality index value, historical map building quality index value, actual data upload stability index value, and historical data upload stability index value of the target participating vehicle can be represented by matrix expansion to obtain the corresponding task execution evaluation index value distribution matrix. Then, the task execution evaluation index value distribution matrix is ​​subjected to negative index normalization processing to obtain the corresponding normalized index distribution matrix. Next, the feature weight of each of the two task execution evaluation indices under different map perception tasks is calculated in the normalized index distribution matrix, and index weight calculation operation is performed based on all the calculated feature weights (which involves index information entropy calculation, index redundancy calculation, and redundancy ratio calculation) to obtain the actual index weight of each of the aforementioned two task execution evaluation indices.

[0120] Next, by performing a weighted summation operation on all matrix elements in the normalized index distribution matrix of the target participating vehicle according to the actual index weights of the two task performance evaluation indicators mentioned above, the corresponding normalized index comprehensive characterization value can be obtained. Then, the ratio between the normalized index comprehensive characterization value and the total number of perception tasks involved in the normalized index distribution matrix is ​​calculated to obtain the initial reputation value of the corresponding target participating vehicle in the current map perception task.

[0121] Step S440: Detect whether the vehicle recruitment probability of the target participating vehicle in the target map construction area exceeds a preset probability threshold.

[0122] In this embodiment, the vehicle recruitment probability of a single target participating vehicle at a certain edge node 11 can be obtained by counting the number of times the target participating vehicle recruits in the corresponding target map construction area and calculating the ratio between the number of vehicle recruitments and the number of perception task releases of the corresponding edge node 11. The higher the vehicle recruitment probability of a target participating vehicle in the corresponding target map construction area, the easier it is for the target participating vehicle to be recruited by the edge node 11. When the vehicle recruitment probability of the target participating vehicle is detected to be less than the preset probability threshold, it indicates that the target participating vehicle is not frequently recruited at the corresponding edge node 11, and the edge node 11 will execute step S460. Conversely, when the vehicle recruitment probability of the target participating vehicle is detected to be greater than the preset probability threshold, it indicates that the target participating vehicle is frequently recruited at the corresponding edge node 11, and the edge node 11 will execute step S450.

[0123] Step S450: Based on the task participation status of the target participating vehicle in the target map construction area, the initial reputation value is corrected by the upper confidence bound algorithm to obtain the task participation reputation value of the target participating vehicle in the current map perception task.

[0124] In this embodiment, for the current map perception task The recruited The initial reputation value of a target vehicle participating in the exploration can be represented by the following formula:

[0125] ;

[0126] in, Used to indicate the first The target vehicles participate in the current map perception task. Task participation reputation value, Used to indicate the first The target vehicles participate in the current map perception task. The initial reputation value, Used to represent the exploration coefficients of the upper confidence bound algorithm; This indicates that the current map perception task has been published at the corresponding edge node 11. In the case of the first The total number of times the target vehicle recruits vehicles at the edge node 11 (i.e., the number of times the task is participated). This is used to indicate that the first [item] was published on the corresponding edge node 11. In the case of a map perception task, the first The total number of times the target vehicle recruits vehicles at edge node 11 (i.e., the number of times the task is participated).

[0127] Step S460: Directly use the initial reputation value as the target participating vehicle's mission participation reputation value in the current map perception task.

[0128] Therefore, by executing the above steps S410 to S460, this application can accurately assess the task participation reputation value of any target participating vehicle at the end of the current map perception task.

[0129] The above descriptions are merely various embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for recruiting a crowd-sensing participant for map data collection, characterized in that, The method is applied to each edge node included in a map construction system, wherein the edge node is individually responsible for a target map construction area, and the method comprises the following steps: obtaining vehicle perception performance information and vehicle motion state information periodically uploaded by each intelligent vehicle in the target map construction area, and querying historical task credit data of the intelligent vehicle; calculating perception utility score data of the intelligent vehicle in the target map construction area according to the vehicle perception performance information and the vehicle motion state information of the intelligent vehicle; determining an actual map defect area corresponding to a current map perception task in the target map construction area according to an actual map integrity matrix of the target map construction area; for each intelligent vehicle, calculating a perception adaptation state tensor of the intelligent vehicle in the target map construction area according to the actual map integrity matrix, the historical task credit data and the perception utility score data of the intelligent vehicle; calling a task participant recruitment model to perform vehicle recruitment evaluation processing based on the perception adaptation state tensor of the intelligent vehicle and the actual map defect area, and obtaining a target participant vehicle matched with the current map perception task. The task participant recruitment model adopts a double deep Q network architecture with noise introduction network, and the task participant recruitment model is used for calculating the instant reward value corresponding to the current map perception task of the intelligent vehicle at the task participant recruitment model. The instant reward value corresponding to the current map perception task of the intelligent vehicle at the task participant recruitment model is calculated by using the following formula. ； in, Used to represent the first A smart vehicle Used to represent the first The task of intelligent vehicles with current map perception. The corresponding instant reward value, , and They are respectively , and Their respective weighting coefficients Used to represent the first A smart vehicle performs a task of perceiving the current map. The perceived coverage of the corresponding actual map defect area. Used to indicate the actual map defect area in the [number]th Perceptual redundancy factors at the location of each intelligent vehicle. Used to represent the first A smart vehicle is performing a perception task on the current map. The most recent historical perception task participation reputation value. A set of defect raster regions used to represent the actual map defect area. Used to represent the first The theoretical perception grid area set of an intelligent vehicle in the target map construction area. Used to represent the first in the target map construction area Line 1 The grid area of ​​the column, Used to represent the theoretically covered sensing grid area The total number of vehicles, Functions used to calculate the number of elements in a set.

2. The method of recruiting of claim 1, wherein, The vehicle perception performance information of each intelligent vehicle includes an effective perception length range and a maximum perception field of view angle of the intelligent vehicle, the vehicle motion state information of each intelligent vehicle includes an actual vehicle position, a vehicle driving direction angle and a vehicle size of the intelligent vehicle, and the perception utility score data of each intelligent vehicle includes actual perception utility scores of the intelligent vehicle for different grid areas in the target map construction area, and the step of calculating the perception utility score data of the intelligent vehicle in the target map construction area according to the vehicle perception performance information and the vehicle motion state information of the intelligent vehicle comprises the following steps: for each intelligent vehicle, performing perception visual field shielding detection on the intelligent vehicle in the target map construction area according to actual vehicle positions and vehicle sizes of other intelligent vehicles, to obtain a shielding mask function of the intelligent vehicle for different grid areas in the target map construction area; performing perception performance attenuation detection on the intelligent vehicle in the target map construction area according to the effective perception length range, the maximum perception field of view angle, the actual vehicle position and the vehicle driving direction angle of the intelligent vehicle, to obtain a perception attenuation function of the intelligent vehicle for different grid areas; calculating actual perception utility scores of the intelligent vehicle for different grid areas according to a mapping importance coefficient and an actual grid area of each grid area in the target map construction area, and the shielding mask function and the perception attenuation function of the intelligent vehicle for different grid areas.

3. The method of recruiting of claim 2, wherein, The target map construction region satisfies a perception occlusion constraint condition when a grid region in the target map construction region is occluded by other intelligent vehicles in the global perception field of view of the intelligent vehicle The target map construction region satisfies a perception occlusion constraint condition when a grid region in the target map construction region is occluded by other intelligent vehicles in the global perception field of view of the intelligent vehicle The target map construction region satisfies a perception occlusion constraint condition when a grid region in the target map construction region is occluded by other intelligent vehicles in the global perception field of view of the intelligent vehicle The target map construction region satisfies a perception occlusion constraint condition when a grid region in the target map construction region and / or ; wherein the actual vehicle position of the first intelligent vehicle, the actual vehicle position of the second intelligent vehicle, the grid position of the grid area in the row , , and the fitted rectangle vertex position of any one of the intelligent vehicles except the first intelligent vehicle in the bird's eye view state, wherein and are on the same diagonal line of the fitted rectangle, and are on the same diagonal line of the fitted rectangle. The first The intelligent vehicle constructs the first [unclear] region based on the target map. Line 1 The occlusion mask function for the grid area of ​​a column is represented by the following function: ； wherein, a function for representing a grid region of a row and a column in the target map construction region, a function for representing an occlusion mask for the first intelligent vehicle with respect to the grid region .

4. The method of recruiting of claim 3, wherein, The target map construction region is divided into a plurality of grid regions, and each grid region is associated with a perception attenuation function. The perception attenuation function of the first row and the first column grid region is represented by the following function: ； The first Intelligent vehicle calculates the actual perception utility score for the grid area of the first row and the first column as follows: ； in, Used to represent the first in the target map construction area Line 1 The grid area of ​​the column, Used to indicate the first A smart vehicle targets the grid area. The actual perceived utility score, Used to indicate the first A smart vehicle targets the grid area. The perceived decay function, Used to represent the first A smart vehicle targets the grid area. The occlusion mask function, Used to represent a grid area The mapping importance coefficient at the target map construction area. Used to represent a grid area The actual grid area at the target map construction region. Used to indicate the first The vehicle's driving direction angle for an intelligent vehicle. Used to indicate the first The driving direction of an intelligent vehicle and the grid area The absolute value of the angle difference between them Used to indicate the first A smart vehicle targets the grid area. The optimal perceived quality value, Used to represent the attenuation coefficient Used to represent the first The actual vehicle location of each intelligent vehicle is located within the grid area. The straight-line distance Used to represent the first The lower limit of the effective sensing length range for an intelligent vehicle. Used to represent the first The upper limit of the effective sensing length range for an intelligent vehicle. Used to represent the first The maximum field of view for a smart vehicle. Used to represent the natural exponential function.

5. The method of recruiting according to any one of claims 2-4, wherein, The historical task credit data of a single intelligent vehicle includes task participation credit values of all historical perception tasks participated by the corresponding intelligent vehicle, and for each intelligent vehicle, the step of calculating a perception adaptation state tensor of the intelligent vehicle in the target map construction area according to the actual map integrity matrix, the historical task credit data and the perception utility score data of the intelligent vehicle comprises the following steps: According to the occlusion mask function, the perception attenuation function, the actual perception utility score and the actual relative distance of the intelligent vehicle for different grid areas in the target map construction area, the spatial perception intensity embedding vector of the intelligent vehicle at the target map construction area is calculated; The task participation reputation value of all historical perception tasks involving the intelligent vehicle is hidden state time sequence coded by calling the gating cycle unit, and the time perception reputation embedding vector of the intelligent vehicle at the target map construction area is obtained. According to the position coupling relationship between the intelligent vehicle and the target map construction area, the standard self-attention mechanism is used to perform defect area spatial attention learning based on the actual map integrity matrix, and the defect area spatial attention vector of the intelligent vehicle at the target map construction area is obtained. The spatial perception intensity embedding vector, the time perception reputation embedding vector and the defect area spatial attention vector of the intelligent vehicle are spliced to obtain the perception adaptation state tensor of the intelligent vehicle at the target map construction area.

6. The method of recruiting of claim 5, wherein, The step of calculating the spatial perception intensity embedding vector of the intelligent vehicle at the target map construction area according to the occlusion mask function, the perception attenuation function, the actual perception utility score and the actual relative distance of the intelligent vehicle for different grid areas in the target map construction area, comprising: The actual perception utility score, the occlusion mask function and the perception attenuation function of the intelligent vehicle for different grid areas are respectively subjected to spatial distribution image construction processing to obtain the grid perception utility distribution image, the grid occlusion distribution image and the grid perception attenuation distribution image of the intelligent vehicle at the target map construction area. The actual relative distance of the intelligent vehicle to different grid areas is subjected to distance normalization image construction to obtain the relative distance distribution image of the intelligent vehicle at the target map construction area. The grid perception utility distribution image, the grid occlusion distribution image, the grid perception attenuation distribution image and the relative distance distribution image corresponding to the intelligent vehicle are spliced to obtain the target spatial perception intensity image; The target spatial perception intensity image is subjected to high-dimensional embedding feature extraction, and the extracted high-dimensional embedding feature tensor is subjected to tensor flattening processing to obtain the spatial perception intensity embedding vector of the intelligent vehicle.

7. The method of recruiting according to any one of claims 1-4, wherein, The recruitment method further comprises: The actual perception images uploaded by all target participating vehicles during the execution of the current map perception task are obtained, and the BEV feature of each actual perception image of all target participating vehicles is extracted to obtain the BEV perception feature tensor of each target participating vehicle; The standard map image of the target map construction area is obtained, and the multi-scale spatial feature of the standard map image is extracted to obtain the multi-scale spatial feature tensor of the standard map image; For each target participating vehicle, the multi-scale spatial feature tensor and the BEV perception feature tensor of the target participating vehicle are aligned and spliced in feature dimension, and the fused feature tensor obtained by splicing is subjected to semantic segmentation decoding processing to obtain a predicted semantic segmentation image of the target participating vehicle at the target map construction area; A non-maximum suppression mechanism is used to perform grid alignment fusion on the predicted semantic segmentation images of all target participating vehicles to obtain a local semantic map for the actual map defect area; Based on the local semantic map, the actual map integrity matrix of the target map construction area is updated in map integrity.

8. The method of recruiting of claim 7, wherein, If the semantic segmentation images, perception data upload status and task participation reputation value of the same vehicle in any map perception task are shared by each edge node of the map construction system, the recruiting method further comprises: For each target participating vehicle, the actual map construction quality index value of the target participating vehicle in the current map perception task is calculated according to the local semantic map, the predicted semantic segmentation image of the target participating vehicle, and the historical task semantic segmentation image of the target participating vehicle; The actual data upload stability index value of the target participating vehicle in the current map perception task is calculated according to the perception data upload status of the target participating vehicle in the current map perception task and the historical task data upload status of the target participating vehicle; The initial reputation value of the target participating vehicle in the current map perception task is calculated based on the entropy weight method according to the actual map construction quality index value and the actual data upload stability index value of the target participating vehicle, and the historical map construction quality index value and the historical data upload stability index value of the target participating vehicle; It is detected whether the vehicle recruitment probability of the target participating vehicle at the target map construction area exceeds a preset probability threshold; If it is detected that the vehicle recruitment probability exceeds the preset probability threshold, the initial reputation value is corrected in reputation exploration by using the upper confidence bound algorithm according to the task participation status of the target participating vehicle at the target map construction area to obtain the task participation reputation value of the target participating vehicle in the current map perception task, otherwise the initial reputation value is directly taken as the task participation reputation value of the target participating vehicle in the current map perception task.

9. A map construction system characterized by comprising: The construction system comprises a cloud server and a plurality of edge nodes, wherein the plurality of edge nodes are respectively in communication connection with the cloud server to realize data sharing function between the plurality of edge nodes through the cloud server, and each edge node is responsible for realizing map construction function of a target map construction area; Each edge node stores a computer program, and can realize the map data acquisition group intelligence perception participant recruiting method of any one of claims 1-8 by running the computer program.

Images