5G MEC task management method based on intelligent vehicle networking

Abstract

The present application relates to Internet of Vehicles technology, and discloses a 5G MEC task management method based on intelligent Internet of Vehicles, the method comprising: a first edge server receives the detection image of the intelligent vehicle it covers, and judges the processing capacity based on its own load. If it can be processed, the local model is used for detection and the result is directly fed back to the vehicle. If the load is too high, the detection data is transferred to the central server, which distributes the task to the second edge server for processing according to the real-time load of each edge server, and returns the result to the first edge server after completion, which then conveys it to the vehicle. The present application realizes efficient processing of intelligent vehicle terminal detection data by intelligently allocating and utilizing edge computing resources, improves the real-time performance and processing capability of the system, and also optimizes the efficiency of resource utilization.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle networking, and in particular to a 5GMEC task management method based on intelligent vehicle networking. Background Art

[0002] With the rapid development of intelligent vehicle networking technology, intelligent vehicle terminals need to process a large amount of detection data, such as detection images, in real time to realize functions such as advanced driver assistance systems (ADAS) and autonomous driving. However, the traditional cloud computing model is difficult to meet the needs of real-time data processing due to network latency and bandwidth limitations. At the same time, the computing power of the on-board computing platform is limited and it is difficult to cope with the growing demand for data processing. Therefore, a more efficient and real-time data processing solution has become particularly important. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a 5G MEC task management method based on the intelligent Internet of Vehicles, which realizes real-time or near real-time processing of detection data in the intelligent Internet of Vehicles, solves the problems of network delay and limited vehicle computing power in the traditional cloud computing model, and improves the efficiency and flexibility of resource allocation.

[0004] According to one aspect of the present application, a 5G MEC task management method based on intelligent vehicle networking is provided, which is applied to a MEC task management system. The MEC task management system includes a central server and multiple edge servers. The method includes:

[0005] The first edge server receives detection data sent by the intelligent vehicle terminal within the processing range of the first edge server, wherein the detection data includes a detection image, and the first edge server is any one of the plurality of edge servers;

[0006] The first edge server determines whether it has the processing capability for the detection image based on the current load information;

[0007] If the first edge server determines that it has the ability to process the detection image, it performs target detection on the detection image based on a locally deployed first detection model, and sends the first target detection result to the smart vehicle terminal; the first detection model is a detection model corresponding to the first edge server;

[0008] If the first edge server determines that it does not have the ability to process the detection image, the detection data is sent to the central server, and the central server assigns a second edge server to perform task processing for the detection image based on the real-time load information reported by each of the multiple edge servers, so that the second edge server performs target detection on the detection image based on the second detection model deployed locally, and sends the second target detection result to the first edge server, and the first edge server sends the second target detection result to the smart vehicle terminal; the second detection model is the detection model corresponding to the second edge server; the second edge server is an edge server other than the first edge server.

[0009] In the embodiment of the present application, optionally, the method further includes:

[0010] Each of the edge servers stores the received detection image and the target detection result corresponding to the detection image as historical detection data, and periodically reports the historical detection data to the central server;

[0011] The central server performs the following processing on the historical detection data sent by each of the edge servers to obtain a training data sample corresponding to each of the edge servers: the central server determines, based on the acquired user feedback information on the historical detection data, the historical detection data for which the user feedback information indicates that the target detection result is correct as the first training data sample; randomly samples the remaining historical detection data except the first training data sample to obtain a first candidate data sample and a second candidate data sample, and verifies the first candidate data sample based on the locally deployed central detection model to obtain a first candidate data sample that passes the verification as the second training data sample, verifies the second candidate data sample based on the expert system to obtain a second candidate data sample that passes the verification as the third training data sample, and merges the first training data sample, the second training data sample and the third training data sample to obtain the training data sample;

[0012] The central server sends the training data samples to the corresponding edge servers, and the edge servers iteratively train their respective locally deployed detection models according to the received training data samples.

[0013] In an embodiment of the present application, optionally, the edge server iteratively trains the detection model deployed locally according to the received training data sample, including:

[0014] The edge server copies the locally deployed detection model to obtain a detection model copy, and iteratively trains the detection model copy using the training data sample to obtain a trained detection model copy, extracts model parameters of the trained detection model copy, deploys the model parameters in the detection model, and deletes the trained detection model copy.

[0015] In the embodiment of the present application, optionally, the detection model deployed by each edge server has the same model structure as the central detection model deployed by the central server;

[0016] After extracting the model parameters of the trained detection model copy, the method further includes:

[0017] Each of the edge servers sends its own model parameters to the central server;

[0018] The central server performs parameter fusion on the model parameters corresponding to each of the edge servers to obtain target model parameters, and deploys the target model parameters in the central detection model.

[0019] In the embodiment of the present application, optionally, the central server performs parameter fusion on the model parameters corresponding to each edge server to obtain the target model parameters, including:

[0020] For each of the edge servers, the central server calculates the sum of the number of candidate samples according to the first number of candidate data samples and the second number of candidate data samples corresponding to each of the edge servers, calculates the sum of the number of training samples according to the second number of training data samples and the third number of training samples corresponding to the edge server, and calculates the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server;

[0021] The central server performs parameter fusion on the model parameters of each edge server according to the parameter weight of each edge server to obtain the target model parameters.

[0022] In the embodiment of the present application, optionally, after each edge server sends its own model parameter to the central server, the method further includes:

[0023] The central server calculates the model parameter similarity between different edge servers according to the model parameters corresponding to each edge server, and determines a similar edge server corresponding to each edge server according to the model parameter similarity;

[0024] Accordingly, the central server assigns the detection data to a second edge server for task processing based on the real-time load information reported by each of the plurality of edge servers, including:

[0025] The central server determines a candidate edge server having processing capability for the detection image based on real-time load information reported by each of the plurality of edge servers;

[0026] If there is one candidate edge server, use the candidate edge server as the second edge server, and assign the detection data to the second edge server for task processing;

[0027] If there are multiple candidate edge servers, query whether there is a similar edge server corresponding to the first edge server among the candidate edge servers, determine the second edge server based on the query result, and assign the detection data to the second edge server for task processing.

[0028] In the embodiment of the present application, optionally, the determining the second edge server based on the query result includes:

[0029] If there are multiple candidate edge servers belonging to the similar edge server, then according to the geographical location corresponding to the first edge server, determine a candidate edge server that is closest to the first edge server among the multiple candidate edge servers belonging to the similar edge server as the second edge server;

[0030] If there is a candidate edge server belonging to the similar edge server, use the edge server as the second edge server;

[0031] If there is no candidate edge server belonging to the similar edge server, a distance score and a model parameter similarity score between the candidate edge server and the first edge server are obtained, and a total score calculation rule is determined according to the current driving environment in the detection data, and a total score is calculated for the distance score and the model parameter similarity score according to the total score calculation rule, and the candidate edge server with the highest total score is determined to be the second edge server.

[0032] According to another aspect of the present application, a 5G MEC task management system based on intelligent Internet of Vehicles is provided, the system comprising:

[0033] A central server and multiple edge servers;

[0034] A first edge server is used to receive detection data sent by an intelligent vehicle terminal within a processing range of the first edge server, wherein the detection data includes a detection image, and the first edge server is any one of the plurality of edge servers; judging whether it has the processing capability of the detection image based on current load information; if the first edge server judges that it has the processing capability of the detection image, performing target detection on the detection image based on a locally deployed first detection model, and sending a first target detection result to the intelligent vehicle terminal; the first detection model is a detection model corresponding to the first edge server; if the first edge server judges that it does not have the processing capability of the detection image, sending the detection data to the central server;

[0035] The central server is used to assign a second edge server to perform task processing for the detection image based on the real-time load information reported by each of the plurality of edge servers;

[0036] The second edge server is configured to perform target detection on the detection image based on a locally deployed second detection model, and send a second target detection result to the first edge server;

[0037] The first edge server is also used to send the second target detection result to the intelligent vehicle terminal; the second detection model is the detection model corresponding to the second edge server; the second edge server is an edge server other than the first edge server.

[0038] In the embodiment of the present application, optionally,

[0039] The edge server is further used to store the received detection image and the target detection result corresponding to the detection image as historical detection data, and periodically report the historical detection data to the central server;

[0040] The central server is further configured to perform the following processing on the historical detection data sent by each edge server to obtain a training data sample corresponding to each edge server: the central server determines, based on the acquired user feedback information on the historical detection data, historical detection data for which the user feedback information indicates that the target detection result is correct as a first training data sample; randomly samples the remaining historical detection data except the first training data sample to obtain a first candidate data sample and a second candidate data sample, and verifies the first candidate data sample based on a locally deployed central detection model to obtain a first candidate data sample that passes the verification as a second training data sample, verifies the second candidate data sample based on an expert system to obtain a second candidate data sample that passes the verification as a third training data sample, and merges the first training data sample, the second training data sample, and the third training data sample to obtain the training data sample;

[0041] The central server is further used to send the training data samples to the corresponding edge servers, and the edge servers iteratively train their respective locally deployed detection models according to the received training data samples.

[0042] In the embodiment of the present application, optionally,

[0043] The edge server is also used to copy the locally deployed detection model to obtain a detection model copy, and iteratively train the detection model copy using the training data sample to obtain a trained detection model copy, extract model parameters of the trained detection model copy, deploy the model parameters in the detection model, and delete the trained detection model copy.

[0044] In the embodiment of the present application, optionally, the detection model deployed by each edge server has the same model structure as the central detection model deployed by the central server;

[0045] The edge server is further used to send the respective model parameters to the central server;

[0046] The central server is further used to perform parameter fusion on the model parameters corresponding to each of the edge servers to obtain target model parameters, and deploy the target model parameters in the central detection model.

[0047] In the embodiment of the present application, optionally,

[0048] The central server is also used to calculate the sum of the number of candidate samples according to the first number of candidate data samples and the second number of candidate data samples corresponding to each edge server, calculate the sum of the number of training samples according to the second number of training data samples and the third number of training samples corresponding to the edge server, and calculate the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server; perform parameter fusion on the model parameters of each edge server according to the parameter weight of each edge server to obtain the target model parameters.

[0049] In the embodiment of the present application, optionally,

[0050] The central server is further configured to calculate the model parameter similarity between different edge servers according to the model parameters corresponding to each edge server, and determine the similar edge server corresponding to each edge server according to the model parameter similarity;

[0051] The central server is further used to determine a candidate edge server that has processing capabilities for the detection image based on real-time load information reported by each of the plurality of edge servers; if there is one candidate edge server, the candidate edge server is used as the second edge server, and the detection data is assigned to the second edge server for task processing; if there are multiple candidate edge servers, query whether there is a similar edge server corresponding to the first edge server among the candidate edge servers, determine the second edge server based on the query result, and assign the detection data to the second edge server for task processing.

[0052] In the embodiment of the present application, optionally, the central server is further used to:

[0053] If there are multiple candidate edge servers belonging to the similar edge server, then according to the geographical location corresponding to the first edge server, determine a candidate edge server that is closest to the first edge server among the multiple candidate edge servers belonging to the similar edge server as the second edge server;

[0054] If there is a candidate edge server belonging to the similar edge server, use the edge server as the second edge server;

[0055] If there is no candidate edge server belonging to the similar edge server, a distance score and a model parameter similarity score between the candidate edge server and the first edge server are obtained, and a total score calculation rule is determined according to the current driving environment in the detection data, and a total score is calculated for the distance score and the model parameter similarity score according to the total score calculation rule, and the candidate edge server with the highest total score is determined to be the second edge server.

[0056] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the above-mentioned 5G MEC task management method based on intelligent vehicle networking is implemented.

[0057] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein when the processor executes the program, the above-mentioned 5G MEC task management method based on intelligent vehicle networking is implemented.

[0058] With the help of the above technical solution, the embodiment of the present application provides a 5G MEC task management method based on the intelligent vehicle network, which realizes the efficient processing of intelligent vehicle terminal detection data by intelligently allocating and utilizing edge computing resources, improves the real-time and processing capabilities of the system, and also optimizes the utilization efficiency of resources.

[0059] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0061] Figure 1 A flow chart of a 5G MEC task management method based on intelligent Internet of Vehicles provided in an embodiment of the present application is shown;

[0062] Figure 2 A flow chart of another 5G MEC task management method based on intelligent Internet of Vehicles provided in an embodiment of the present application is shown;

[0063] Figure 3 A structural diagram of a 5G MEC task management system based on intelligent Internet of Vehicles provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0064] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0065] In this embodiment, a 5G MEC task management method based on intelligent vehicle networking is provided. Figure 1 As shown, it is applied to a MEC task management system, the MEC task management system includes a central server and multiple edge servers, and the method includes:

[0066] Step 101: A first edge server receives detection data sent by an intelligent vehicle terminal within a processing range of the first edge server, wherein the detection data includes a detection image, and the first edge server is any one of the plurality of edge servers.

[0067] Step 102: The first edge server determines whether it has the ability to process the detection image based on current load information.

[0068] Step 103: If the first edge server determines that it has the ability to process the detection image, it performs target detection on the detection image based on a locally deployed first detection model, and sends the first target detection result to the smart vehicle terminal; the first detection model is the detection model corresponding to the first edge server.

[0069] Step 104: If the first edge server determines that it does not have the ability to process the detection image, the detection data is sent to the central server, and the central server assigns a second edge server to perform task processing for the detection image based on the real-time load information reported by each of the plurality of edge servers, so that the second edge server performs target detection on the detection image based on the second detection model deployed locally, and sends the second target detection result to the first edge server, and the first edge server sends the second target detection result to the smart vehicle terminal; the second detection model is the detection model corresponding to the second edge server; the second edge server is an edge server other than the first edge server.

[0070] The embodiment of the present application mainly provides a solution for efficiently processing and distributing detection data (mainly detection images) from intelligent vehicle terminals in an intelligent transportation system, performing target detection and returning the results to the vehicle. First, a first edge server (as one of multiple edge servers) receives detection data sent by intelligent vehicle terminals within its processing range. These data typically include detection images captured by vehicle sensors (such as cameras) for subsequent target detection analysis. Receiving detection data through edge servers can reduce the delay in data transmission to remote central servers and improve real-time performance. At the same time, utilizing edge computing resources can reduce the burden on central servers and optimize resource allocation for the overall system.

[0071] Secondly, after receiving the detection images, the first edge server will determine whether it has the ability to process these images based on its current load information (such as CPU usage, memory usage, etc.). By evaluating its own capabilities, it can avoid overload and ensure system stability and processing efficiency. If the server load is too high, continuing to process new tasks may result in performance degradation or task failure.

[0072] Finally, if the first edge server determines that it has sufficient processing power, it will use the locally deployed first detection model (i.e., the detection model corresponding to the first edge server) to perform target detection on the received detection image. After the detection is completed, the detection results (such as identified vehicles, pedestrians, etc.) are sent back to the intelligent vehicle terminal. Direct processing at the edge can greatly reduce data transmission delay and improve response speed. At the same time, the use of locally deployed detection models can reduce dependence on the central server and enhance the autonomy and robustness of the system. If the first edge server determines that its processing power is insufficient, it will forward the detection data to the central server. The central server assigns a second edge server with processing power (i.e., one of the edge servers other than the first edge server) to these images for task processing based on the real-time load information of each edge server. After the second edge server completes the target detection, it sends the result directly back to the first edge server, which is then forwarded to the intelligent vehicle terminal by the first edge server, thereby realizing dynamic task allocation and load balancing. Through the unified scheduling of the central server, it can ensure that the detection task is assigned to the most suitable edge server for processing, thereby maximizing the processing power and efficiency of the overall system. At the same time, this flexible scheduling mechanism also improves the scalability and fault tolerance of the system.

[0073] The present invention realizes efficient processing of intelligent vehicle terminal detection data by intelligently allocating and utilizing edge computing resources, improves the real-time performance and processing capability of the system, and also optimizes the utilization efficiency of resources.

[0074] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another 5G MEC task management method based on intelligent vehicle networking is provided, such as Figure 2 As shown, the method includes:

[0075] In step 201, each edge server stores the received detection image and the target detection result corresponding to the detection image as historical detection data, and periodically reports the historical detection data to the central server.

[0076] In the embodiment of the present application, the mechanism of the edge server and the central server working together can also be used in the intelligent transportation system to optimize and improve the detection model. Wherein, each edge server, when processing the detection image from the intelligent vehicle terminal, not only outputs the target detection result, but also can store these detection images and their corresponding target detection results as historical detection data. Subsequently, these edge servers can periodically report these historical detection data to the central server. Here, the historical detection data reported to the central server by a certain edge server refers to the detection image contained in the historical detection data as the detection image directly received by the edge server from the intelligent vehicle terminal. The historical detection data generated by the edge server processing the task assigned by the central server will not be reported to the central server. The historical detection data reported to the central server is used to screen the samples of the detection model in the subsequent training of the edge server. Such processing can make the detection model more adaptable to the detection environment within the processing range of the edge server during the iterative training process, and help to improve the accuracy of the detection image taken by the detection model for the detection environment within the specific processing range. Storing historical detection data provides a rich data resource for subsequent model training. By reporting these data to the central server, centralized management and utilization of data can be achieved, which is convenient for unified analysis and processing.

[0077] In step 202, the central server performs the following processing on the historical detection data sent by each of the edge servers to obtain a training data sample corresponding to each of the edge servers: the central server determines, based on the user feedback information obtained for the historical detection data, the historical detection data for which the user feedback information indicates that the target detection result is correct as the first training data sample; randomly samples the remaining historical detection data except the first training data sample to obtain a first candidate data sample and a second candidate data sample, and verifies the first candidate data sample based on the locally deployed central detection model to obtain a verified first candidate data sample as the second training data sample, verifies the second candidate data sample based on the expert system to obtain a verified second candidate data sample as the third training data sample, and merges the first training data sample, the second training data sample and the third training data sample to obtain the training data sample.

[0078] In an embodiment of the present application, after the central server receives the historical detection data reported by each edge server, it can generate training data samples according to a series of rules and processing procedures. First, determine which historical detection data has the correct target detection results based on user feedback information, and the historical detection data corresponding to these correct user feedback information is used as the first training data sample. Then, the remaining historical detection data is randomly sampled to obtain the first candidate data sample and the second candidate data sample. Next, the first candidate data sample is verified using the central detection model, and the first candidate data sample that passes the verification is used as the second training data sample; at the same time, the second candidate data sample is verified using the expert system, and the second candidate data sample that passes the verification is used as the third training data sample. Finally, these three types of data samples are merged to obtain the final training data sample. The correct historical detection data is determined through user feedback, ensuring the accuracy and reliability of the training data. The candidate data samples are verified in combination with the central detection model and the expert system, which improves the accuracy and efficiency of data screening. Through random sampling and multiple verification methods, the diversity and comprehensiveness of the training data samples are ensured, which helps to improve the generalization ability of the model.

[0079] In step 203, the central server sends the training data samples to the corresponding edge servers, and the edge servers iteratively train their locally deployed detection models according to the received training data samples.

[0080] In the embodiment of the present application, the central server sends the generated training data samples back to the corresponding edge server. After receiving these training data samples, the edge server will iteratively train the locally deployed detection model according to these data to optimize the performance of the model. Through iterative training, the detection model on the edge server can continuously learn and adapt to new data, thereby improving the accuracy and efficiency of detection. The training data samples sent by the central server to a certain edge server are completely derived from the historical detection data reported by the edge server, so that the training data samples used by the edge server for training are derived from the local historical detection data. Such a sample selection method can make the local detection model of each edge server better adapt to the detection environment within the processing range of the edge server, which helps to improve the accuracy of the detection images taken by the detection model for the detection environment within a specific processing range. The collaborative work of the central server and the edge server realizes the centralized management of data and the distributed training of the model, and improves the flexibility and scalability of the entire system. In summary, the scheme realizes the continuous optimization and improvement of the detection model in the intelligent transportation system by storing and reporting historical detection data, generating diversified training data samples, and iteratively training the detection model on the edge server, thereby improving the overall performance and user experience of the system.

[0081] In an embodiment of the present application, optionally, the edge server in step 203 iteratively trains the respective locally deployed detection models according to the received training data samples, including: step 203-1, the edge server copies the locally deployed detection model to obtain a detection model copy, and iteratively trains the detection model copy using the training data samples to obtain a trained detection model copy, extracts model parameters of the trained detection model copy, deploys the model parameters in the detection model, and deletes the trained detection model copy.

[0082] In this embodiment, in order to ensure that the detection model is updated without interrupting the current service, in the embodiment of the present application, the edge server first copies the detection model deployed locally to generate a detection model copy. This step is to avoid affecting the running original model during the training process, thereby maintaining the stability and continuity of the system. The detection model copy is iteratively trained using the received training data samples. In this process, the model learns and adjusts its internal parameters according to the training data to improve the accuracy and efficiency of target detection. After the iterative training is completed, the optimized model parameters are extracted from the trained detection model copy. These parameters are the result of model learning and contain the knowledge learned from the training data. The extracted model parameters are deployed to the original detection model to replace the old parameters. In this way, the original detection model is updated and has better detection capabilities. Finally, the no longer needed copy of the trained detection model is deleted to release system resources. The embodiment of the present application performs iterative training by copying the detection model, thereby avoiding interruption or affecting the running service during the training process and ensuring the stability and continuity of the system. Iterative training of the model using training data samples can continuously optimize the parameters and performance of the model and improve the accuracy and efficiency of target detection. Deleting model copies that are no longer needed after training is completed helps free up system resources and improve resource utilization. And because model updates are performed locally on the edge server, data transmission delays and bandwidth consumption can be greatly reduced, enabling fast response and real-time processing.

[0083] In an embodiment of the present application, optionally, the detection model deployed by each of the edge servers has the same model structure as the central detection model deployed by the central server; after extracting the model parameters of the trained detection model copy in step 203-1, the method also includes: step 203-2, each of the edge servers sends its own model parameters to the central server; the central server performs parameter fusion on the model parameters corresponding to each of the edge servers to obtain target model parameters, and deploys the target model parameters in the central detection model.

[0084] In this embodiment, after each edge server completes the iterative training of the training model copy and extracts the model parameters, the model parameters can be further shared and integrated to optimize the performance of the central detection model. Among them, each edge server sends the model parameters of the respective trained detection model copies to the central server. These model parameters contain the knowledge and features learned by the edge server based on local data and a specific environment. After the central server receives the model parameters from each edge server, it will perform parameter integration. The parameter fusion process may include a variety of methods such as weighted averaging, feature selection, and model integration, which aims to integrate the model parameters of each edge server to obtain a more comprehensive and generalized target model parameter. The central server deploys the fused target model parameters in the central detection model. In this way, the central detection model can absorb the knowledge and experience from each edge server and improve its own detection performance and generalization ability. In the embodiment of the present application, through the sharing of model parameters, each edge server can learn from each other and share the knowledge and features learned by each other in a specific environment and data, thereby accelerating the learning process of the model. By integrating the model parameters from each edge server, the central server can obtain a more comprehensive and generalized central detection model. This central detection model can better adapt to different scenarios and data, and improve the accuracy and robustness of detection. The collaborative work of edge servers and central servers enables the system to use resources more efficiently. The edge server is responsible for processing local data and optimizing local models, while the central server is responsible for integrating global knowledge and optimizing the central model, thereby achieving optimal resource allocation. As the edge server continues to learn and optimize the local model and feeds the model parameters back to the central server for fusion, the entire intelligent transportation system can continue to evolve and develop to adapt to more complex and changing environments.

[0085] In an embodiment of the present application, optionally, the central server in step 203-2 performs parameter fusion on the model parameters corresponding to each of the edge servers to obtain target model parameters, including: for each of the edge servers, the central server calculates the sum of the number of candidate samples based on the first number of candidate data samples and the second number of candidate data samples corresponding to each of the edge servers, calculates the sum of the number of training samples based on the second number of training data samples and the third number of training samples corresponding to the edge servers, and calculates the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server; the central server performs parameter fusion on the model parameters of each of the edge servers according to the parameter weights of each of the edge servers to obtain the target model parameters.

[0086] In this embodiment, in the process of parameter fusion of the model parameters corresponding to each edge server by the central server, a parameter weight calculation method based on the number of candidate samples and the number of training samples can be used to more reasonably allocate the contribution of each edge server in the fusion process. Among them, for each edge server, the central server first calculates the sum of the number of candidate samples according to the first number of candidate data samples and the second number of candidate data samples of each edge server. This reflects the contribution of the edge server in generating candidate data samples. Next, the central server calculates the sum of the number of training samples according to the second number of training data samples and the third number of training data samples corresponding to each edge server. This represents the contribution of the edge server in providing effective training data. Then, the central server uses the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server. This ratio takes into account both the diversity of candidate samples (embodied by the sum of candidate samples) and the effectiveness of training samples (embodied by the sum of training samples). After obtaining the parameter weights of each edge server, the central server will fuse the model parameters of each edge server according to these weights. Specifically, the central server can perform weighted averaging on the model parameters of each edge server, where the weight is the parameter weight calculated previously. In this way, the central server can integrate the knowledge and experience of each edge server to obtain a more comprehensive and generalized target model parameter. This target model parameter not only reflects the local characteristics of each edge server, but also integrates global information, which helps to improve the performance of the central detection model. In the embodiment of the present application, the central server can more fairly evaluate the contribution of each edge server through a parameter weight calculation method based on the number of candidate samples and the number of training samples, and avoid some edge servers being ignored due to small data volume or low quality. The parameter fusion process fully considers the local characteristics and global information of each edge server, making the fused target model parameters more effective and generalized. As the edge server continues to learn and optimize the local model, and reports the updated model parameters and sample data to the central server to optimize the central detection model, the central detection model can also continuously adapt to new data and scenarios, and improve its own detection performance and robustness.

[0087] In the embodiment of the present application, optionally, after each edge server sends its own model parameter to the central server, the method further includes: the central server calculates the model parameter similarity between different edge servers according to the model parameters corresponding to each edge server, and determines a similar edge server corresponding to each edge server according to the model parameter similarity;

[0088] Correspondingly, the central server assigns the second edge server to the detection data for task processing based on the real-time load information reported by each of the plurality of edge servers, including: the central server determines a candidate edge server having processing capability for the detection image based on the real-time load information reported by each of the plurality of edge servers; if there is one candidate edge server, the candidate edge server is used as the second edge server, and the detection data is assigned to the second edge server for task processing; if there are multiple candidate edge servers, query whether there is a similar edge server corresponding to the first edge server among the candidate edge servers, determine the second edge server based on the query result, and assign the detection data to the second edge server for task processing.

[0089] In this embodiment, after the edge servers send their respective model parameters to the central server, the central server will further use these model parameters to optimize the task allocation process, determine the similarity between the edge servers by calculating the model parameter similarity, and assign the second edge server for task processing for the detection data based on the real-time load information and similarity. First, after the central server receives the model parameters sent by each edge server, it will calculate the model parameter similarity between different edge servers, such as calculating the cosine similarity, Euclidean distance, etc. between the parameter vectors. Then, based on the calculated model parameter similarity, the central server will determine the similar edge servers corresponding to each edge server. These similar edge servers have high similarity in model parameters, which means that they may have similar performance and effects when processing similar tasks. Then, when there is detection data to be processed, the central server will first determine the candidate edge servers that have the processing capability for the detection image based on the real-time load information reported by each of the multiple edge servers. These candidate edge servers are currently under low load and have sufficient resources to process new tasks. Among them, if there is only one candidate edge server, the central server will directly use this candidate edge server as the second edge server to assign tasks to the detection data. If there are multiple candidate edge servers, the central server will further query whether there are similar edge servers corresponding to the first edge server (i.e., the edge server that initially receives the detection data) among these candidate edge servers, so that the similar edge server can obtain a processing effect closer to the first edge server, that is, the similar edge server may have better effect and efficiency when processing similar tasks. If there are similar edge servers, the central server will use the candidate edge servers belonging to the similar edge servers as the second edge servers to assign tasks to the detection data. Doing so can further improve the efficiency and accuracy of task processing. If there are no similar edge servers, the central server can select the second edge server according to other strategies (such as based on distance, etc.). The embodiment of the present application determines the similarity between edge servers by calculating the similarity of model parameters, and optimizes task allocation based on similarity and real-time load information, which can ensure that the detection data is assigned to the edge server that is most suitable for processing, thereby improving the efficiency and accuracy of task processing. In the case of multiple candidate edge servers, similar edge servers are preferentially selected for processing, which can reduce processing errors caused by model differences and improve the stability and reliability of the system. The method can dynamically adjust the task allocation strategy according to the real-time load information and model parameter similarity of the edge server to adapt to the task processing requirements under different scenarios and requirements.

[0090] In the embodiment of the present application, optionally, determining the second edge server based on the query result includes:

[0091] If there are multiple candidate edge servers belonging to the similar edge server, then according to the geographical location corresponding to the first edge server, determine a candidate edge server that is closest to the first edge server among the multiple candidate edge servers belonging to the similar edge server as the second edge server;

[0092] If there is a candidate edge server belonging to the similar edge server, use the edge server as the second edge server;

[0093] If there is no candidate edge server belonging to the similar edge server, a distance score and a model parameter similarity score between the candidate edge server and the first edge server are obtained, and a total score calculation rule is determined according to the current driving environment in the detection data, and a total score is calculated for the distance score and the model parameter similarity score according to the total score calculation rule, and the candidate edge server with the highest total score is determined to be the second edge server.

[0094] In the above embodiment, when determining to use a candidate edge server belonging to a similar edge server as the second edge server, the central server will further consider factors such as the geographical location, the number of candidate edge servers, and the similarity of model parameters. Among them, if there are multiple candidate edge servers belonging to similar edge servers, the central server will further consider the geographical location relationship between these candidate edge servers and the first edge server (i.e., the edge server that initially receives the detection data). The specific steps are as follows: First, the central server calculates the distance between each candidate edge server and the first edge server. Then, among the candidate edge servers belonging to the similar edge servers, the candidate edge server closest to the first edge server is selected as the second edge server. Doing so can reduce the delay and cost of data transmission while maintaining processing efficiency. If only one candidate edge server belongs to the similar edge server, then the edge server is naturally selected as the second edge server without further comparison or calculation.

[0095] If there is no candidate edge server that belongs to the similar edge server, the central server will use a comprehensive scoring mechanism to determine the second edge server. The specific steps are as follows: First, calculate the distance score between each candidate edge server and the first edge server. The closer the distance, the higher the score. Specifically, the distance between all edge servers can be calculated, and the distance can be normalized after taking the inverse of the distance to obtain the distance score between each edge server. Secondly, although these candidate edge servers do not belong to similar edge servers, the central server can still calculate their similarity scores with the first edge server in model parameters. The higher the similarity, the higher the score. Then, according to the current driving environment in the detection data (such as city, suburb, highway, etc.), the current driving environment can be determined according to the location of the intelligent vehicle terminal, and the central server will determine a total score calculation rule. This rule may consider the relative importance of distance and model parameter similarity in different environments. Among them, the impact of the driving environment on the total score rule is mainly reflected in the fact that the relative importance of the distance score and the model parameter similarity score will be different in different environments. For example: for the urban driving environment, the characteristics are dense roads, large traffic flow, many obstacles, and high real-time requirements. Total scoring rules: Distance score: Due to the complex and changeable traffic conditions in urban environments, the real-time requirements for tasks are high, so the weight of the distance score in the total score can be appropriately increased. Closer edge servers can respond and process tasks faster, reducing the delay in data transmission. Model parameter similarity score: Although it is also important, in urban environments, more emphasis may be placed on rapid response and processing of tasks, so the weight of the model parameter similarity score can be relatively low. Therefore, in urban driving environments, the total scoring rule is more inclined to select candidate edge servers that are closer and can respond quickly. For highway driving environments, the characteristics are wide roads and relatively stable traffic flow, but the speed is fast, and the requirements for safety and stability are high. Total scoring rules: Distance score: Although distance is still a consideration, on highways, due to the high speed and relatively stable road conditions, the requirements for real-time may not be as high as in urban environments. Therefore, the weight of the distance score can be appropriately reduced. Model parameter similarity score: On highways, due to the high speed and serious consequences of accidents, the weight of the model parameter similarity score should be increased. Selecting edge servers with similar model parameters can reduce processing errors caused by model differences and improve the accuracy and safety of task processing. Therefore, in the highway driving environment, the total scoring rule tends to select candidate edge servers with higher model parameter similarity and more stable processing. In general, this flexible scoring rule helps ensure that the most suitable edge server can be selected for task processing in different driving environments. Finally, according to the scoring calculation rule, the distance score and model parameter similarity score are calculated in total.Then, the candidate edge server with the highest total score is selected as the second edge server. This ensures that when there are no similar edge servers to choose from, the most suitable edge server can still be selected based on a comprehensive evaluation of multiple factors to process the task.

[0096] By applying the technical solution of this embodiment: First, the real-time load information of the edge server is monitored by the central server, and tasks are dynamically assigned to the detection data based on this information, which can effectively optimize resource allocation and ensure the balance of system load. This avoids the situation where some edge servers are overloaded while other servers are idle, and improves the overall performance and response speed of the system. Second, by calculating the similarity of model parameters between different edge servers, the central server can identify edge servers with similar performance and effects when processing similar tasks. When assigning tasks, giving priority to these similar edge servers can reduce processing errors caused by model differences and improve the efficiency and accuracy of task processing. Third, the solution can adapt to task processing requirements under different scenarios and requirements. Whether it is the case of a single candidate edge server, or multiple candidate edge servers with similar edge servers or no similar edge servers, the central server can comprehensively evaluate multiple factors such as real-time load information, geographic location, model parameter similarity, and flexibly select the most suitable edge server for task processing. This adaptability and flexibility enables the system to better cope with complex and changing application scenarios. Fourth, by optimizing task allocation and improving processing efficiency, the solution can respond to user requests faster, reduce user waiting time, and thus improve user experience. At the same time, due to the improved accuracy of task processing, the results obtained by users will also be more reliable and useful. Fifth, the edge servers send their own model parameters to the central server for parameter fusion, which helps to achieve data sharing and collaborative learning between different edge servers. This can not only improve the generalization ability and robustness of each model, but also promote the development of the entire system to a higher level through unified management and optimization of the central server.

[0097] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a 5G MEC task management system based on intelligent vehicle networking, such as Figure 3 As shown, the system includes:

[0098] A central server 301 and multiple edge servers 302;

[0099] The first edge server 3021 is used to receive detection data sent by the intelligent vehicle terminal within the processing range of the first edge server 3021, wherein the detection data includes a detection image, and the first edge server 3021 is any one of the multiple edge servers 302; based on the current load information, it is judged whether it has the processing capability of the detection image; if the first edge server 3021 judges that it has the processing capability of the detection image, it performs target detection on the detection image based on the locally deployed first detection model, and sends the first target detection result to the intelligent vehicle terminal; the first detection model is the detection model corresponding to the first edge server 3021; ​​if the first edge server 3021 judges that it does not have the processing capability of the detection image, it sends the detection data to the central server 301;

[0100] The central server 301 is used to assign a second edge server 3022 to perform task processing for the detection image based on the real-time load information reported by each of the plurality of edge servers 302;

[0101] The second edge server 3022 is used to perform target detection on the detection image based on a second detection model deployed locally, and send the second target detection result to the first edge server 3021;

[0102] The first edge server 3021 is also used to send the second target detection result to the intelligent vehicle terminal; the second detection model is the detection model corresponding to the second edge server 3022; the second edge server 3022 is an edge server 302 other than the first edge server 3021.

[0103] In the embodiment of the present application, optionally,

[0104] The edge server 302 is further used to store the received detection image and the target detection result corresponding to the detection image as historical detection data, and periodically report the historical detection data to the central server;

[0105] The central server 301 is further used to perform the following processing on the historical detection data sent by each of the edge servers 302 to obtain a training data sample corresponding to each of the edge servers 302: the central server 301 determines, based on the acquired user feedback information on the historical detection data, the historical detection data for which the user feedback information indicates that the target detection result is correct as the first training data sample; randomly samples the remaining historical detection data except the first training data sample to obtain a first candidate data sample and a second candidate data sample, and verifies the first candidate data sample based on the locally deployed central detection model to obtain a first candidate data sample that passes the verification as the second training data sample, verifies the second candidate data sample based on the expert system to obtain a second candidate data sample that passes the verification as the third training data sample, and merges the first training data sample, the second training data sample and the third training data sample to obtain the training data sample;

[0106] The central server 301 is further used to send the training data samples to the corresponding edge servers 302, and the edge servers 302 perform iterative training on their respective locally deployed detection models according to the received training data samples.

[0107] In the embodiment of the present application, optionally,

[0108] The edge server 302 is also used to copy the locally deployed detection model to obtain a detection model copy, and iteratively train the detection model copy using the training data sample to obtain a trained detection model copy, extract model parameters of the trained detection model copy, deploy the model parameters in the detection model, and delete the trained detection model copy.

[0109] In the embodiment of the present application, optionally, the detection model deployed by each edge server 302 has the same model structure as the central detection model deployed by the central server 301;

[0110] The edge server 302 is also used to send the respective model parameters to the central server 301;

[0111] The central server 301 is further used to perform parameter fusion on the model parameters corresponding to each of the edge servers 302 to obtain target model parameters, and deploy the target model parameters in the central detection model.

[0112] In the embodiment of the present application, optionally,

[0113] The central server 301 is also used for each of the edge servers 301. The central server 301 calculates the sum of the number of candidate samples according to the first number of candidate data samples and the second number of candidate data samples corresponding to each of the edge servers 302, calculates the sum of the number of training samples according to the second number of training data samples and the third number of training samples corresponding to the edge server 302, and calculates the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server 302; and performs parameter fusion on the model parameters of each of the edge servers 302 according to the parameter weight of each of the edge servers 302 to obtain the target model parameters.

[0114] In the embodiment of the present application, optionally,

[0115] The central server 301 is further used to calculate the model parameter similarity between different edge servers 302 according to the model parameters corresponding to each edge server 302, and determine the similar edge server corresponding to each edge server 302 according to the model parameter similarity;

[0116] The central server 301 is further used to determine a candidate edge server that has processing capabilities for the detection image based on real-time load information reported by each of the plurality of edge servers 302; if there is one candidate edge server, the candidate edge server is used as the second edge server, and the detection data is assigned to the second edge server 3022 for task processing; if there are multiple candidate edge servers, it is queried whether there is a similar edge server corresponding to the first edge server 3021 among the candidate edge servers, and the second edge server 3022 is determined based on the query result, and the detection data is assigned to the second edge server 3022 for task processing.

[0117] In the embodiment of the present application, optionally, the central server 301 is further used for:

[0118] If there are multiple candidate edge servers belonging to the similar edge server, then according to the geographical location corresponding to the first edge server 3021, a candidate edge server closest to the first edge server 3021 is determined from the multiple candidate edge servers belonging to the similar edge server as the second edge server 3022;

[0119] If there is a candidate edge server belonging to the similar edge server, the edge server is used as the second edge server 3022;

[0120] If there is no candidate edge server belonging to the similar edge server, the distance score and the model parameter similarity score between the candidate edge server and the first edge server 3021 are obtained, and a total score calculation rule is determined according to the current driving environment in the detection data, and a total score calculation is performed on the distance score and the model parameter similarity score according to the total score calculation rule, and the candidate edge server with the highest total score is determined to be the second edge server 3022.

[0121] It should be noted that for other corresponding descriptions of the functional units involved in the 5G MEC task management system based on intelligent vehicle networking provided in the embodiment of the present application, please refer to Figure 1 to Figure 2 The corresponding description in the method will not be repeated here.

[0122] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0123] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0124] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0125] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0126] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0127] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0128] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A 5G MEC task management method based on intelligent vehicle networking, characterized in that: Applied in an MEC task management system, the MEC task management system includes a central server and multiple edge servers, and the method includes: The first edge server receives detection data sent by the intelligent vehicle terminal within the processing range of the first edge server, wherein the detection data includes a detection image, and the first edge server is any one of the plurality of edge servers; The first edge server determines whether it has the processing capability for the detection image based on the current load information; If the first edge server determines that it has the ability to process the detection image, it performs target detection on the detection image based on a locally deployed first detection model, and sends the first target detection result to the smart vehicle terminal; the first detection model is a detection model corresponding to the first edge server; If the first edge server determines that it does not have the ability to process the detection image, the detection data is sent to the central server, and the central server assigns a second edge server to perform task processing for the detection image based on the real-time load information reported by each of the multiple edge servers, so that the second edge server performs target detection on the detection image based on the second detection model deployed locally, and sends the second target detection result to the first edge server, and the first edge server sends the second target detection result to the smart vehicle terminal; the second detection model is the detection model corresponding to the second edge server; the second edge server is an edge server other than the first edge server.

2. The method according to claim 1, characterized in that The method further comprises: Each of the edge servers stores the received detection image and the target detection result corresponding to the detection image as historical detection data, and periodically reports the historical detection data to the central server; The central server performs the following processing on the historical detection data sent by each of the edge servers to obtain a training data sample corresponding to each of the edge servers: the central server determines, based on the acquired user feedback information on the historical detection data, the historical detection data for which the user feedback information indicates that the target detection result is correct as the first training data sample; randomly samples the remaining historical detection data except the first training data sample to obtain a first candidate data sample and a second candidate data sample, and verifies the first candidate data sample based on the locally deployed central detection model to obtain a first candidate data sample that passes the verification as the second training data sample, verifies the second candidate data sample based on the expert system to obtain a second candidate data sample that passes the verification as the third training data sample, and merges the first training data sample, the second training data sample and the third training data sample to obtain the training data sample; The central server sends the training data samples to the corresponding edge servers, and the edge servers iteratively train their respective locally deployed detection models according to the received training data samples.

3. The method according to claim 2, characterized in that The edge servers iteratively train the detection models deployed locally according to the received training data samples, including: The edge server copies the locally deployed detection model to obtain a detection model copy, and iteratively trains the detection model copy using the training data sample to obtain a trained detection model copy, extracts model parameters of the trained detection model copy, deploys the model parameters in the detection model, and deletes the trained detection model copy.

4. The method according to claim 3, characterized in that The detection model deployed by each of the edge servers has the same model structure as the central detection model deployed by the central server; After extracting the model parameters of the trained detection model copy, the method further includes: Each of the edge servers sends its own model parameters to the central server; The central server performs parameter fusion on the model parameters corresponding to each of the edge servers to obtain target model parameters, and deploys the target model parameters in the central detection model.

5. The method according to claim 4, characterized in that The central server performs parameter fusion on the model parameters corresponding to each edge server to obtain target model parameters, including: For each of the edge servers, the central server calculates the sum of the number of candidate samples according to the first number of candidate data samples and the second number of candidate data samples corresponding to each of the edge servers, calculates the sum of the number of training samples according to the second number of training data samples and the third number of training samples corresponding to the edge server, and calculates the ratio of the sum of the number of candidate samples to the sum of the number of training samples as the parameter weight of the edge server; The central server performs parameter fusion on the model parameters of each edge server according to the parameter weight of each edge server to obtain the target model parameters.

6. The method according to claim 4 or 5, characterized in that: After each edge server sends its own model parameter to the central server, the method further includes: The central server calculates the model parameter similarity between different edge servers according to the model parameters corresponding to each edge server, and determines a similar edge server corresponding to each edge server according to the model parameter similarity; Accordingly, the central server assigns the detection data to a second edge server for task processing based on the real-time load information reported by each of the plurality of edge servers, including: The central server determines a candidate edge server having processing capability for the detection image based on real-time load information reported by each of the plurality of edge servers; If there is one candidate edge server, use the candidate edge server as the second edge server, and assign the detection data to the second edge server for task processing; If there are multiple candidate edge servers, query whether there is a similar edge server corresponding to the first edge server among the candidate edge servers, determine the second edge server based on the query result, and assign the detection data to the second edge server for task processing.

7. The method according to claim 6, characterized in that The determining the second edge server based on the query result includes: If there are multiple candidate edge servers belonging to the similar edge server, determine, according to the geographical location corresponding to the first edge server, a candidate edge server closest to the first edge server among the multiple candidate edge servers belonging to the similar edge server as the second edge server; If there is a candidate edge server belonging to the similar edge server, use the edge server as the second edge server; If there is no candidate edge server belonging to the similar edge server, a distance score and a model parameter similarity score between the candidate edge server and the first edge server are obtained, and a total score calculation rule is determined according to the current driving environment in the detection data, and a total score is calculated for the distance score and the model parameter similarity score according to the total score calculation rule, and the candidate edge server with the highest total score is determined to be the second edge server.

8. A 5G MEC task management system based on intelligent Internet of Vehicles, characterized in that: The system comprises: A central server and multiple edge servers; A first edge server is used to receive detection data sent by an intelligent vehicle terminal within a processing range of the first edge server, wherein the detection data includes a detection image, and the first edge server is any one of the plurality of edge servers; judging whether it has the processing capability of the detection image based on current load information; if the first edge server judges that it has the processing capability of the detection image, performing target detection on the detection image based on a locally deployed first detection model, and sending a first target detection result to the intelligent vehicle terminal; the first detection model is a detection model corresponding to the first edge server; if the first edge server judges that it does not have the processing capability of the detection image, sending the detection data to the central server; The central server is used to assign a second edge server to perform task processing for the detection image based on the real-time load information reported by each of the plurality of edge servers; The second edge server is configured to perform target detection on the detection image based on a locally deployed second detection model, and send a second target detection result to the first edge server; The first edge server is also used to send the second target detection result to the intelligent vehicle terminal; the second detection model is the detection model corresponding to the second edge server; the second edge server is an edge server other than the first edge server.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images