A scenario-based algorithm configuration method and device
Through the cloud center, the comparison of multiple algorithm packages is performed on the sampled data of the terminal device, and the dynamic decision-making optimal algorithm package solves the problem that the static optimal algorithm in the existing technology cannot meet the real-time scenario analysis needs, and improves the user experience and analysis results. accuracy.
Patent Information
- Application Number
- CN202110867970.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-29
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2041-07-29
AI Technical Summary
The existing video analysis methods are too static when determining the optimal algorithm, which cannot meet the user's real-time needs for scene analysis, and rely on information such as the number and type of video devices, such as the information of non-analytical content, resulting in high implementation complexity.
The cloud center receives the sampled data sent by the terminal device, determines multiple algorithm packets corresponding to the data type, and decides the optimal algorithm packet through algorithm comparison, dynamically adjusts the algorithm to adapt to scene changes, avoid relying on the number and type of video devices, etc. information.
It realizes the user's real-time need for different scenarios to analyze different scenarios, improves user experience, and reduces the complexity of algorithm configuration, ensuring that the analysis results are in line with the actual scenario.
Smart Images

Figure CN113596158B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of Internet of Things technology, and in particular to a scenario-based algorithm configuration method and device. Background Art
[0002] As a highly integrated and comprehensive application of the new generation of information technology, the Internet of Things has the characteristics of strong penetration, great driving effect and good comprehensive benefits. It is another promoter of the development of the information industry after computers, the Internet and mobile communication networks. The application and development of the Internet of Things will help promote the transformation of production, life and social management methods towards intelligence, refinement and networking, and greatly improve the level of social management and public services.
[0003] At present, the Internet of Things in the fields of medical care, transportation, environmental protection and security in urban construction needs to accelerate the development of intelligence with various technologies based on artificial intelligence, and video / image intelligent analysis is an indispensable component. When existing video analysis is used for video surveillance, it can obtain scene information (including analysis targets, data volume of analysis targets and parameters affecting video devices), determine the number and type of video devices according to the scene information, and then obtain sample video sequences of analysis targets and device application additional information of video devices, and determine the optimal algorithm corresponding to the scene according to the number and type of video devices, sample video sequences and device application additional information, so as to perform video analysis according to the optimal algorithm.
[0004] It can be understood that this video analysis method is a static method for determining the optimal algorithm, which is determined based on the established parameters determined once, namely the number and type of video devices, sample video sequences, and additional information used by the devices. Once the optimal algorithm is determined, the optimal algorithm is no longer updated. In this way, the static optimal algorithm determined cannot meet the user's real-time needs for scene analysis. Moreover, the method for determining the optimal algorithm relies too much on information other than the content to be analyzed (sample video sequence), namely the number and type of video devices, and additional information used by the devices, which makes the implementation of determining the optimal algorithm more complicated. Summary of the invention
[0005] The embodiments of the present application provide a scenario-based algorithm configuration method and device, which can configure algorithm packages to edge domain nodes according to users' real-time needs for scenario analysis, so as to analyze data received from terminal devices according to the configured algorithm packages, thereby improving user experience and avoiding the problem of complex implementation caused by the participation of additional information of device applications in analysis, such as the number and type of video devices other than data obtained by terminal devices.
[0006] In order to achieve the above purpose, the embodiment of the present application adopts the following technical solution:
[0007] In a first aspect, a scenario-based algorithm configuration method is provided, including: a cloud center receives sampled data sent by a first edge domain node, and the cloud center determines multiple algorithm packages corresponding to the data type of the sampled data; the data type of the sampled data is used to reflect the scenario when the terminal device obtains the sampled data. The cloud center executes each algorithm package in the multiple algorithm packages on the sampled data to obtain an algorithm result corresponding to each algorithm package; the cloud center determines a target algorithm package from the multiple algorithm packages according to the algorithm result corresponding to each algorithm package, and the target algorithm package is used to analyze the data collected by the terminal device.
[0008] Therefore, the scenario-based algorithm configuration method provided by the present application can report the sampled data obtained from the terminal device through the edge domain node, and the cloud center performs algorithm comparison of multiple algorithm packages on the sampled data, and decides the optimal algorithm package, that is, the target algorithm package. Since the type of sampled data can reflect the scenario when the terminal device obtains the sampled data, when the scenario changes when the terminal device obtains the sampled data, the data type of the sampled data determined by the cloud center will also change with the scenario, and the multiple algorithm packages determined by the cloud center will also change with the data type, and the target algorithm package in the multiple algorithm packages finally determined by the cloud center from the multiple algorithm packages will also change. Therefore, in the present application, the target algorithm package decided by the cloud center can be different with the different scenarios when the terminal device obtains the data, which can meet the user's needs for real-time analysis of different scenarios and improve the user experience. Moreover, in the present application, when the cloud center decides the optimal algorithm package, it does not need to rely on the number and type of video devices other than the sampled data in the prior art and the additional information of the device application of the video device to participate in the decision-making, and the implementation complexity is low.
[0009] In one possible design, the sampled data is data obtained by extracting frames from the video data shot by the terminal device, or the sampled data is picture data captured by the terminal device. The sampled data here can be understood as real-time, or acquired at a fixed time, or acquired periodically. The terminal device can be understood as a camera connected to the first edge domain node. The camera can report the current scene data within the viewing angle, that is, the video frame data or the captured data, so that the cloud center can decide the optimal algorithm for the first edge domain node to process the current scene data, meet the user's real-time scene analysis needs, and improve the user experience.
[0010] In a possible design, the cloud center executes each algorithm package in multiple algorithm packages on the sampled data, and obtains the algorithm result corresponding to each algorithm package, including: the cloud center determines multiple evaluation indicators in the algorithm evaluation system corresponding to the data type of the sampled data; the cloud center refers to the multiple evaluation indicators, performs analysis on the sampled data through each algorithm package in the multiple algorithm packages, and obtains the values of the multiple evaluation indicators corresponding to each algorithm package; the cloud center determines the algorithm evaluation value corresponding to each algorithm package according to the values of the multiple evaluation indicators corresponding to each algorithm package, and uses the algorithm evaluation value corresponding to each algorithm package as the algorithm result corresponding to each algorithm package;
[0011] The cloud center determines the target algorithm package from multiple algorithm packages according to the algorithm results corresponding to each algorithm package, including: the cloud center determines the algorithm package with the highest algorithm evaluation value among the multiple algorithm packages as the target algorithm package according to the algorithm evaluation value corresponding to each algorithm package.
[0012] In other words, according to the type of sampling data, the optimal algorithm among multiple algorithms of this type can be determined through algorithm comparison, that is, algorithm evaluation, to meet the user's real-time demand for scene analysis and improve user experience.
[0013] In a possible design, the method further includes: the cloud center sends a target algorithm package and an algorithm analysis task to the first edge domain node, and the algorithm analysis task is used to instruct the first edge domain node to analyze the data received from the terminal device through the target algorithm package. Since the target algorithm package is determined according to the data type of the sampled data obtained by the terminal device, that is, determined according to the scene, when the first edge domain node performs the algorithm analysis task according to the target algorithm package, it can be more in line with the current scene of the terminal device, so that the algorithm analysis result is more in line with reality and meets the user's real-time scene analysis needs.
[0014] In a possible design, the method further includes: the cloud center determines the remaining computing power corresponding to the plurality of edge domain nodes managed by the cloud center, the plurality of edge domain nodes including the first edge domain node; the cloud center determines the second edge domain node according to the remaining computing power corresponding to the plurality of edge domain nodes, the second edge domain node being the edge domain node with the most remaining computing power among the plurality of nodes; the cloud center sends the target algorithm package and the algorithm analysis task to the second edge domain node, the algorithm analysis task being used to instruct the second edge domain node to analyze the data collected by the terminal device received from the first edge domain node through the target algorithm package. That is to say, after the cloud center determines the target algorithm package according to the sampled data collected by the terminal device reported by the first edge domain node, it is not in a hurry to directly send the target algorithm package to the first edge domain node, but can first determine the second edge domain node with the strongest remaining computing power among the plurality of edge domain nodes under the jurisdiction of the cloud center, and can send the target algorithm package and the algorithm analysis task to the second edge domain node. In this way, the purpose of reasonably using the computing power of the edge domain nodes can be achieved. Among them, the second edge domain node may also be the same node as the first edge domain node.
[0015] In one possible design, the cloud center determines the remaining computing power corresponding to each of the multiple edge domain nodes managed by the cloud center, including: the cloud center determines the remaining computing power corresponding to each edge domain node according to the hardware resource performance, data access volume, and the number of terminal devices connected to each edge domain node in the multiple edge domain nodes. Among them, the hardware resource performance can be understood as the remaining computing power of the CPU and GPU of the edge domain node. The data access volume can be used to reflect the total number of real-time algorithm executions of the edge domain node. The number of terminal devices connected to the edge domain node can be understood as the number of terminal devices that establish a communication connection with the edge domain node.
[0016] In a second aspect, a scenario-based algorithm configuration method is provided, comprising:
[0017] The edge domain node sends the sampled data to the cloud center; the edge domain node receives the first target algorithm package sent by the cloud center, where the first target algorithm package is determined by the cloud center executing each of the multiple algorithm packages on the sampled data, and obtaining the analysis result corresponding to each algorithm package; the multiple algorithm packages are determined by the cloud center according to the data type of the sampled data; the type of the sampled data is used to reflect the scenario when the terminal device obtains the sampled data; the edge domain node loads the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package.
[0018] Therefore, for the edge domain nodes, the edge domain nodes can analyze the data received from the terminal device according to the first target algorithm package determined by the cloud center that changes with the scene of the sampled data. That is, the first target algorithm package can be different depending on the results of the algorithm comparison, which can meet the user's real-time scene analysis needs and improve the user experience.
[0019] In a possible design, the sampled data is data obtained by extracting frames from video data captured by the terminal device, or the sampled data is picture data captured by the terminal device. The beneficial effects of this design can be seen in the description of the first aspect.
[0020] In a possible design, the edge domain node loads the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package, including: the edge domain node determines whether the first target algorithm package has been loaded locally at the edge domain node; if it is determined that the first target algorithm package is not integrated locally at the edge domain node, the edge domain node loads the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package. That is, it is possible that the optimal algorithm (first target algorithm package) determined by the cloud center is the algorithm currently being loaded and run by the edge domain node, and at this time the edge domain may not load the first target algorithm package.
[0021] In a possible design, the edge domain node receives the second target algorithm package and algorithm analysis task sent by the cloud center. The second target algorithm package is determined by the cloud center based on the sampled data sent by other edge domain nodes, and the algorithm analysis task includes the terminal device information to be analyzed; the edge domain node obtains the data collected by the terminal device corresponding to the terminal device information to be analyzed from other edge domain nodes; the edge domain node loads the second target algorithm package to analyze the data received from other edge nodes through the second target algorithm package. In other words, after an edge domain node reports the data collected from the terminal device to the cloud center, it is possible that the cloud center does not directly send the target algorithm package to the edge domain node, but determines the optimal edge domain node that can execute the target algorithm package, and sends the target algorithm package to the optimal edge domain node for recording. In this way, the computing power of each edge domain node can be balanced, and an edge domain node may have a mixture of algorithm packages under multiple scenarios, which can realize the computing power sharing of edge domain node clusters based on mixed scenarios to a certain extent.
[0022] According to a third aspect, a cloud center is provided, including: a communication module for receiving sampling data sent by a first edge domain node; a processing module for determining a plurality of algorithm packages corresponding to a data type of the sampling data, the data type of the sampling data being used to reflect a scenario when a terminal device acquires the sampling data; the processing module is also used to execute each of the plurality of algorithm packages on the sampling data to obtain an algorithm result corresponding to each algorithm package; the processing module is also used to determine a target algorithm package from the plurality of algorithm packages according to the algorithm result corresponding to each algorithm package, the target algorithm package being used to analyze data collected by the terminal device.
[0023] The beneficial effects of the third aspect can refer to the description of the first aspect.
[0024] In a possible design, the sampled data is data obtained by extracting frames from video data captured by the terminal device, or the sampled data is picture data captured by the terminal device.
[0025] In one possible design, the processing module is specifically used to: determine multiple evaluation indicators in an algorithm evaluation system corresponding to the data type of the sampling data; refer to the multiple evaluation indicators, execute each of the multiple algorithm packages on the sampling data, and obtain the values of the multiple evaluation indicators corresponding to each algorithm package; determine the algorithm evaluation value corresponding to each algorithm package according to the values of the multiple evaluation indicators corresponding to each algorithm package, and use the algorithm evaluation value corresponding to each algorithm package as the algorithm result corresponding to each algorithm package; according to the algorithm evaluation value corresponding to each algorithm package, determine the algorithm package with the highest algorithm evaluation value among the multiple algorithm packages as the target algorithm package.
[0026] In a possible design, the transceiver module is also used to: send a target algorithm package and an algorithm analysis task to the first edge domain node, where the algorithm analysis task is used to instruct the first edge domain node to analyze data received from the terminal device through the target algorithm package.
[0027] In a possible design, the processing module is also used to: determine the remaining computing power corresponding to multiple edge domain nodes managed by the cloud center, the multiple edge domain nodes including a first edge domain node; determine the second edge domain node according to the remaining computing power corresponding to the multiple edge domain nodes, the second edge domain node is the edge domain node with the largest remaining computing power among the multiple edge domain nodes; the transceiver module is also used to: send a target algorithm package and an algorithm analysis task to the second edge domain node, the algorithm analysis task is used to instruct the second edge domain node to analyze the data collected by the terminal device received from the first edge domain node through the target algorithm package.
[0028] In one possible design, the processing module is specifically used to determine the remaining computing power corresponding to each edge domain node according to the hardware resource performance, data access volume and the number of connected terminal devices corresponding to each edge domain node among the multiple edge domain nodes.
[0029] In a fourth aspect, an edge domain node is provided, and the edge domain node includes: a communication module, which is used to send sampling data to a cloud center; a transceiver module, which is also used to receive a first target algorithm package sent by the cloud center, the first target algorithm package is determined by the cloud center executing each of multiple algorithm packages on the sampling data, and the algorithm result corresponding to each algorithm package is obtained; the multiple algorithm packages are determined by the cloud center according to the data type of the sampling data; the type of the sampling data is used to reflect the scenario when the terminal device obtains the sampling data; a processing module, which is used to load the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package.
[0030] The beneficial effects of the fourth aspect can refer to the description of the second aspect.
[0031] In a possible design, the sampled data is data obtained by extracting frames from video data captured by the terminal device, or the sampled data is picture data captured by the terminal device.
[0032] In a possible design, the processing module is specifically used to: determine whether the edge domain node has loaded the first target algorithm package locally;
[0033] If it is determined that the edge domain node has not locally loaded the first target algorithm package, the first target algorithm package is loaded to analyze the data received from the terminal device through the first target algorithm package.
[0034] In one possible design, the communication module is also used to receive a second target algorithm package and an algorithm analysis task sent by the cloud center, where the second target algorithm package is determined by the cloud center based on sampling data sent by other edge domain nodes, and the algorithm analysis task includes terminal device information to be analyzed; the processing module is also used to obtain data collected by terminal devices corresponding to the terminal device information to be analyzed from other edge domain nodes; the processing module is also used to load the second target algorithm package to analyze the data received from other edge nodes through the second target algorithm package.
[0035] In a fifth aspect, a cloud center is provided, the cloud center comprising: a memory and a processor. The memory and the processor are coupled. The memory is used to store computer program code, the computer program code comprising computer instructions. The transceiver is used to receive data and send data. When the processor executes the computer instructions, the cloud center executes any one of the scenario-based algorithm configuration methods provided in the first aspect or its corresponding possible design.
[0036] In a sixth aspect, an edge domain node is provided, the edge domain node comprising: a memory and a processor. The memory and the processor are coupled. The memory is used to store a computer program code, the computer program code comprising a computer instruction. The transceiver is used to receive data and send data. When the processor executes the computer instruction, the edge domain node executes any scenario-based algorithm configuration method provided in the second aspect or its corresponding possible design.
[0037] In a seventh aspect, the present application provides a chip system, which is applied to a cloud center. The chip system includes one or more interface circuits and one or more processors. The interface circuit and the processor are interconnected by a line; the interface circuit is used to receive a signal from a memory of the cloud center and send the signal to the processor, and the signal includes a computer instruction stored in the memory. When the processor executes the computer instruction, the cloud center executes the scenario-based algorithm configuration method provided by the first aspect or its corresponding possible design.
[0038] In an eighth aspect, the present application provides a chip system, which is applied to an edge domain node. The chip system includes one or more interface circuits and one or more processors. The interface circuit and the processor are interconnected by a line; the interface circuit is used to receive a signal from a memory of the edge domain node and send the signal to the processor, the signal including a computer instruction stored in the memory. When the processor executes the computer instruction, the edge domain node executes a scenario-based algorithm configuration method as provided in the second aspect or its corresponding possible design.
[0039] In a ninth aspect, the present application provides a computer-readable storage medium, which includes computer instructions. When the computer instructions are executed on a computer, the computer executes the method provided in the first aspect or its corresponding possible design.
[0040] In a tenth aspect, the present application provides a computer-readable storage medium, the computer-readable storage medium comprising computer instructions, when the computer instructions are executed on a computer, the computer executes the method provided in the second aspect or its corresponding possible design
[0041] In an eleventh aspect, the present application provides a computer program product, which includes computer instructions. When the computer instructions are executed on a computer, the computer executes the method provided in the first aspect or its corresponding possible design.
[0042] In a twelfth aspect, the present application provides a computer program product, which includes computer instructions. When the computer instructions are executed on a computer, the computer executes the method provided in the second aspect or its corresponding possible design.
[0043] It can be understood that any of the cloud centers, edge domain nodes, chip systems, computer-readable storage media or computer program products provided above can be applied to the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods and will not be repeated here.
[0044] These and other aspects of the present application will become more apparent from the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 A schematic diagram of a network architecture provided for an embodiment of the present application;
[0046] Figure 2 A schematic diagram of a scenario-based algorithm configuration method flow chart provided in an embodiment of the present application;
[0047] Figure 3 A schematic diagram of a method for selecting an edge domain node provided in an embodiment of the present application;
[0048] Figure 4 A schematic diagram of a software framework of a cloud center and a schematic diagram of a software framework of each node in an edge domain provided in an embodiment of the present application;
[0049] Figure 5 A schematic diagram of a process flow of a method for managing terminal devices by a cloud center provided in an embodiment of the present application;
[0050] Figure 6 A schematic diagram of a process flow of a method for managing a terminal device by a node in an edge domain provided in an embodiment of the present application;
[0051] Figure 7 A possible schematic diagram of the composition of a server in a cloud center provided in an embodiment of the present application;
[0052] Figure 8 A possible schematic diagram of the composition of a server in a cloud center and a server in an edge domain provided in an embodiment of the present application;
[0053] Fig. 9 A possible composition diagram of an edge domain node provided in an embodiment of the present application. DETAILED DESCRIPTION
[0054] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two.
[0055] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this embodiment, unless otherwise specified, "plurality" means two or more.
[0056] The embodiments of the present application can be applied to scenarios where a cloud center in the Internet of Things manages edge domain nodes, for example, it can be applied to scenarios where edge domain nodes are managed to analyze video data or image data.
[0057] In the scenario where the cloud center manages the edge domain nodes to analyze video data or image data, the optimal algorithm determined by the static optimal algorithm determination method when the edge domain nodes perform video analysis cannot meet the user's real-time scene analysis needs. This application proposes a scene-based algorithm configuration method that enables the edge domain nodes to support algorithm integration and expansion. That is, the edge domain node can send sampled data to the cloud center, and the cloud center can determine multiple algorithm packages corresponding to the data type of the sampled data. The data type of the sampled data is used to reflect the scene when the terminal device obtains the sampled data. The cloud center executes each algorithm package in the multiple algorithm packages on the sampled data, and obtains the target algorithm package by comparing the algorithm results obtained by executing the multiple algorithm packages. The target algorithm package can be understood as the optimal algorithm package among the multiple algorithm packages obtained by comparison. In this way, since the type of sampled data can reflect the scene when the terminal device obtains the sampled data, when the scene when the terminal device obtains the sampled data changes, the data type of the sampled data determined by the cloud center will also change with the scene, and the multiple algorithm packages determined by the cloud center will also change with the data type, and the target algorithm package in the multiple algorithm packages finally determined by the cloud center from the multiple algorithm packages will also change.
[0058] Therefore, in this application, the target algorithm package decided by the cloud center can be different depending on the scenario when the terminal device obtains data. The target algorithm package decided by the cloud center can meet the user's real-time needs for scene analysis and improve the user experience. Moreover, in this application, when the cloud center decides the target algorithm package, it does not need to rely on the number and type of video devices other than the sampled data and the additional information of the device application of the video device to participate in the decision-making, and the implementation complexity is relatively low.
[0059] It should be noted that the algorithms involved in this application, such as multiple algorithm packages, each algorithm package and target algorithm package, etc., can be applied not only to the calculation of media data, but also to the calculation of non-media data. In other words, the data involved in this application can be media data or non-media data. When it is media data, the sampled data of this application can be understood as media sampled data; when it is non-media data, the sampled data of this application can be understood as non-media sampled data.
[0060] Exemplarily, the media data or media sampling data may be, for example, video data or picture data acquired through a camera.
[0061] Non-media sampling data may be, for example, intelligent structured data or sensor data other than camera data, including smoke sensor data, door magnetic switch sensor data, water level sensor data, temperature and humidity sensor data, natural gas sensor data, noise sensor data, wind speed sensor data, ambient light sensor data, sunlight sensor data, infrared sensor data, and ultraviolet sensor data.
[0062] The scenario-based algorithm configuration method proposed in this application can be applied to Figure 1 The network architecture shown in FIG. 1 can be applied to the Internet of Things. The network architecture can include a cloud center, multiple edge domain nodes in the edge domain (e.g. Figure 1 In the edge domain node 1, edge domain node 2, ..., edge domain node n, and multiple terminal devices managed by each edge domain node (terminal device 1, terminal device 2, ..., terminal device m). m and n are integers greater than 1.
[0063] In this application, the cloud center, which can be referred to as "cloud", can be a public cloud center for Internet scenarios or a private cloud center for intranets / private networks; the cloud center focuses on business data fusion and multi-dimensional analysis applications of big data. It should be noted that the functions of the cloud center can be implemented by one device or by multiple devices, and this embodiment of the application does not limit this. It should be understood that in the embodiment of the application, the steps performed by the cloud center are specifically performed by one or more devices in the cloud center.
[0064] The edge domain, which can be referred to as "edge", can reflect the logical position relative to the cloud center. The cloud center can be referred to as the edge domain below, and the communication network can be between the edge domain and the cloud center. The edge domain focuses on the aggregation, storage, processing and intelligent application of perception data. In this application, the edge domain can include multiple edge domain nodes, and each edge domain node can be used to perform algorithm analysis on the data sent by multiple terminal devices managed by the edge domain node.
[0065] Terminal devices focus on the collection of multi-dimensional perception data and front-end intelligent processing. For example, terminal devices can be cameras, sensor devices and smart devices, such as enhanced cameras, smart cameras, IoT sensor devices (such as Global Positioning System (GPS), Wireless Fidelity (Wifi) probes, access control alarm products, smart helmets, environmental monitoring and card-entry mobile terminals, etc.), industrial smart devices, robots and drones. For example, cameras can perform video monitoring of the environment, people, vehicles and objects within the field of view, and transmit the video data collected by monitoring to the edge domain node that manages the camera for algorithm analysis.
[0066] That is, in the network architecture, data can be transferred from the terminal device to the edge domain, which can realize "gathering from edge to domain"; data can be transferred from the edge domain to the cloud center, which can realize "data entering the cloud". The edge domain and cloud center can be multi-level and multi-class. Depending on different applications, the data gathered in the edge domain and the data transmitted to the cloud center may be different in format and content.
[0067] Applying this network architecture, the scenario-based algorithm configuration method provided in this application can refer to Figure 2 , the method comprising:
[0068] 201. The terminal device obtains sampled data and sends the sampled data to the first edge domain node. That is, the first edge domain node receives the sampled data sent by the terminal device.
[0069] In some embodiments, the sampled data may be obtained by sampling the data acquired in real time by the terminal device, and the terminal device then sends the sampled data to the first edge domain node.
[0070] In some embodiments, the sampled data may also be data acquired in real time and sent by the terminal device to the first edge domain node, and the first edge domain node samples the data received in real time.
[0071] Exemplarily, when the terminal device is a camera, the sampling data can be understood as media sampling data, for example, the sampling data is data obtained by extracting frames from video data shot by the terminal device, or the sampling data is picture data captured by the camera during real-time shooting.
[0072] When the sampling data is video data, the sampling data may be obtained by extracting frames from the video data obtained by real-time shooting by the camera, and the camera then sends the sampling data to the first edge domain node; or, the sampling data may be obtained by extracting frames from the video data obtained by real-time shooting by the camera to the first edge domain node, and the first edge domain node extracts frames from the received video data.
[0073] It can be understood that the sampling data can be obtained regularly, or periodically, by the terminal device or the first edge domain node.
[0074] It is understood that when the terminal device is a camera, the video data or image data may be data corresponding to a scene within the viewing angle captured by the camera, such as objects in the scene including pedestrians, vehicles, animals, buildings, signs, flowers, plants, and trees.
[0075] 202. The first edge domain node sends sampled data to the cloud center, that is, the cloud center receives the sampled data sent by the first edge domain node.
[0076] When the first edge domain node receives the sampled data regularly or periodically, the received sampled data may be sent to the cloud center for analysis.
[0077] 203. The cloud center determines a plurality of algorithm packages corresponding to data types of the sampled data, where the data types of the sampled data are used to reflect the scenario when the terminal device acquires the sampled data.
[0078] When the cloud center receives sampled data, it can first determine the data type of the sampled data.
[0079] In some embodiments, for example, the terminal device is a camera, which is located at the school gate, and the scene around the road at the school gate can be captured within its field of view. Taking the sampled data as image data as an example, the cloud center can first determine the number of people in the image data through a face detection algorithm. When the number of students and parents at the school gate is large during the school dismissal period, for example, when a certain threshold is reached, the cloud center can believe that there is a gathering of people in the scene corresponding to the image data. Therefore, the cloud center can determine that the type of the image data is a gathering of people type. In other words, the gathering of people type can reflect that when the camera acquires the image data, there is a gathering of people in the scene.
[0080] When the cloud center determines that the data type of the image data is a people gathering type, the multiple algorithm packages corresponding to the people gathering type may be, for example, multiple people violent motion detection algorithms. Each people violent motion detection algorithm can be used to detect whether there may be a situation of people chasing each other, or people fighting, etc.
[0081] Alternatively, the multiple algorithm packages corresponding to the type of personnel gathering may be, for example, multiple gait detection algorithms. Gait detection algorithm is a non-contact biometric recognition technology. For example, the Condensation algorithm can be used to perform pedestrian analysis in combination with prior knowledge such as human body models, motion models, and motion constraints. Then, the angle change trajectory of the main joints of the human body is obtained from the analysis results. After structural and temporal normalization, these trajectories are used as dynamic features for identity recognition. For example, wanted persons can be identified from a crowd.
[0082] It should be understood that the reason why there are multiple algorithms for detecting violent movements of people or multiple algorithms for detecting gait in the cloud center is that different algorithm developers design different algorithms.
[0083] In addition, the multiple algorithm packages corresponding to the types of people gatherings may be, for example, multiple face image recognition and comparison algorithms, which can identify special people from the crowd, etc.; or the multiple algorithm packages corresponding to the types of people gatherings may be, for example, multiple face detection algorithms, which can detect the positions of people in the crowd in the scene, etc.
[0084] In some embodiments, taking the above-mentioned camera shooting the scene around the road in front of the school as an example, when the cloud center detects that there are fewer people at the school gate and more passing vehicles during non-school hours, the cloud center can believe that there are more vehicles in the scene corresponding to the image data. Therefore, the cloud center can determine that the type of the image data is a vehicle gathering type. In other words, the vehicle gathering type can reflect that when the camera acquires the image data, vehicles gather in the scene.
[0085] When the cloud center determines that the data type of the image data is a vehicle aggregation type, the multiple algorithm packages corresponding to the vehicle aggregation type may be, for example, multiple solid line lane change detection algorithms. Each solid line lane change detection algorithm can be used to detect whether there is a vehicle crossing a solid line to change lanes, so as to obtain driving conditions that do not comply with traffic regulations.
[0086] Similarly, the reason why the cloud center has multiple solid line lane change detection algorithms is that different algorithm developers design different algorithms.
[0087] It should be noted that when the cloud center analyzes the sampled data, it can be done under the assumption that the service configuration and computing power of each node under the cloud center are similar. Among them, similar service configuration can be understood, for example, that each node in the edge domain is a server, and the hardware parameters such as the central processing unit (CPU) and graphics processing unit (GPU) of each server are the same; similar computing power can be understood, for example, that the computing power of the CPU and the computing power of the GPU are similar.
[0088] 204. The cloud center executes each algorithm package of the multiple algorithm packages on the sampled data to obtain an algorithm result corresponding to each algorithm package.
[0089] In some embodiments, the cloud center may evaluate each algorithm package in the multiple algorithm packages (algorithm sets) determined by the cloud center through an algorithm evaluation system corresponding to the data type of the sampling data to determine the algorithm package with the highest evaluation value among the multiple algorithm packages.
[0090] In some embodiments, specifically, the cloud center may first determine multiple evaluation indicators in the algorithm evaluation system corresponding to the data type of the sampled data, refer to the multiple evaluation indicators, calculate the sampled data through each algorithm package in the multiple algorithm packages, and obtain the values of multiple evaluation indicators corresponding to each algorithm package. The cloud center then determines the algorithm evaluation value corresponding to each algorithm package based on the values of the multiple evaluation indicators corresponding to each algorithm package, and uses the algorithm evaluation value corresponding to each algorithm package as the algorithm result corresponding to each algorithm package.
[0091] Exemplarily, assuming that the data type of the sampled data is a personnel gathering type, therefore, after the cloud center determines through step 203 that the multiple algorithm packages corresponding to this type are multiple face detection algorithm packages, the cloud center can first determine the multiple evaluation indicators in the algorithm evaluation system corresponding to this type. It can be understood that when the algorithm evaluation system is an algorithm evaluation system for face detection, the goal of face detection is to find the corresponding positions of all faces in the picture, and the output of the algorithm is the coordinates of the circumscribed rectangle of the face in the picture, which may also include posture information such as tilt angle. At this time, multiple evaluation indicators for evaluating the quality of a face detection algorithm package may include detection rate, false alarm rate, frame rate (Frames Per Second, FPS), intersection-over-Union (Intersection-over-Union, IoU) and other indicators.
[0092] The detection rate can be understood as: the number of detected faces / the number of all faces in the picture, the number of detected faces can be understood as the number of correctly detected faces, and the number of all faces in the picture can be understood as the total number of faces that should be detected.
[0093] The false alarm rate is understood as: number of false alarms / number of all non-face scanning windows in the image.
[0094] FPS can be understood as the number of images that the face detection algorithm can process in one second, that is, FPS is used to evaluate the speed of the face detection algorithm.
[0095] IoU can be understood as the overlap rate between the candidate bound and the ground truth bound in the face's bounding box, that is, the ratio of their intersection to their union. The ideal situation is complete overlap, that is, the ratio is 1.
[0096] Therefore, referring to these four evaluation indicators, the sampled data can be calculated by each face detection algorithm package in a plurality of face detection algorithm packages (face detection algorithm set) to obtain the values of the four evaluation indicators corresponding to each face detection algorithm package. The cloud center then determines the algorithm evaluation value corresponding to each face detection algorithm package based on the values of the four evaluation indicators corresponding to each face detection algorithm package, and uses the algorithm evaluation value corresponding to each face detection algorithm package as the algorithm result corresponding to each algorithm package.
[0097] Among them, according to the values of the four evaluation indicators corresponding to each face detection algorithm package, determining the algorithm evaluation value corresponding to each face detection algorithm package can be understood as, referring to the scoring standards corresponding to the four evaluation indicators, scoring each of the four evaluation indicators of each face detection algorithm, and obtaining a comprehensive score value for each face detection algorithm, and using the comprehensive score value as the algorithm evaluation value corresponding to the face detection algorithm package.
[0098] For another example, when the multiple algorithm packages are multiple face image recognition and comparison algorithms, the corresponding evaluation indicators for the multiple face image recognition and comparison algorithm packages may include false rejection rate (False Rejection Rate, FRR) and false acceptance rate (False Acceptance Rate, FAR). Among them, FRR can be understood as the proportion of false rejection cases in all similar matching cases. For example, if two samples are of the same type (the same person), but are mistakenly regarded as different types (not the same person) by the system, it is a false rejection case. FAR can be understood as the proportion of false acceptance cases in all different matching cases. For example, if two samples are different types (not the same person), but are mistakenly regarded as the same type (the same person) by the system, it is a false acceptance case. Similar to the face detection algorithm package, the implementation method of using multiple face recognition algorithm packages to analyze the image data and obtain the analysis results corresponding to each face recognition algorithm package can refer to the implementation method of the above-mentioned face detection algorithm package.
[0099] 205. The cloud center determines the target algorithm package from the multiple algorithm packages according to the algorithm results corresponding to each algorithm package.
[0100] It can be understood that the target algorithm package is used by the edge domain node to analyze the data collected by the terminal device. For example, when the target algorithm package is a person violent movement detection algorithm, it can be used to analyze the person in the video captured by the camera to determine whether the person is violently moving.
[0101] In some embodiments, the cloud center may determine the target algorithm package from multiple algorithm packages according to the algorithm results corresponding to each algorithm package, which may include:
[0102] The cloud center determines the algorithm package with the highest algorithm evaluation value among multiple algorithm packages as the target algorithm package based on the algorithm evaluation value corresponding to each algorithm package.
[0103] For example, when the comprehensive score of each face detection algorithm is obtained according to the scoring criteria for the face detection algorithm package, the face detection algorithm with the highest comprehensive score can be used as the target face detection algorithm. It can be understood that the face detection algorithm with the highest comprehensive score is more suitable for analyzing the image data captured in the current scene than other algorithms in the set of face detection algorithm packages, that is, the target face detection algorithm is more accurate in face detection of the current scene, and is more conducive to the edge domain node to perform the face detection algorithm analysis task in the current shooting scene of the camera.
[0104] 206. The cloud center sends the target algorithm package and the algorithm analysis task to the first edge domain node. That is, the first edge domain node receives the target algorithm package and the algorithm analysis task sent by the cloud center.
[0105] The target algorithm package can be understood as the algorithm framework used by the first edge domain node to perform the algorithm analysis task. The algorithm analysis task can be understood as a command instructing the first edge domain node to execute the target algorithm package, that is, instructing the first edge domain node to analyze the data received from the terminal device through the target algorithm package.
[0106] Exemplarily, the cloud center may send an algorithm integration instruction to the first edge domain node, the algorithm integration instruction including a target algorithm package (e.g., a target face detection algorithm package) or an algorithm analysis task, the target algorithm package or the algorithm analysis task may include an identifier of the target algorithm package, the identifier may include, for example, a version number or a serial number of the target algorithm package, etc. Alternatively, the identifier of the target algorithm package is indicated in a field other than the target algorithm package and the algorithm analysis task.
[0107] 207. The first edge domain node loads a target algorithm package to analyze data received from the terminal device through the target algorithm package.
[0108] In some embodiments, it is possible that the target algorithm package determined by the cloud center through algorithm comparison is already the algorithm package currently running on the first edge domain node. In this case, the first edge domain node does not need to load the target algorithm package again.
[0109] Exemplarily, when the first edge domain node receives the algorithm integration instruction, the first edge domain node determines whether the target algorithm package has been loaded locally on the first edge domain node; if it is determined that the target algorithm package has not been loaded locally on the first edge domain node, the first edge domain node loads the target algorithm package to analyze the data received from the terminal device through the target algorithm package. For example, the first edge domain node can determine whether the target algorithm package has been loaded and run locally on the first edge domain node based on fields such as the version number or serial number of the target algorithm package. If it is determined that it has been loaded and run, the first edge domain node does not process it. If it is determined that it has not been loaded and run, the first edge domain node can load the received target algorithm package to use the target algorithm package to analyze the data received from the terminal device.
[0110] It can be understood that in this application, the data received from the terminal device is the scene data received by the first edge domain node from the terminal device and acquired by the terminal device in real time. The sampled data is obtained by the terminal device or the first edge domain node by periodically sampling the real-time scene data, such as the above-mentioned video frame extraction data or captured picture data.
[0111] Exemplarily, the algorithm analysis task can be understood as, for example, the first edge domain node performs algorithm analysis on the video data captured by the camera using the face image recognition and comparison algorithm package to determine the face within the camera's viewing angle, etc. This face image recognition and comparison algorithm package can be used in security monitoring, human-document comparison, human-computer interaction, social interaction, and entertainment, for example.
[0112] Therefore, the scenario-based algorithm configuration method provided by the present application can report the sampled data obtained from the terminal device through the edge domain node, and the cloud center performs algorithm comparison of multiple algorithm packages on the sampled data, and decides the optimal algorithm package, that is, the target algorithm package. Since the type of sampled data can reflect the scenario when the terminal device obtains the sampled data, when the scenario changes when the terminal device obtains the sampled data, the data type of the sampled data determined by the cloud center will also change with the scenario, and the multiple algorithm packages determined by the cloud center will also change with the data type, and the target algorithm package in the multiple algorithm packages finally determined by the cloud center from the multiple algorithm packages will also change. Therefore, in the present application, the target algorithm package decided by the cloud center can be different with the different scenarios when the terminal device obtains the sampled data, meeting the user's needs for real-time analysis of different scenarios and improving the user experience. Moreover, in the present application, when the cloud center decides the optimal algorithm package, it does not need to rely on the number and type of video devices other than the sampled data in the prior art and the additional information of the device application of the video device to participate in the decision-making, and the implementation complexity is relatively low.
[0113] In addition, it can be understood that among the multiple edge domain nodes in the edge domain managed by the cloud center, the algorithm analysis tasks performed by each edge domain node may be different, so the remaining computing power of each edge domain node may also be different. Therefore, considering that the remaining computing power of multiple edge domain nodes is different, before executing step 206, the cloud center can first estimate the remaining computing power of each edge domain node in the multiple edge domain nodes to determine the edge domain node with the most remaining computing power, and send the target algorithm package to the edge domain node with the most remaining computing power. It can be understood that the edge domain node with the most remaining computing power here may be the first edge domain node, or it may be another edge domain node other than the first edge domain node.
[0114] Therefore, if Figure 3 As shown, a method for selecting edge domain nodes is provided. After step 205, step 206 may not be performed, but step 208 may be performed instead.
[0115] 208. The cloud center determines the remaining computing power corresponding to each of the plurality of edge domain nodes managed by the cloud center, wherein the plurality of edge domain nodes include the first edge domain node.
[0116] In some embodiments, the cloud center can determine the corresponding remaining computing power in each edge domain node based on the hardware resource performance, data access volume, and the number of connected terminal devices corresponding to each node in the multiple edge domain nodes.
[0117] For example, for any edge domain node, its hardware resource performance V server It can be understood as the remaining computing power of the CPU and GPU of the edge domain node. It can be understood that if the CPU and / or GPU in the edge domain node does not support the operation of the target algorithm, the remaining computing power V of the edge domain node can be directly determined. computer is 0;
[0118] Data access volume V data , can be used to reflect the total number of real-time algorithm executions of edge domain nodes, or to reflect the real-time execution of algorithms. That is, the data access volume V of edge domain nodes data The larger it is, the greater the computing power consumption of the edge domain node.
[0119] The number of terminal devices connected to the edge domain node V device ,It can be understood that under normal circumstances, the more real-time and effective terminal devices access to the edge ,domain nodes, the fewer remaining computing resources of the edge ,domain nodes.
[0120] Combine the above three indicators to calculate the remaining computing power V of the edge domain node computer It can be expressed as:
[0121] V computer =f(V server .βserver +V data .β data +V device .β device )
[0122] Among them, f() represents the calculation of the remaining computing power V computer The formula, β server Indicates the hardware resource performance V server The weight coefficient, β data Indicates the data access volume V data The weight coefficient, β device Indicates the number of terminal devices connected to the edge domain node V device The weight coefficient of .
[0123] It is understandable that the remaining computing power V computer The larger the value, the smaller the remaining computing power of the edge domain node.
[0124] 209. The cloud center determines a second edge domain node according to the remaining computing power corresponding to each of the plurality of edge domain nodes. The second edge domain node is an edge domain node having the largest remaining computing power among the plurality of edge domain nodes.
[0125] For example, the cloud center can arrange the remaining computing power corresponding to multiple edge domain nodes in order of size, and take V computer The edge domain node with the smallest value is the second edge domain node, that is, the second edge domain node is considered to have the largest remaining computing power. At this time, the second edge domain node here may be the first edge domain node mentioned above, or may be another edge domain node other than the first edge domain node.
[0126] If the cloud center determines that the second edge domain node is the first edge domain node, after step 209, step 206 can be continued. If the cloud center determines that the second edge domain node is not the first edge domain node, that is, it is not the edge domain node that sends sampled data to the cloud center, step 210 can be continued.
[0127] 210. The cloud center sends the target algorithm package and the algorithm analysis task to the second edge domain node. That is, the second edge domain node receives the target algorithm package and the algorithm analysis task sent by the cloud center.
[0128] In step 210, the algorithm analysis task may include address information of the first edge domain node.
[0129] Similar to step 206, when the second edge domain node receives the target algorithm package and the algorithm analysis task, it can also first determine whether the second edge domain node has already loaded and run the algorithm framework of the target algorithm package locally. The determination method can refer to step 206.
[0130] 211. The second edge domain node loads a target algorithm package to analyze data received from the terminal device through the target algorithm package.
[0131] For the specific implementation of step 211 , reference may be made to the implementation of the first edge domain node in step 207 .
[0132] It can be understood that, assuming that the above-mentioned first edge domain node (or second edge domain node) currently stores the first target algorithm package and the corresponding algorithm analysis task, it is possible that it also receives the second target algorithm package and the corresponding algorithm analysis task sent by the cloud center, and the second target algorithm package is determined by the cloud center based on the sampling data sent by other edge domain nodes, and the algorithm analysis task includes the terminal device information to be analyzed; at this time, the first edge domain node (or second edge domain node) can obtain the data collected by the terminal device corresponding to the terminal device information to be analyzed from other edge domain nodes; the first edge domain node (or second edge domain node) loads the second target algorithm package to analyze the data received from other edge nodes through the second target algorithm package.
[0133] Therefore, when the cloud center sends the target algorithm package to the edge domain node, it can first compare the remaining computing power of each edge domain node, and select the edge domain node with the largest remaining computing power to execute the algorithm analysis task of the target algorithm package, so as to achieve the reasonable use of the edge domain node computing power. In this way, when determining the edge domain node to receive the target algorithm package based on the remaining computing power, an edge domain node may have a mixture of algorithm packages under multiple scenarios, which can realize the computing power sharing of the edge domain node cluster based on mixed scenarios to a certain extent.
[0134] In addition, in the Internet of Things, when a new terminal device is connected to an edge domain node, an adapter corresponding to the new terminal device can be created in the edge domain node; or when a terminal device communicating with an edge domain node changes, the adapter corresponding to the terminal device can be updated in the edge domain node. The adapter here can be understood as the device access driver for the terminal device to access the edge domain node. However, in this case, only the access adaptation of the terminal device is reflected, and the management of the terminal device access is not reflected.
[0135] Therefore, in this application, a software module for managing terminal devices accessing edge domain nodes can be added in the cloud center, and a software module for managing terminal devices accessing edge domain nodes can be added in each edge domain node in the edge domain. Users can add terminal devices in the cloud center and perform operations to bind terminal devices to edge domain nodes. Alternatively, users can also add terminal devices to edge domain nodes and perform operations to bind terminal devices to edge domain nodes to achieve management of terminal devices when accessed by the cloud center or edge domain nodes.
[0136] like Figure 4As shown, in the present application, the software framework of the cloud center may include an edge management module, a device model warehouse, a device access driver warehouse, an algorithm warehouse, and an edge application warehouse.
[0137] The software framework of each edge domain node in the edge domain may include an IoT edge service module, which may include an edge agent and an IoT edge service framework supporting extensions, which may include a device management framework, an application integration framework, a device access framework, and an algorithm integration framework.
[0138] Among them, the above modules and warehouses can be understood as a software module.
[0139] The edge management module can be used to manage each edge domain node in the edge domain. For example, the edge agent in the edge domain node can register with the edge management module. When the registration is successful, the edge management module can send the model in the device model warehouse or the driver in the device access driver warehouse to the registered node.
[0140] The device model warehouse can be used to store basic information of terminal devices, operation interfaces that can be provided to terminal devices, and alarm functions, etc. It can be understood as being used to digitally model terminal devices.
[0141] The device access driver repository can be used to support terminal devices accessing edge domain nodes.
[0142] The algorithm warehouse can be used to store multiple algorithm packages, which can be applied to edge domain nodes.
[0143] The edge application warehouse can be understood as an application warehouse for deployment in edge domain nodes. For example, the application can be a monitoring system and an access control system. The application can be used as a data display window on the edge domain node, and terminal device access information and algorithm integration framework can be displayed on the application window.
[0144] The edge agent in the edge domain can be used to support cloud-edge communication. For example, the edge agent can register with the edge management module in the cloud center. After successful registration, the edge management module can communicate with the edge agent.
[0145] The device management framework corresponds to the device model warehouse in the cloud center and can be used to load the device models in the device model warehouse to generate device control, device configuration, and device addition and deletion forms and interfaces for device management.
[0146] The application integration framework corresponds to the edge application warehouse in the cloud center and can be used to integrate applications in the edge application warehouse and load and run them.
[0147] The device access framework corresponds to the device access driver repository in the cloud center and can be used to load and run the device access driver.
[0148] The algorithm integration framework corresponds to the algorithm warehouse in the cloud center and can be used to integrate the algorithm framework and load the algorithm to run, that is, to perform algorithm analysis tasks. For example, the algorithm analysis task can be understood as the above-mentioned face detection.
[0149] It should be noted that, using the software framework of the cloud center and edge domain nodes provided by the present application, the above-mentioned step 202 can be the sampling data sent by the edge agent in the first edge domain node to the edge management module of the cloud center; step 203 can be implemented by the edge management module in the cloud center according to multiple algorithm packages corresponding to the type of sampling data in the algorithm warehouse; step 204 can be the algorithm analysis process triggered by the edge management module in the cloud center to execute multiple algorithm packages; step 205 can be the edge management module in the cloud center determining the target algorithm package based on the analysis results corresponding to each algorithm package; step 206 can be the target algorithm package and algorithm analysis task sent by the edge management module in the cloud center to the first edge domain node; step 207 can be the IoT edge service module in the first edge domain node triggering the algorithm integration framework to integrate the target algorithm package, and performing the algorithm analysis task on the data type corresponding to the sampling data through the target algorithm package.
[0150] The above steps 208 to 210 can all be executed by the edge management module in the cloud center, and step 211 can be executed by the algorithm integration framework in the edge domain node.
[0151] The software framework of the above cloud center and edge domain nodes can be applied to the following two scenarios where terminal devices access the edge domain.
[0152] Scenario 1: Assume that the edge proxy in the edge domain nodes A, B, and C of the edge domain is registered with the edge management module of the cloud center, then the edge management module determines that the edge domain nodes A, B, and C exist. Assume that the user determines that the terminal device D is close to the edge domain node A, or the user determines that the computing power of the edge domain node A is strong. Therefore, when the user adds the terminal device D to the edge management module of the cloud center and adds the terminal device D to bind the edge domain node A in the edge management module of the cloud center, as shown in the following figure: Figure 5 As shown, the process of the cloud center managing the terminal device D may include:
[0153] 501. The cloud center sends basic information of the terminal device D to the edge domain node A. That is, the edge domain node A receives the basic information of the terminal device D sent by the cloud center.
[0154] The edge management module in the cloud center can send basic information of the terminal device D to the edge agent in the edge domain node A.
[0155] Exemplarily, the basic information here may include the Internet Protocol (IP) port of the terminal device D, the user name, the user password, the device name, and the protocol type used for communication, etc.
[0156] The above-mentioned user adds the terminal device D in the edge management module of the cloud center, which can be understood as adding the basic information of the terminal device D in the edge management module.
[0157] 502. The edge domain node A determines whether the edge domain node A has the capability to access and manage the device type of the terminal device D. If it is determined that the capability exists, step 508 is executed. If it is determined that the capability does not exist, step 503 is executed.
[0158] After receiving the basic information of the terminal device D, the edge agent of the edge domain node A may synchronize the basic information to the IoT edge service module of the edge domain node A. The IoT edge service module may identify whether the edge domain node A has the ability to access and manage the device type of the terminal device D based on the basic information. For example, when the IoT edge service module determines that the device access framework includes the device access driver of the terminal device D, and the device management framework includes the device model of the terminal device D, it is determined that the edge domain node A supports access and management of the terminal device D.
[0159] Exemplarily, the terminal device D is a camera, and the edge domain node A may determine whether there is a capability to access and manage camera-type devices.
[0160] 503. The edge domain node A requests the cloud center to download the device model and device access driver of the terminal device D. That is, the cloud center receives the request sent by the edge domain node A to download the device model and device access driver of the terminal device D.
[0161] When the IoT edge service module in edge domain node A determines that it supports access and management of terminal device D, the IoT edge service module can instruct the edge agent to send a request message to the edge management module of the cloud center, where the request message is used to request downloading of the device model and device access driver of terminal device D.
[0162] The request message may include basic information of the terminal device D.
[0163] 504. The cloud center determines whether there is a device model of the terminal device D locally. If yes, step 505 is executed. If no, a response indicating that there is no matching model is returned to the edge domain node A.
[0164] When the edge management module in the cloud center receives the request message, it can determine whether there is a device model corresponding to the basic information in the device model warehouse of the cloud center according to the basic information.
[0165] When there is no device model corresponding to the basic information in the device model warehouse, the edge service module of the cloud center has a problem support function that can be used for problem feedback. For example, the edge service module can submit the basic information of terminal device D and the device model definition application of terminal device D to other servers that can provide problem feedback, so as to obtain the device model of terminal device D for use when a node requests to obtain the device model later.
[0166] 505. The cloud center determines whether there is a device access driver for the terminal device D locally. If yes, step 506 is executed. If no, a response indicating that there is no matching driver is returned to the edge domain node A.
[0167] The edge management module of the cloud center can determine whether there is a device access driver corresponding to the basic information of the terminal device D in the device access driver warehouse in the cloud center.
[0168] When there is no device access driver corresponding to the basic information in the device access driver warehouse, the edge service module of the cloud center can also obtain the device access driver of the terminal device D through problem feedback, so that a node can request to obtain the device access driver later.
[0169] 506. The cloud center sends the download link of the device model and the download link of the device access driver of the terminal device D to the edge domain node A. That is, node A receives the download link of the device model and the download link of the device access driver of the terminal device D sent by the cloud center.
[0170] In some embodiments, the edge management module of the cloud center can also directly obtain the device model of the terminal device D from the device model warehouse and send it to the edge agent of the edge domain node A, and directly obtain the device access driver of the terminal device D from the device access driver warehouse and send it to the edge agent of the edge domain node A.
[0171] In step 506, the cloud center sends the download link of the device model and the device access driver to the terminal device D. This is because the storage space of the cloud center may be limited, or when the cloud center directly obtains the device model from the device model warehouse and obtains the device access driver from the device access driver warehouse and returns it to the edge domain node A, it is easy to cause interaction blockage or data loss. Therefore, the device model warehouse of the cloud center may only store the link of the device model of the terminal device D, and the device access driver warehouse may only store the link of the device access driver of the terminal device D.
[0172] 507 . The edge domain node A downloads the device model and the device access driver according to the download link of the device model and the download link of the device access driver, and loads the device model and the device access driver.
[0173] When the edge agent of edge domain node A can obtain the device model of terminal device D according to the download link of the device model of terminal device D, the edge agent in edge domain node A can execute the loading of the device model in edge domain node A; and, when the edge agent of edge domain node A can obtain the device access driver of terminal device D according to the download link of the device access driver of terminal device D, the edge agent in edge domain node A can execute the loading of the device access driver in edge domain node A.
[0174] Exemplarily, the edge agent may instruct the device model framework to load the device model to generate a device management form for the terminal device D, which may include add, delete, modify, and query items of the terminal device D, configuration items of the terminal device D, event monitoring items, operation event items, and operation attribute items. The edge agent may also instruct the device access driver framework to complete the installation and deployment of the device access driver of the terminal device D on the edge domain node A side.
[0175] In addition, the user can also complete the device information in the device management form of terminal device D, for example, to add basic information of terminal device D, connection information (such as the IP address, port, user name, password and supported protocol types of terminal device D, etc.), tag information and location information, etc. in the device management form.
[0176] 508 . The edge domain node A obtains the device connection information between the edge domain node A and the terminal device D, and executes the process of the terminal device D accessing the node A.
[0177] The device access driver in the edge domain node A can obtain the device connection information of the terminal device D from the device management form, and communicate with the terminal device D according to the device connection information, so that the terminal device D can access the edge domain node A.
[0178] Scenario 2: Assume that the user determines that the terminal device D is close to the node B, or the user determines that the computing power of the node B is strong. At this time, the user can also add the terminal device D in the IoT edge service module of the node B, which means that the terminal device D is bound to the node B. Figure 6 As shown, the process of node B managing terminal device D may include:
[0179] 601. A user adds basic information of a terminal device D in the edge domain node B.
[0180] For example, the user may add basic information of the terminal device D in the edge proxy of the edge domain node B. The content of the basic information may be referred to in step 501 .
[0181] 602. The edge domain node B determines whether the edge domain node B has the capability to access and manage the device type of the terminal device D. If it is determined that the capability exists, step 608 is executed; if it is determined that the capability does not exist, step 603 is executed.
[0182] The manner in which the edge domain node B executes step 602 may refer to step 502 .
[0183] 603. The edge domain node B requests the cloud center to download the device model and device access driver of the terminal device D. That is, the cloud center receives the request from the edge domain node B to download the device model and device access driver of the terminal device D.
[0184] The way in which the edge domain node B executes step 603 may refer to step 503 .
[0185] 604. The cloud center determines whether there is a device model of the terminal device D locally. If yes, execute step 605. If no, return a response of no matching model to the edge domain node B.
[0186] The way the cloud center executes step 604 can refer to step 504.
[0187] 605. The cloud center determines whether there is a device access driver for the terminal device D locally. If yes, step 606 is executed. If no, a response indicating that there is no matching driver is returned to the edge domain node B.
[0188] The way the cloud center executes step 605 can refer to step 505.
[0189] 606. The cloud center sends the download link of the device model and the download link of the device access driver of the terminal device D to the edge domain node B. That is, the edge domain node B receives the download link of the device model and the download link of the device access driver of the terminal device D sent by the cloud center.
[0190] The way the cloud center executes step 606 can refer to step 506.
[0191] 607 . The edge domain node B downloads the device model and the device access driver according to the download link of the device model and the download link of the device access driver, and loads the device model and the device access driver.
[0192] The way in which the edge domain node B executes step 607 may refer to step 507 .
[0193] 608. Node B obtains device connection information between edge domain node B and terminal device D, and executes a process for terminal device D to access edge domain node B.
[0194] The manner in which the edge domain node B executes step 608 may refer to step 508 .
[0195] Therefore, in scenario 1, the basic information of the terminal device that will be connected to the edge domain node can be added in the cloud center, and the cloud center will then feedback the device model resources and device access driver resources to the edge domain node. After the device model resources and device access driver resources are loaded and run in the edge domain node, the edge domain node can execute the process of the terminal device accessing the node, and the cloud center can manage the terminal device when it is accessed. In scenario 2, the basic information of the terminal device that will be connected to the edge domain node can be added in the node, and the edge domain node will request the cloud center to obtain the device model resources and device access driver resources. After the device model resources and device access driver resources are loaded and run in the edge domain node, the edge domain node can execute the process of the terminal device accessing the edge domain node, and the edge domain node can manage the terminal device when it is accessed.
[0196] Similar to the above scenario of adding and binding terminal devices, users can also configure applications in edge domain nodes in the edge application warehouse of the cloud center and bind applications to edge domain nodes. After receiving the binding message, the edge domain node can download the application to the edge application warehouse of the cloud center through the node agent. The node agent performs the installation and deployment of the application on the edge domain node side, and the deployed edge application runs in the edge domain node.
[0197] For scenarios 1 and 2, if the IoT edge service module in edge domain node A or edge domain node B is running, once edge domain node A or edge domain node B needs to be completely unbound from terminal device D, the node agent in edge domain node A or edge domain node B can uninstall the device access driver corresponding to terminal device D, thereby reducing the resource usage of edge domain node A or edge domain node B.
[0198] Exemplarily, the node agent can scan the device access driver framework of the current node (edge domain node A or edge domain node B) to determine whether the device access driver framework of terminal device D is still communicating with terminal device D. If it is determined that the communication connection has been disconnected, the node agent can uninstall the device access driver framework of terminal device D.
[0199] This cloud-edge collaborative device access management mechanism can make the use of edge domain node resources in the edge domain more reasonable, that is, resources are used when needed and released in time when not needed.
[0200] In some embodiments, when in scenario 1, that is, the user adds terminal device D in the edge management module of the cloud center, if any edge domain node in the edge domain supports the access of terminal device D, the cloud center can also recommend to the user, or even help the user automatically determine the node object to which the terminal device D is bound.
[0201] For example, the edge management module in the cloud center can make recommendations or decisions on selecting edge domain nodes based on the following three indicators:
[0202] 1) The number of valid terminal devices accessing edge domain nodes in real time.
[0203] In some embodiments, after each terminal device accesses the edge domain node, the edge domain node may notify the edge management module of the cloud center. At the same time, the edge domain node may also notify the edge management module of the cloud center of the number of online terminal devices N. onlineDevice , and the number of offline terminal devices N otherDevice In this way, the edge management module of the cloud center can be based on the number of online terminal devices N of each edge domain node. onlineDevice And the number of offline terminal devices N otherDevice Determine the number of valid terminal devices currently connected to the edge domain node N device .
[0204] For example, for each edge domain node, the number of valid terminal devices connected is N device It can be expressed as:
[0205] N device =N onlineDevice .α online +N otherDevice .α other
[0206] Where 1 = α online +α other , α online Indicates the number of online terminal devices N onlineDevice The weight coefficient, α other Indicates the number of offline terminal devices N otherDevice The weight coefficient of online is a larger weight value, α other is a smaller weight value.
[0207] It can be understood that the number of effective terminal devices connected is N device The smaller the number, the greater the probability that the corresponding edge domain node is called a recommended edge domain node.
[0208] 2) Hardware resource availability of edge domain nodes.
[0209] In some embodiments, the edge agent in the edge domain node can notify the edge management module of the cloud center of the relevant information of the hardware resource utilization rate of the edge domain node. The relevant information of the hardware resource utilization rate may include: the bandwidth, CPU, memory, hard disk and other indicators currently occupied by the edge domain node. The edge management module can calculate the hardware resource utilization rate of the node based on the relevant information of the hardware resource utilization rate.
[0210] For example, for each edge domain node, its hardware resource availability V server It can be expressed as:
[0211] V server =f(V cpu .α cpu +V ram .α ram +V disk .α disk +V net .α net )
[0212] Where 1 = α cpu +α ram +α disk +α net , V cpu Indicates the remaining CPU resources of the node, α upu express
[0213] V cpu The weight coefficient of V ram Indicates the remaining resources used by the node memory, α ram Indicates V ram The weight coefficient, V disk Indicates the remaining disk capacity of the node, α disk Indicates V disk The weight coefficient of V net represents the remaining bandwidth resources of the node, α net Indicates V net The weight coefficient of .
[0214] 3) The network distance between the edge domain node and the terminal device to be connected.
[0215] The network distance measurement here can reflect the relative distance between the edge domain node and the terminal device through the round-trip delay of the Transmission Control Protocol (TCP) three-way handshake when the edge domain node and the terminal device are connected. It can be understood that a small relative distance indicates that the network communication quality between the edge domain node and the terminal device is good and more stable.
[0216] For example, the network distance between the edge domain node and the terminal device to be connected It can be expressed as:
[0217]
[0218] Among them, RTT i represents the i-th round trip delay, n=3.
[0219] It can be understood that the smaller the value of the network distance is, the better and more stable the communication quality between the edge domain node and the terminal device is.
[0220] In some embodiments, the edge management module in the cloud center can comprehensively calculate the ranking of the edge domain nodes under the cloud center through the values of the above three indicators, and thus determine the recommended edge domain nodes according to the ranking.
[0221] There are many algorithms for comprehensive calculation, such as normalization, weighted average, least squares method and other common algorithms, and more complex algorithms such as Kalman filter, extended Kalman filter algorithm and the like can also be used.
[0222] In this way, when the cloud center uses these three indicators to determine the recommended edge domain node when the terminal device accesses the node, it can more reasonably utilize the resources of each edge domain node, so that the communication efficiency of the terminal device after accessing the edge domain node is improved.
[0223] It is understandable that in order to implement the above functions, the above cloud center or edge domain node includes hardware and / or software modules corresponding to the execution of each function. In combination with the algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application in combination with the embodiments, but such implementation should not be considered to be beyond the scope of this application.
[0224] This embodiment can divide the cloud center or edge domain node into functional modules according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is schematic and is only a logical function division. There may be other division methods in actual implementation.
[0225] In the case of dividing each functional module into corresponding functional modules, Figure 7 A possible composition diagram of the server in the cloud center involved in the above embodiment is shown, such as Figure 7As shown, the server 70 may include: a determining unit 701 , an analyzing unit 702 , a sending unit 703 and a receiving unit 704 .
[0226] Among them, the determination unit 701 can be used to support the server 70 to execute the above-mentioned steps 203, step 205, step 208, step 209, step 504, step 505, step 604, step 605, etc., and / or other processes for the technology described in this article.
[0227] The analysis unit 702 may be used to support the server 70 in executing the above-mentioned step 204 and / or other processes of the technology described herein.
[0228] The sending unit 703 can be used to support the server 70 to execute the above-mentioned steps 206, 210, 501, 506, 606, etc., and / or other processes for the technology described in this document.
[0229] The receiving unit 704 may be used to support the server 70 in executing the above-mentioned steps 202 , 503 , etc., and / or other processes for the technology described herein.
[0230] It should be noted that all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.
[0231] The server 70 provided in this embodiment is used to execute the above-mentioned scenario-based algorithm configuration method, and thus can achieve the same effect as the above-mentioned implementation method.
[0232] In the case of using an integrated unit, the server 70 may include a processing module, a storage module and a communication module. Among them, the processing module can be used to control and manage the actions of the server 70, for example, it can be used to support the server 70 to execute the steps performed by the above-mentioned determination unit 701 and analysis unit 702. The storage module can be used to support the server 70 to store program codes and data, etc. The communication module, which may include the above-mentioned sending unit 703 and receiving unit 704, can be used to support the communication between the server 70 and other devices, such as communication with edge domain nodes.
[0233] In the present application, the communication module and the processing module included in the server 70 may be distributed in one device or in different devices. If they are distributed in different devices, the server 70 may be understood as a server cluster.
[0234] Among them, the processing module can be a processor or a controller. It can implement or execute various exemplary logic boxes, modules and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory. The communication module can specifically be a device that interacts with other electronic devices, such as a radio frequency circuit, a Bluetooth chip, a Wi-Fi chip, etc.
[0235] In one embodiment, when the processing module is a processor, the storage module is a memory, and the communication module is a transceiver, the server 70 involved in this embodiment can be a server having Figure 8 A server 80 of the structure shown.
[0236] In the case of dividing each functional module into corresponding functional modules, Fig. 9 A possible schematic diagram of the composition of the edge domain nodes involved in the above embodiments is shown. Fig. 9 As shown, the edge domain node 90 may include: a receiving unit 901 , a sending unit 902 , an integrating unit 903 and a determining unit 904 .
[0237] The receiving unit 901 may be used to support the edge domain node 90 to execute the above steps 201, 206, 210, 607, 501, 506, etc., and / or other processes for the technology described herein.
[0238] The sending unit 902 may be used to support the edge domain node 90 to execute the above steps 202, 503, 603, etc., and / or other processes for the technology described herein.
[0239] The integration unit 903 may be used to support the edge domain node 90 to execute the above steps 207 , 211 , 507 , etc., and / or other processes for the technology described herein.
[0240] The determining unit 904 may be used to support the edge domain node 90 to execute the above steps 502 , 508 , 602 , 608 , etc., and / or other processes for the technology described herein.
[0241] It should be noted that all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.
[0242] The edge domain node 90 provided in this embodiment is used to execute the above scenario-based algorithm configuration method, and thus can achieve the same effect as the above implementation method.
[0243] In the case of using an integrated unit, the edge domain node 90 may include a processing module, a storage module and a communication module. Among them, the processing module can be used to control and manage the actions of the edge domain node 90, for example, it can be used to support the edge domain node 90 to execute the steps performed by the above-mentioned integrated unit 903 and the determination unit 904. The storage module can be used to support the server node 90 to store program codes and data, etc. The communication module, which may include the above-mentioned sending unit 902 and receiving unit 901, can be used to support the communication between the edge domain node 90 and other devices, such as communicating with the cloud center.
[0244] Among them, the processing module can be a processor or a controller. It can implement or execute various exemplary logic boxes, modules and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory. The communication module can specifically be a device that interacts with other electronic devices, such as a radio frequency circuit, a Bluetooth chip, a Wi-Fi chip, etc.
[0245] In one embodiment, when the processing module is a processor, the storage module is a memory, and the communication module is a transceiver, the edge domain node 90 involved in this embodiment may be a Figure 8 A server with the structure shown.
[0246] An embodiment of the present application also provides a computer storage medium, which stores computer instructions. When the computer instructions are executed on an electronic device (cloud center and edge domain nodes), the electronic device executes the above-mentioned related method steps to implement the scenario-based algorithm configuration method in the above-mentioned embodiment.
[0247] An embodiment of the present application also provides a computer program product. When the computer program product runs on a computer, it enables the computer to execute the above-mentioned related steps to implement the scenario-based algorithm configuration method executed by the electronic device (cloud center and edge domain node) in the above-mentioned embodiment.
[0248] Another embodiment of the present application provides a system, which may include the above-mentioned cloud center, the above-mentioned at least one edge domain node and at least one terminal device, and can be used to implement the above-mentioned scenario-based algorithm configuration method.
[0249] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.
[0250] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0251] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0252] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0253] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
[0254] The above contents are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A scenario-based algorithm configuration method, characterized in that: include: The cloud center receives the sampling data sent by the first edge domain node; The cloud center determines a plurality of algorithm packages corresponding to a data type of the sampled data, where the data type of the sampled data is used to reflect a scenario when a terminal device acquires the sampled data; The cloud center executes each of the multiple algorithm packages on the sampled data to obtain an algorithm evaluation value corresponding to each of the algorithm packages, wherein the algorithm evaluation value is determined based on multiple evaluation indicators in the algorithm evaluation system corresponding to the data type of the sampled data; The cloud center determines a target algorithm package from the multiple algorithm packages according to the algorithm evaluation value corresponding to each algorithm package, and the target algorithm package is used to analyze the data collected by the terminal device.
2. The method according to claim 1, characterized in that The cloud center executes each of the multiple algorithm packages on the sampled data, and obtains an algorithm evaluation value corresponding to each of the algorithm packages, including: The cloud center determines a plurality of evaluation indicators in an algorithm evaluation system corresponding to the data type of the sampled data; The cloud center refers to the multiple evaluation indicators, performs analysis on the sampled data through each algorithm package in the multiple algorithm packages, and obtains the values of the multiple evaluation indicators corresponding to each algorithm package; The cloud center determines the algorithm evaluation value corresponding to each algorithm package according to the values of the multiple evaluation indicators corresponding to each algorithm package; The cloud center determines the target algorithm package from the multiple algorithm packages according to the algorithm evaluation value corresponding to each algorithm package, including: The cloud center determines, according to the algorithm evaluation values corresponding to each algorithm package, the algorithm package with the highest algorithm evaluation value among the multiple algorithm packages as the target algorithm package.
3. The method according to claim 1 or 2, characterized in that: The method further comprises: The cloud center sends the target algorithm package and the algorithm analysis task to the first edge domain node, where the algorithm analysis task is used to instruct the first edge domain node to analyze the data received from the terminal device through the target algorithm package.
4. The method according to claim 1 or 2, characterized in that: The method further comprises: The cloud center determines the remaining computing power respectively corresponding to a plurality of edge domain nodes managed by the cloud center, wherein the plurality of edge domain nodes include the first edge domain node; The cloud center determines a second edge domain node according to the residual computing power respectively corresponding to the plurality of edge domain nodes, the second edge domain node being the edge domain node with the largest residual computing power among the plurality of edge domain nodes; The cloud center sends the target algorithm package and the algorithm analysis task to the second edge domain node, and the algorithm analysis task is used to instruct the second edge domain node to analyze the data collected by the terminal device and received from the first edge domain node through the target algorithm package.
5. The method according to claim 4, characterized in that The cloud center determines the remaining computing power corresponding to the plurality of edge domain nodes managed by the cloud center, including: The cloud center determines the remaining computing power corresponding to each edge domain node according to the hardware resource performance, data access volume and the number of connected terminal devices corresponding to each edge domain node among the multiple edge domain nodes.
6. A scenario-based algorithm configuration method, characterized in that: include: The edge domain nodes send sampled data to the cloud center; The edge domain node receives a first target algorithm package sent by the cloud center, where the first target algorithm package is determined by the cloud center executing each algorithm package in multiple algorithm packages on the sampled data, and an algorithm evaluation value corresponding to each algorithm package is obtained, and the multiple algorithm packages are determined by the cloud center according to the data type of the sampled data; the type of the sampled data is used to reflect the scenario when the terminal device obtains the sampled data; The edge domain node loads the first target algorithm package to analyze data received from the terminal device through the first target algorithm package.
7. The method according to claim 6, characterized in that The edge domain node loading the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package includes: The edge domain node determines whether the first target algorithm package has been loaded locally on the edge domain node; If it is determined that the edge domain node has not locally loaded the first target algorithm package, the edge domain node loads the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package.
8. The method according to claim 6, characterized in that The method further comprises: The edge domain node receives a second target algorithm package and an algorithm analysis task sent by the cloud center, where the second target algorithm package is determined by the cloud center based on sampling data sent by other edge domain nodes, and the algorithm analysis task includes terminal device information to be analyzed; The edge domain node obtains data collected by the terminal device corresponding to the terminal device information to be analyzed from the other edge domain nodes; The edge domain node loads the second target algorithm package to analyze data received from other edge nodes through the second target algorithm package.
9. A cloud center, characterized in that: include: A communication module, configured to receive sampling data sent by the first edge domain node; a processing module, used to determine a plurality of algorithm packages corresponding to a data type of the sampled data, where the data type of the sampled data is used to reflect a scenario when a terminal device acquires the sampled data; The processing module is further used to execute each algorithm package of the multiple algorithm packages on the sampled data to obtain an algorithm evaluation value corresponding to each algorithm package, wherein the algorithm evaluation value is determined based on multiple evaluation indicators in the algorithm evaluation system corresponding to the data type of the sampled data; The processing module is further used to determine a target algorithm package from the multiple algorithm packages according to the algorithm evaluation value corresponding to each algorithm package, and the target algorithm package is used to analyze the data collected by the terminal device.
10. An edge domain node, characterized in that: The edge domain node includes: Communication module, used to send sampling data to the cloud center; The communication module is further used to receive a first target algorithm package sent by the cloud center, where the first target algorithm package is determined by the cloud center executing each algorithm package in the multiple algorithm packages on the sampled data, and the algorithm evaluation value corresponding to each algorithm package is obtained; the multiple algorithm packages are determined by the cloud center according to the data type of the sampled data; the type of the sampled data is used to reflect the scenario when the terminal device obtains the sampled data; A processing module is used to load the first target algorithm package to analyze the data received from the terminal device through the first target algorithm package.
11. An electronic device, characterized in that: include: A memory, a transceiver and a processor; the memory, the transceiver and the processor are coupled; the memory is used to store computer program code, the computer program code includes computer instructions; the transceiver is used to receive data and send data; when the processor executes the computer instructions, the electronic device executes the method as described in any one of claims 1-8.
12. A computer-readable storage medium, characterized in that: The method comprises computer instructions, which, when executed on an electronic device, cause the electronic device to execute the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Distributed flow collection system and edge calculation method
CN111131379A
Edge calculation method and system for video monitoring
CN111405241A
Data analysis method, device and system and storage medium
CN111464650A
Fine granularity real-time supervision system based on edge computing
US20210096911A1
Algorithm configuration method, device, system, and movable platform
WO2021003692A1
