Image recognition type edge computing server application platform based on cluster mode
By adopting a cluster-mode distributed architecture and custom communication protocol in the image recognition intelligent system, dynamically allocating computing power, solving the problems of high hardware cost and poor scalability, and realizing a low-cost and high-scalability image recognition system. Suitable for small industry sectors.
Patent Information
- Application Number
- CN202411585486.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-11-08
AI Technical Summary
The existing image recognition intelligent systems have high hardware costs and poor subsequent scalability, which makes it difficult for small industries to adopt AI intelligent recognition, and the system scalability is limited, requiring early redundant hardware configuration.
Adopt a cluster-based image recognition edge computing server application platform, adopts a distributed cluster architecture in master-slave node mode, and realizes communication between the center service node and the edge service node through custom communication protocols. The edge service node executes the center service node The task process is issued to control the operation of the image recognition algorithm model and dynamically allocate computing power to meet the expansion needs.
It reduces hardware costs, achieves the same recognition effect as existing systems, and breaks through the upper limit of subsequent expansion, without the need for early redundant computing resources, and supports the increase and decrease of the needs of edge computing servers at any time.
Smart Images

Figure CN119090701A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of edge computing server application technology, and specifically to an image recognition-based edge computing server application platform in a cluster mode. Background Art
[0002] In recent years, as the popularity of artificial intelligence technologies such as ChatGPT developed by OpenAI has increased significantly, the demand for related hardware required for the practical application of AI technology has also been increasing. Among them, the hardware performance demand for edge computing servers that are responsible for data processing and real-time computing capabilities has risen sharply.
[0003] Current artificial intelligence systems, especially image recognition-type intelligent systems, mainly improve the configuration of related hardware on edge computing servers to achieve the capability support for the system computing power requirements. Among them, graphics cards with high video memory and high computing power are indispensable for servers.
[0004] Although this method can achieve the required image recognition function, the high hardware cost and the subsequent poor scalability are also particularly prominent: First, the price of graphics cards with high video memory and high computing power is constantly rising, which leads to an increase in the cost of the entire system, which means that the system cannot be extended to some smaller industries with vertical demands, such as: smart security systems for individual gas stations, smart inspection systems for small enterprises, etc.; these industries often only focus on the intelligent image recognition needs of one or several vertical segments, such as smoking and making phone calls in gas stations, and not wearing safety helmets as required in corporate factories. However, due to the high hardware costs, these industries often use manual management rather than AI intelligent recognition to solve related scenario needs.
[0005] Secondly, the issue of subsequent scalability of the system cannot be ignored. The scalability of a single edge computing server is destined to be limited. If we take into account the continuous increase in subsequent business needs and the continuous expansion of the types of image recognition processing, we must consider the scalability of the edge computing server. This may lead to more redundant configuration of idle hardware configurations in the early stage of system design, and this redundant design also increases the cost of the entire system. Summary of the invention
[0006] In view of the above shortcomings and deficiencies, the present invention provides an image recognition edge computing server application platform based on a cluster mode to overcome the problems of high hardware cost and poor subsequent scalability in existing image intelligent recognition systems.
[0007] In order to achieve the above-mentioned purpose, the main technical solutions adopted by the present invention include: an image recognition edge computing server application platform based on cluster mode, which is constructed using a distributed cluster architecture of a master-slave node mode, including a central service node and multiple edge service nodes. The central service node and the edge service node communicate with each other through a custom communication protocol. The edge service node executes the task execution platform operation process issued by the central service node, controls the operation of the image recognition algorithm model, and achieves the purpose of image recognition; The edge service node executes the task execution platform operation process issued by the central service node, and controls the operation of the image recognition algorithm model, which specifically includes the following steps: S1. The edge service node starts and establishes a connection with the central service node according to the service configuration startup parameters, and sends a registration request to the central service node; S2. The central service node issues task information according to the registration node and sends it to the edge service node. The edge service node starts the image recognition algorithm model according to the task allocation, binds it with the task ID, and then sends it back to the central service node; S3. After the central service node receives the startup response from the edge service node, it searches for the configuration information of the node task and sends it to the edge service node through the task ID. After receiving the configuration information sending message, the edge service node loads the configuration information parameters into the image recognition algorithm model and then sends the configuration response status back to the central service node. S4. After the central service node receives the configuration response from the edge service node, it searches for the video stream information of the node task and sends it to the edge service node through the task ID. After the edge service node receives the video stream sending message, it controls the image recognition algorithm model to pull the video stream. The image recognition algorithm model recognizes the image and returns the result data. After the edge service node obtains the result data, it uploads it to the central service node. S5. After receiving the data, the central service node performs data shaping, ROI filtering, and business processing operations on the identified data to generate effectively identified data, and archives the processed data for storage or uploads it to the relevant business system connected to this platform.
[0008] The central service node uniformly manages the service configuration startup parameters of all image recognition algorithm models and the video streams of video acquisition devices, including the push streaming of image recognition algorithm models and the monitoring of the real-time status of video acquisition devices.
[0009] The central service node uniformly manages the status of all edge service nodes. The edge service nodes maintain their running status by regularly sending heartbeat data. If the central service node finds that a certain edge service node has a status timeout, it will reallocate tasks based on the status of the current online edge service node to keep all tasks in running status.
[0010] The message encapsulation format of the self-defined communication protocol used in the network communication between the central service node and the edge service node includes a start identification bit, a message header, a message body, a verification code and an end identification bit.
[0011] The start identifier and the end identifier adopt a fixed format to indicate the start and end of the message; the message header includes the message type, message body attributes, protocol version, message serial number and message encapsulation attributes; the message body includes the response serial number, the corresponding request message type and the message content; the verification code is used to verify the data integrity and correctness.
[0012] The customized communication protocol used in the network communication between the central service node and the edge service node includes central service node general response, edge service node general response, node registration, task delivery, task delivery response, parameter delivery, video stream address delivery, heartbeat upload, status upload, result upload and image upload.
[0013] The present invention has the following beneficial effects and advantages: 1. The present invention is based on the application method of image recognition edge computing servers under cluster mode. In the image recognition intelligent system, the communication protocol between the edge computing server and the devices under the cluster mode is designed based on the cluster mode to solve the hardware cost problem of the current image recognition intelligent system. The computing resources of multiple edge computing servers with lower configurations are integrated to achieve the same recognition effect as the existing system.
[0014] 2. The present invention solves the problem of subsequent expansion capabilities of current image recognition intelligent systems. The method can meet the needs of increasing or decreasing edge computing servers at any time. The central service node will reallocate computing power based on the number of edge service nodes actually registered online. Theoretically, it breaks through the upper limit of subsequent expansion and does not require redundant computing power resources in the early stage. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a flow chart of the image recognition edge computing server application platform based on the cluster mode of the present invention. DETAILED DESCRIPTION
[0016] The present invention will be further described below in conjunction with the accompanying drawings. Figure 1As shown, the present invention is an image recognition edge computing server application platform based on a cluster mode, which is constructed using a distributed cluster architecture of a master-slave node mode, including a central service node and multiple edge service nodes. The central service node runs alone on a service host, and multiple edge service nodes can run together on a service host or run on different service hosts according to the usage scenario; all edge service nodes need to establish a network connection with the central service node, and the central service node needs to be responsible for maintaining the status of these network connections. The central service node and the edge service node communicate with each other through a custom communication protocol. The edge service node executes the task execution platform operation process issued by the central service node to control the operation of the image recognition algorithm model. All tasks on the edge service node are uniformly allocated and issued by the central service node. All edge service nodes do not need to be associated with specific business scenarios, but only need to execute related tasks according to the allocation of the central service node, thereby realizing the "plug and play" of the edge service node and achieving the purpose of image recognition.
[0017] The central service node uniformly manages the service configuration startup parameters of all image recognition algorithm models, and sends them to the corresponding edge service nodes when the model is started or the parameters are changed.
[0018] The central service node uniformly manages the video streams of video acquisition devices of all image recognition algorithm models, including the push of image recognition algorithm models on each edge service node, the monitoring of the real-time status of video acquisition devices, etc., and can control the video acquisition devices in accordance with the national standard GB / T 28181.
[0019] The central service node manages the status of the edge service nodes in a unified manner. The edge service nodes maintain their operating status by regularly sending heartbeat data, and can also report their current operating status by regularly uploading real-time monitoring data. Once the central service node finds that an edge service node has timed out, it will reallocate tasks based on the current online edge service node status to ensure that all tasks are in operation.
[0020] Furthermore, the message encapsulation format of the custom communication protocol used in the network communication between the central service node and the edge service node includes a start flag, a message header, a message body, a verification code and an end flag, as shown below: The start flag and the end flag are in a fixed format to indicate the start and end of the message. If the flag content appears in the message header, message body and verification code, it can be circumvented by escaping in a fixed format.
[0021] The message header usually includes a request message and a response message, including a message type, a message body attribute, a protocol version, a message serial number, and a message encapsulation attribute; the specific format is as follows: The message body includes the response serial number, the corresponding request message type and the message content; the specific format is as follows: The verification code is used to verify data integrity and correctness.
[0022] The communication protocol also includes: Central service node general response: Type ID: 0x0001 The central service node responds to the request from the edge service node; General response from edge service node: Type ID: 0x8001 The edge service node responds to the request from the central service node; Node registration: Type ID: 0x8002 The edge service node uploads registration information; Task distribution: Type ID: 0x0003 The central service node sends task start information to the edge service node; Task delivery response: Type ID: 0x8003 The edge service node responds to the central service node task start response; Parameter delivery: Type ID: 0x0004 The central service node sends the parameters of a single task; Video stream address delivery Type ID: 0x0005 The central service node sends the video stream address of a single task; Heartbeat upload: Type ID: 0x8006 The edge service node sends a heartbeat to the central service node; Status upload: Type ID: 0x8007 The edge service node sends node status information to the central service node; Upload result: Type ID: 0x8008 The edge service node sends the image recognition results to the central service node.
[0023] Image upload: Type ID: 0x8009 The edge service node sends the image results to the central service node; The edge service node executes the task execution platform operation process issued by the central service node, and controls the operation of the image recognition algorithm model, which specifically includes the following steps: S1. The edge service node is started to ensure that the edge service node process is running at any time, and establishes a connection with the central service node according to the service configuration startup parameters, and sends a registration request to the central service node. The registration information includes the number of computing cards of the current node device.
[0024] S2. The central service node allocates and binds the relationship between nodes, computing cards and tasks according to the current online edge service node status and the recognition task status of the current business configuration, and then issues the tasks to be executed by the node through the connection of different nodes; after the edge service node receives the task issued by the central service node, it first checks whether the image recognition algorithm model task is currently running. If so, it stops all algorithm model task processes, then generates the corresponding task id according to the new task allocation, starts the algorithm model according to the task allocation, and binds it to the task id. Return to the central service node.
[0025] S3. After the central service node receives the startup response from the edge service node, it searches for the configuration information of the node task and sends it to the edge service node through the task id; after the edge service node receives the configuration information sending message, it loads the configuration information parameters into the image recognition algorithm model and then transmits the configuration response status back to the central service node.
[0026] S4. After the central service node receives the configuration response from the edge service node, it searches for the video stream information of the node task and sends it to the edge service node through the task ID. After the edge service node receives the video stream sending message, it controls the image recognition algorithm model to pull the video stream. The image recognition algorithm model recognizes the image and returns the result data. After the edge service node obtains the result data, it uploads it to the central service node. S5. After receiving the data, the central service node performs data shaping, ROI filtering, and business processing operations on the identification data to generate valid identification data that can be used by the system, and archives the processed data or uploads it to the relevant business system.
[0027] The above steps only involve a single-graphics card edge computing server or development board. This method is also applicable to scenarios with multiple graphics card devices. This scenario involves secondary allocation of computing power by edge service nodes. That is, after receiving the task sent by the central service node, the edge service node allocates the current task to one or more graphics cards based on the computing power of multiple graphics cards configured on the current server to meet actual business needs.
[0028] The present invention is based on the application method of edge computing servers of image recognition in cluster mode. In the image recognition intelligent system, the communication protocol between the edge computing server and the devices in the cluster mode is designed based on the cluster mode, so as to solve the hardware cost problem of the current image recognition intelligent system, and utilize the computing resources of multiple edge computing servers with lower configurations to achieve the same recognition effect as the existing system. To solve the problem of subsequent expansion capability of the current image recognition intelligent system, this method can meet the demand of increasing or decreasing edge computing servers at any time. The central service node will redistribute the computing power according to the number of edge service nodes actually registered online, theoretically breaking through the upper limit of subsequent expansion, and no redundant computing power resources are required in the early stage.
Claims
1. An edge computing server application platform for image recognition based on cluster mode. It is characterized in that It is constructed using a distributed cluster architecture in a master-slave node mode, including a central service node and multiple edge service nodes. The central service node and the edge service node communicate with each other through a custom communication protocol. The edge service node executes the task execution platform operation process issued by the central service node, controls the operation of the image recognition algorithm model, and achieves the purpose of image recognition. The specific steps include: S1. The edge service node starts and establishes a connection with the central service node according to the service configuration startup parameters, and sends a registration request to the central service node; S2. The central service node issues task information according to the registration node and sends it to the edge service node. The edge service node starts the image recognition algorithm model according to the task allocation, binds it with the task ID, and then sends it back to the central service node; S3. After the central service node receives the startup response from the edge service node, it searches for the configuration information of the node task and sends it to the edge service node through the task ID. After receiving the configuration information sending message, the edge service node loads the configuration information parameters into the image recognition algorithm model and then sends the configuration response status back to the central service node. S4. After the central service node receives the configuration response from the edge service node, it searches for the video stream information of the node task and sends it to the edge service node through the task ID. After the edge service node receives the video stream sending message, it controls the image recognition algorithm model to pull the video stream. The image recognition algorithm model recognizes the image and returns the result data. After the edge service node obtains the result data, it uploads it to the central service node. S5. After receiving the data, the central service node performs data shaping, ROI filtering, and business processing operations on the identified data to generate effectively identified data, and archives the processed data for storage or uploads it to the relevant business system connected to this platform.
2. According to claim 1, an image recognition edge computing server application platform based on cluster mode, Features: The central service node uniformly manages the service configuration startup parameters of all image recognition algorithm models and the video streams of video acquisition devices, including the push streaming of image recognition algorithm models and the monitoring of the real-time status of video acquisition devices.
3. According to claim 1, an image recognition edge computing server application platform based on cluster mode, Features: The central service node uniformly manages the status of all edge service nodes. The edge service nodes maintain their running status by regularly sending heartbeat data. If the central service node finds that a certain edge service node has a status timeout, it will reallocate tasks based on the status of the current online edge service node to keep all tasks in running status.
4. According to claim 1, an image recognition edge computing server application platform based on cluster mode, Features: The message encapsulation format of the self-defined communication protocol used in the network communication between the central service node and the edge service node includes a start identification bit, a message header, a message body, a verification code and an end identification bit.
5. According to claim 4, an image recognition edge computing server application platform based on cluster mode, Features: The start identifier and the end identifier adopt a fixed format to indicate the start and end of the message; the message header includes the message type, message body attributes, protocol version, message serial number and message encapsulation attributes; the message body includes the response serial number, the corresponding request message type and the message content; the verification code is used to verify the data integrity and correctness.
6. According to claim 1, an image recognition edge computing server application platform based on cluster mode, Features: The customized communication protocol used in the network communication between the central service node and the edge service node includes central service node general response, edge service node general response, node registration, task delivery, task delivery response, parameter delivery, video stream address delivery, heartbeat upload, status upload, result upload and image upload.
Citation Information
Patent Citations
Collaborative computing platform system and method in edge environment
CN109889575A
Communication method based on digital retina system
CN112929351A
Task processing method, device and system, computer equipment and storage medium
CN115729683A
Calculation cloud network integrated analysis system for video streams
CN115967719A
Video stream algorithm scheduling system and method and electronic equipment
CN118827683A
