System architecture and deployment method based on global digital space project

Abstract

The invention discloses a system architecture based on a global digital space project and a deployment method. The system architecture comprises a terminal user module, an edge device module, a regional server module and a regional digital space platform module. The system realizes multi-terminal access through the terminal user module, and a camera, a switch and an NVR are deployed on the edge side to complete video acquisition and local storage. And the regional server establishes connection with the NVR, receives a video stream and performs primary processing, and the regional digital space platform splices a multi-view video to generate a panoramic digital twinning scene. And the group-level platform realizes data integration with the Internet of Things platform through the AI algorithm center and provides a standardized interface service. According to the method, a cloud edge cooperative computing mode is adopted, the network bandwidth requirement is reduced, the data processing real-time performance and the system flexibility are improved, cross-region deployment is supported, and the method is suitable for digital space management scenes such as intelligent agriculture and intelligent manufacturing.

Description

Technical Field

[0001] This invention belongs to the technical field of computer vision algorithm deployment, specifically relating to a system architecture and deployment method based on a global digital space project. Background Technology

[0002] With the rapid development of information technology, digital transformation to improve efficiency and enhance competitiveness has become a trend across industries. Particularly in agriculture, industry, and urban management, multimodal data fusion and intelligent analysis combining video technology, the Internet of Things (IoT), and artificial intelligence (AI) are becoming core applications in these fields. However, in practice, a series of challenges remain.

[0003] In the traditional "status quo" model, services are typically operated independently by multiple departments, leading to ineffective resource sharing and difficulties in unified coordination and management. Relying primarily on on-site operations and local support, this traditional model is not only inefficient but also geographically limited. Furthermore, the fixed service model means each service team is only responsible for specific clients and tasks, lacking the necessary flexibility to respond to changing demands. Existing system architectures often lack modular design and standardized interfaces, limiting their adaptability and scalability across different business scenarios, thus impacting the system's versatility and expansion potential. Especially in specific application scenarios such as livestock and industrial monitoring, the lack of real-time analysis capabilities and dynamic scheduling algorithms limits the level of intelligent management, making it difficult to fully meet actual needs.

[0004] To address these issues, this invention introduces the concept of digital twins. By employing real-time video fusion and stitching technology, it creates panoramic videos of pig farm pens, enabling comprehensive monitoring of the pen environment. The generated panoramic videos are then combined with AI algorithms and IoT data to construct a fully perceptive and remotely controllable digital space platform for the entire pig farm. Furthermore, this platform is integrated with various business systems to develop multiple AI applications, including remote pen inspection, remote diagnostics, and asset management. Ultimately, this forms an efficient centralized operation system for pig farmers, achieving efficient and standardized management of pig farming assets and production processes. Summary of the Invention

[0005] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide a system architecture and deployment method based on a full-domain digital space project. This invention improves the real-time performance of video data processing and the flexibility of the system through a cloud-edge collaborative computing model, supports large-scale deployment across regions, and is applicable to various digital space management scenarios such as smart agriculture and intelligent manufacturing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a system architecture based on a full-domain digital space project, including an end-user module, an edge device module, a regional server module, a regional digital space platform module, and a group-level platform module;

[0008] The terminal user module is used to support multi-terminal access and enable real-time video viewing, device management, and AI analysis result display;

[0009] The edge device module is used for video data acquisition and preprocessing, and uploads the preprocessed data to the regional digital space platform module. The edge device module includes cameras, switches and network video recorders (NVRs) deployed at the acquisition end.

[0010] The regional server module adopts a distributed deployment approach to configure a regional server for each target region in order to realize point-to-point video interaction, local AI analysis and dynamic resource scheduling.

[0011] The regional digital space platform module integrates video stitching, device management, user management, and algorithm scheduling functions to generate panoramic monitoring images of the target area.

[0012] The group-level platform module includes an AI algorithm center, an IoT data platform, an algorithm classification unit, and a priority and business classification management unit. The AI ​​algorithm center is used for intelligent analysis of multiple types of data to complete unified computing tasks. The algorithm classification unit is used to dynamically classify the intelligent algorithms within the AI ​​algorithm center, based on the average computing resource consumption of each intelligent algorithm during operation, thereby determining the optimal deployment location for each intelligent algorithm. The priority and business classification management unit is used to assign running priority scores to each intelligent algorithm and attach corresponding business tag groups to the intelligent algorithms. The IoT data platform collects IoT sensor data from regional servers in real time.

[0013] As a preferred technical solution, in the edge device module, cameras are deployed above the target area to capture dynamic data and environmental status data within the target area. The captured data is sent to the NVR for local storage via a switch. The cameras are multiple fisheye cameras deployed at the top of the target area to capture high-resolution video streams from a top-down angle. The switch is a gigabit industrial-grade switch, which ensures the bandwidth for video stream transmission through VLAN segmentation. The NVR encodes and compresses the original video, stores historical data locally, and pushes the real-time stream to the regional server via the RTSP protocol.

[0014] As a preferred technical solution, the regional server module deploys a local AI algorithm to perform frame extraction, resolution adaptation, and Gaussian filtering noise reduction on the video stream transmitted from the edge device module. Based on the deployed local AI algorithm, it detects abnormal behavior and identifies features in the target area in real time, and adjusts computing resources and data flow according to the real-time needs in the target area.

[0015] As a preferred technical solution, the regional digital space platform module receives video streams from multiple cameras, performs correction and stitching processing on the multiple video streams, and generates a complete scene image. The correction and stitching processing specifically includes:

[0016] A preset distortion correction algorithm is provided, which includes a checkerboard correction algorithm and a latitude and longitude projection correction algorithm.

[0017] When the acquired video stream is a planar image, a distortion correction algorithm is used for processing. The original image is corrected by using an intrinsic parameter matrix and a set of distortion coefficients and an anti-distortion formula to generate a rectangular planar image.

[0018] When the captured video stream is a fisheye image, a latitude and longitude projection correction algorithm is used to convert the fisheye image to a latitude and longitude plane based on the projection mapping relationship between fisheye pixels and focal length.

[0019] On standard test set or online calibration board images, correction experiments are performed on each correction algorithm and a comprehensive score is calculated according to preset quality indicators and weights. The distortion correction algorithm with the highest score is selected as the current optimal correction scheme.

[0020] For any two corrected images, extract the grayscale vectors of the first and second pixels in their overlapping area, calculate the similarity between the first and second pixel grayscale vectors, and when the similarity is greater than a preset similarity threshold, perform image registration and fusion based on minimizing the pixel error in the overlapping area, and finally generate a seamless panoramic monitoring video stream covering the entire pig farm.

[0021] As a preferred technical solution, the dynamic classification in the group-level platform module specifically refers to:

[0022] When the computational resource consumption of a certain intelligent algorithm is lower than the preset lightweight resource threshold, the intelligent algorithm will be sent to the edge device module to run, so as to reduce data transmission loss and improve real-time processing efficiency.

[0023] When the computational resource consumption of the intelligent algorithm exceeds the lightweight resource threshold but does not exceed the preset regional resource threshold, the intelligent algorithm will be deployed on the regional server module to adapt to the medium-scale data processing needs within the region.

[0024] When the computational resource consumption of the intelligent algorithm exceeds the regional-level resource threshold, the intelligent algorithm will be retained within the group-level platform module.

[0025] The priority and business classification management unit is specifically as follows:

[0026] The business tag group includes eight categories: supply chain management, production operation, quality management, safety management, equipment maintenance, environmental monitoring, energy consumption optimization, and risk warning. Based on the relationship between each tag and the preset pig farm production monitoring task table, the intelligent algorithm that needs to be evaluated periodically is assigned a call cycle and the algorithm process is dynamically triggered according to the cycle, realizing the business-oriented scheduling mechanism of the intelligent algorithm.

[0027] For each target region, the AI ​​algorithm center will first generate a business weight configuration based on the business priorities of that region, clarifying the relative importance of each type of business tag in that region;

[0028] Set a tag association status for each intelligent algorithm. If the function of the intelligent algorithm matches a certain type of business tag, it is marked as "associated"; otherwise, it is marked as "not associated".

[0029] The priority score of the intelligent algorithm in a certain target area is obtained by multiplying the importance of each type of business tag in that area with the association status between the intelligent algorithm and the corresponding tag and then summing the results. This ensures that intelligent algorithms with higher priority get computing resources first and adapt to the core business needs of the area.

[0030] Secondly, the present invention provides a deployment method for a system architecture based on a global digital space project, comprising the following steps:

[0031] (1) Multi-terminal access is achieved through the terminal user module, supporting real-time video viewing, equipment scheduling and abnormal warning;

[0032] (2) Deploy cameras, switches and NVRs at the acquisition end to complete video acquisition, aggregation and local storage;

[0033] (3) The regional server module establishes a data link with the NVR, receives video streams and performs preliminary processing, and loads local AI algorithms for real-time behavior analysis;

[0034] (4) The regional digital space platform module performs video stitching, algorithm scheduling and equipment management, and generates panoramic digital twin scenes by stitching together multi-view videos;

[0035] (5) The group-level platform module integrates data through the AI ​​algorithm center and the Internet of Things platform and provides standardized interface services.

[0036] As a preferred technical solution, step (2) specifically includes:

[0037] (2-1) Install multiple cameras above the target area to collect real-time monitoring video data of the target area in order to capture animal dynamics and environmental status information in the farm, wherein the environmental status includes the density and behavior of animals in the target area;

[0038] (2-2) The switch is responsible for the aggregation and forwarding of data collected by multiple cameras, ensuring efficient data transmission to the NVR;

[0039] (2-3) The NVR stores the video data collected by the camera locally for review or analysis at any time;

[0040] (2-4) Transmit video data to the regional server module in real time to provide data support for edge and cloud processing.

[0041] 8. The deployment method of the system architecture based on the full-domain digital space project according to claim 6, characterized in that step (3) specifically comprises:

[0042] (3-1) The regional servers adopt a distributed deployment approach, configuring a regional server for each farm to reduce network load;

[0043] (3-2) Establish a data link with the NVR through the regional server to receive video streams in real time;

[0044] (3-3) Deploy AI algorithms to perform local analysis on the received video data, and adjust computing resources and data flow according to the real-time needs of the farm to ensure efficient operation;

[0045] (3-4) The processed results are uploaded to the group's AI algorithm platform or digital space platform to support further analysis and decision-making.

[0046] As a preferred technical solution, step (4) specifically includes:

[0047] (4-1) Receive video streams from multiple cameras, stitch and process the video streams to generate a complete scene picture, receive video streams transmitted by the regional server, obtain real-time pictures of multiple cameras through point-to-point data communication, collect video streams from edge servers or NVRs, and ensure smooth transmission of raw data.

[0048] (4-2) Perform timestamp calibration on the video stream to ensure that the video times of multiple cameras are consistent, and use spatial mapping technology to adjust the edge position of the camera screen to achieve geometric alignment;

[0049] (4-3) Using auxiliary calibration software, the camera is subjected to distortion correction, perspective transformation, and cropping, and the fused effect is converted into a corresponding configuration file and sent to the camera;

[0050] (4-4) The NVR receives the video streams from the camera group and merges multiple video streams into one output;

[0051] (4-5) The terminal user module provides real-time preview of the panoramic image and the linked independent camera stream.

[0052] As a preferred technical solution, step (5) specifically includes:

[0053] (5-1) Data acquisition and transmission: Collect multi-source data from the regional digital space platform and perform preliminary summarization. The multi-source data includes video, audio, and sensor data within the target area.

[0054] (5-2) Real-time collection of IoT sensor data from regional servers via an IoT platform, the sensor data including environmental parameters and device status;

[0055] (5-3) Receive videos and splicing results uploaded from the regional digital space platform, integrate data from different sources, and complete preliminary data standardization and format conversion;

[0056] (5-4) Unified scheduling and computation of algorithms: Through the AI ​​algorithm center, intelligent analysis of multiple types of data is performed to complete unified computation tasks. The AI ​​algorithm center receives analysis requests sent by business systems or regional platforms, selects appropriate algorithm models according to task types, and dynamically allocates computing resources.

[0057] (5-5) Standardized interface service: Provide a unified and standardized data service interface to external systems. Based on HTTP / REST or gRPC protocol, design a general interface and provide interface documentation, covering data query, analysis result acquisition and status update functions. Business systems can obtain analysis results and environmental status information in real time through API interface, and support users to submit new tasks through standard interface.

[0058] (5-6) System status monitoring and feedback: Real-time monitoring of data transmission links to ensure stable platform operation, automatic detection of algorithm scheduling load, dynamic adjustment of task distribution, and sending alarms and status update information to business systems or regional platforms via API.

[0059] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0060] 1. This invention adopts the concept of digital twin and constructs a panoramic video of pig farm pens through real-time video fusion and splicing technology to achieve panoramic monitoring of the pens.

[0061] 2. This invention integrates panoramic video, AI algorithms and IoT data to form a comprehensive perception and remote control of the on-site environment, and builds a digital space platform for the entire breeding farm.

[0062] 3. This invention connects the entire digital space with relevant business systems to develop AI applications such as remote patrol, remote diagnosis, and asset management, thereby building a centralized operation system for farmers and achieving efficient and standardized management of farmers' assets and production.

[0063] 4. This invention adopts a cloud-edge collaborative computing architecture, with preliminary data processing performed at the edge, reducing network bandwidth pressure, improving the real-time performance and stability of data processing, and adapting to the needs of multi-regional deployment.

[0064] 5. This invention supports the access and unified management of multi-source heterogeneous devices, has good scalability and platform compatibility, and can be widely used in digital space scenarios such as smart agriculture, intelligent manufacturing, and smart parks.

[0065] 6. This invention has multi-terminal access capability, allowing users to remotely access and schedule data via various terminals such as mobile phones, tablets, and computers, thereby improving user experience and management efficiency. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a system architecture diagram of the present invention based on the global digital space project;

[0068] Figure 2 This is a schematic diagram of the deployment and installation of the fisheye camera of the present invention;

[0069] Figure 3 This is a deployment flowchart of the system architecture based on the whole-domain digital space project of this invention;

[0070] Figure 4 This is the original image taken by the fish-eye camera in the pigsty of this invention;

[0071] Figure 5 This is an image obtained by correcting a fisheye camera image according to the present invention;

[0072] Figure 6 This is the image after the corrected image has been transformed according to the present invention;

[0073] Figure 7 This invention is a large image created by moving, cropping, and merging images captured by a fisheye camera;

[0074] Figure 8 This is a schematic diagram of the system interface of the present invention. Detailed Implementation

[0075] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0076] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0077] like Figure 1 As shown, this embodiment provides a system architecture based on a global digital space project, including an end-user module, an edge device module, a regional server module, a regional digital space platform module, and a group-level platform module;

[0078] The terminal user module is used to support multi-terminal access and provide real-time video viewing, device management and AI analysis result display functions;

[0079] Furthermore, the end-user module serves as the system's interaction entry point, supporting access from multiple devices including PCs and mobile devices (such as smartphones and tablets). Users can view real-time panoramic video streams of the farm via the web or a dedicated client, receive AI analysis results (such as alerts for abnormal animal behavior), and remotely control equipment status (such as temperature control equipment). Simultaneously, the end-user module has a built-in notification service, sending alerts to users via a push notification interface for abnormal events such as detected coughing sounds in the pig herd.

[0080] The edge device module is used for video data acquisition and preprocessing, and uploads the processed data to the regional digital space platform module. The edge device module includes cameras, switches, and network video recorders (NVRs) deployed at the acquisition end. The cameras are deployed above the target area to capture dynamic data and environmental status data within the target area. The captured data is sent to the NVR for local storage via the switch. The cameras use multiple fisheye cameras, deployed at the top of the target area, to acquire high-resolution video streams from a top-down angle. The switches are gigabit industrial-grade switches, and VLAN segmentation ensures the bandwidth for video stream transmission. The NVR encodes and compresses the raw video, stores historical data locally, and pushes the real-time stream to the regional server via the RTSP protocol.

[0081] like Figure 2 As shown, in one specific embodiment, the camera uses multiple fisheye cameras, which are deployed on the top of the farm pens to collect high-resolution video streams from a top-down angle, covering data such as animal activity, environmental temperature and humidity.

[0082] The switch uses a gigabit industrial-grade switch to aggregate data from multiple cameras and ensures video stream transmission bandwidth through VLAN segmentation.

[0083] The NVR encodes and compresses the original video using H.265, stores historical data locally, and pushes the real-time stream to the regional server via the RTSP protocol.

[0084] The regional server module adopts a distributed deployment approach to configure a regional server for each target region in order to realize point-to-point video interaction, local AI analysis, and dynamic resource scheduling.

[0085] Furthermore, the regional server module deploys local AI algorithms to analyze data transmitted from the edge device modules and schedule resources based on real-time demand within the target area. The functions of the regional server module include:

[0086] Video preprocessing: This involves frame extraction, resolution adaptation, and Gaussian filtering for noise reduction of the video stream.

[0087] Local AI analytics: Deploy lightweight AI models to detect abnormal animal behaviors (such as falling down or gathering) in real time, and identify cough sound characteristics through Mel spectrum analysis.

[0088] Dynamic load balancing: Adjusts computing resources and data flow based on real-time needs within the farm to ensure efficient operation.

[0089] The regional digital space platform module integrates video stitching, device management, user management, and algorithm scheduling functions to generate panoramic monitoring images of the target area. The module receives video streams from multiple cameras, performs correction and stitching processing on these streams, and generates a complete scene image. The correction and stitching processing specifically involves:

[0090] A preset distortion correction algorithm is provided, which includes a checkerboard correction algorithm and a latitude and longitude projection correction algorithm.

[0091] When the acquired video stream is a planar image, a distortion correction algorithm is used for processing. The original image is corrected by using an intrinsic parameter matrix and a set of distortion coefficients and an anti-distortion formula to generate a rectangular planar image.

[0092] When the captured video stream is a fisheye image, a latitude and longitude projection correction algorithm is used to convert the fisheye image to a latitude and longitude plane based on the projection mapping relationship between fisheye pixels and focal length.

[0093] On standard test set or online calibration board images, correction experiments are performed on each correction algorithm and a comprehensive score is calculated according to preset quality indicators and weights. The distortion correction algorithm with the highest score is selected as the current optimal correction scheme.

[0094] For any two corrected images, extract the grayscale vectors of the first and second pixels in their overlapping area, calculate the similarity between the first and second pixel grayscale vectors, and when the similarity is greater than a preset similarity threshold, perform image registration and fusion based on minimizing the pixel error in the overlapping area, and finally generate a seamless panoramic monitoring video stream covering the entire pig farm.

[0095] In a specific example, the regional digital space platform module integrates a multi-pigpen camera correction and stitching algorithm, specifically including:

[0096] Candidate correction algorithm library: Preset Distortion correction algorithm It supports two correction modes: checkerboard and latitude / longitude, and can be dynamically expanded to new algorithms;

[0097] Checkerboard distortion correction submodule: uses an intrinsic parameter matrix With the set of distortion coefficients The original image is processed using the following formula. Perform anti-distortion to generate a rectangular planar image:

[0098] ;

[0099] Latitude and longitude projection correction submodule: based on fisheye pixels With focal length The projection mapping relationship is used to transform the fisheye image to the latitude and longitude plane, as shown in the following formula:

[0100] ;

[0101] ;

[0102] in, The original image plane coordinates, The equivalent focal length of the lens; This represents the longitude angle of the corresponding pixel in the spherical coordinate system. The corresponding latitude angle;

[0103] Automatic evaluation and selection submodule: For each correction algorithm, evaluates and selects on a standard test set or online calibration board image. Perform correction experiments and calculate three quality indicators respectively. Based on preset weights Calculate the overall score using the following formula:

[0104] ;

[0105] Choose the algorithm with the highest score. As the current optimal correction scheme, when a new algorithm is added, a re-evaluation process is automatically triggered to include the new algorithm in the candidate library in order to continuously improve the correction quality.

[0106] The stitching submodule: stitches any two corrected images in their overlapping area. Extract pixel grayscale vectors And calculate the similarity using the following formula:

[0107] ;

[0108] when ( When the preset similarity threshold is used, image registration and fusion are performed based on minimizing the pixel error in the overlapping area, and finally a seamless panoramic monitoring video stream covering the entire pig farm is generated.

[0109] The group-level platform module includes an AI algorithm center, an IoT data platform, an algorithm classification unit, and a priority and business classification management unit. The AI ​​algorithm center is used for intelligent analysis of multiple types of data to complete unified computing tasks. The algorithm classification unit is used to dynamically classify the intelligent algorithms within the AI ​​algorithm center, based on the average computing resource consumption of each intelligent algorithm during operation, thereby determining the optimal deployment location for each intelligent algorithm. The priority and business classification management unit is used to assign running priority scores to each intelligent algorithm and attach corresponding business tag groups to the intelligent algorithms. The IoT data platform collects IoT sensor data from regional servers in real time.

[0110] Furthermore, the dynamic classification specifically refers to:

[0111] When the computational resource consumption of a certain intelligent algorithm is lower than the preset lightweight resource threshold, the intelligent algorithm will be sent to the edge device module to run, so as to reduce data transmission loss and improve real-time processing efficiency.

[0112] When the computational resource consumption of the intelligent algorithm exceeds the lightweight resource threshold but does not exceed the preset regional resource threshold, the intelligent algorithm will be deployed on the regional server module to adapt to the medium-scale data processing needs within the region.

[0113] When the computational resource consumption of an intelligent algorithm exceeds the regional-level resource threshold, the intelligent algorithm will be retained within the group-level platform module.

[0114] Furthermore, the classification is calculated based on the following formula:

[0115] ;

[0116] in, Let be the resource consumption of the i-th intelligent algorithm. Its average CPU utilization, Its average GPU memory usage;

[0117] when At that time, the intelligent algorithm is distributed to the edge device module for execution;

[0118] when At that time, the intelligent algorithm was deployed on the regional server module;

[0119] when At that time, execution is retained in the group-level platform module;

[0120] in, and These are the preset lightweight and regional resource thresholds, respectively.

[0121] Furthermore, the priority and business classification management unit specifically includes:

[0122] The business tag group includes eight categories: supply chain management, production operation, quality management, safety management, equipment maintenance, environmental monitoring, energy consumption optimization, and risk warning. Based on the relationship between each tag and the preset pig farm production monitoring task table, the intelligent algorithm that needs to be evaluated periodically is assigned a call cycle and the algorithm process is dynamically triggered according to the cycle, realizing the business-oriented scheduling mechanism of the intelligent algorithm.

[0123] For each target region, the AI ​​algorithm center will first generate a business weight configuration based on the business priorities of that region, clarifying the relative importance of each type of business tag in that region;

[0124] Set a tag association status for each intelligent algorithm. If the function of the intelligent algorithm matches a certain type of business tag, it is marked as "associated"; otherwise, it is marked as "not associated".

[0125] The priority score of the intelligent algorithm in a certain target area is obtained by multiplying the importance of each type of business tag in that area with the association status between the intelligent algorithm and the corresponding tag and then summing the results. This ensures that intelligent algorithms with higher priority get computing resources first and adapt to the core business needs of the area.

[0126] Furthermore, the specific implementation process of the priority and business classification management unit is as follows:

[0127] The business tag set This includes: supply chain management, production operations, quality management, safety management, equipment maintenance, environmental monitoring, energy consumption optimization, and risk warning. The group-level platform allocates call cycles for intelligent algorithms that require periodic evaluation based on each tag and a pre-set table of pig farm production monitoring tasks. The algorithm process is dynamically triggered according to the cycle, realizing a business-oriented scheduling mechanism for intelligent algorithms.

[0128] Algorithm Center According to the first The business evaluation results of each pig farm are used to generate a weight vector. ,in Indicates the first Business category tags in the first Relative importance in individual pig farms;

[0129] Each intelligent algorithm With label vector When intelligent algorithms Related to the first When using class tags, ,otherwise ;

[0130] intelligent algorithms At the pig farm Run priority score The calculation is as follows:

[0131] ;

[0132] This invention integrates multi-terminal access, video acquisition and preprocessing, distributed AI analysis, panoramic stitching (including multi-scene camera correction and seamless fusion), and multi-source data integration through a cloud-edge collaborative computing model. It innovatively introduces a dynamic classification and deployment mechanism for algorithm components (adapting edge, regional, and group-level deployments based on computing resource consumption) and a business-oriented algorithm scheduling system (allocating priorities and call cycles based on regional business priorities). This achieves efficient sharing of data across the entire domain, high-quality panoramic video presentation, optimized configuration of AI algorithm resources, and deep adaptation to business needs. It effectively reduces network bandwidth pressure, improves the real-time performance of data processing and system flexibility, supports large-scale deployment across regions, and significantly enhances the intelligence and standardization of full-domain monitoring and management in scenarios such as smart farming and intelligent manufacturing, helping to achieve core objectives such as remote control, risk warning, and efficient operation.

[0133] like Figure 3 As shown, another embodiment of this application also provides a deployment method for a system architecture based on a global digital space project, including the following steps:

[0134] (1) Multi-terminal access is achieved through the terminal user module, supporting real-time video viewing, equipment scheduling and abnormal warning;

[0135] Furthermore, in step (1), the end-user module supports multi-terminal access, including PC, mobile (phone, tablet) and dedicated client. Users can view the panoramic video stream of the farm in real time through the web interface or mobile application, receive AI analysis results (such as early warning of abnormal animal behavior and prompts of excessive environmental parameters), and remotely control equipment (such as ventilation system and feeding device).

[0136] Real-time viewing includes panoramic stitched video of the farm and video from individual cameras in each channel; it supports AI-based early warning display, including animal gathering, low activity, fighting, equipment failure, and environmental parameter exceeding standards; users can issue remote control commands, such as adjusting fan speed, controlling feeding cycle, and controlling the on / off of lights.

[0137] (2) Deploy cameras, switches and NVRs at the acquisition end to complete video acquisition, aggregation and local storage;

[0138] Furthermore, step (2) specifically involves:

[0139] (2-1) Multiple cameras are installed above the target area to collect real-time monitoring video data of the target area, in order to capture animal dynamics and environmental status information within the farm. The environmental status includes the density and behavior of animals within the target area. Multiple high-resolution wide-angle cameras are installed on the top of the target breeding area (such as pigsty) and key passages, and PoE-powered industrial Ethernet switches are used to aggregate and forward the video streams. One camera is configured for every 30 square meters, with a shooting coverage angle of 130°.

[0140] (2-2) The switch is responsible for the aggregation and forwarding of data collected by multiple cameras, ensuring efficient data transmission to the NVR; the video stream is stored in the NVR according to the channel encoding and is set to be covered in a 7-day cycle; the data is uploaded to the regional server through the local area network, and the transmission protocol is RTSP or ONVIF standard interface.

[0141] (2-3) The NVR stores the video data collected by the camera locally, which is convenient for review or analysis at any time;

[0142] (2-4) Transmit video data to the regional server module in real time to provide data support for edge and cloud processing.

[0143] (3) The regional server module establishes a data link with the NVR, receives video streams and performs preliminary processing, and loads local AI algorithms for real-time behavior analysis;

[0144] Furthermore, step (3) specifically involves:

[0145] (3-1) The regional servers adopt a distributed deployment approach, configuring a regional server for each farm to reduce network load. The percentage improvement in performance is shown in the following formula:

[0146] ;

[0147] Where P represents the percentage performance improvement, P d P represents the performance of distributed deployment. t Represents the performance of traditional deployments;

[0148] (3-2) Establish a data link with the NVR through the regional server to receive video streams in real time;

[0149] (3-3) Deploy AI algorithms to perform local analysis on the received video data, and adjust computing resources and data flow according to the real-time needs of the farm to ensure efficient operation; the regional server establishes a data link with the NVR through the LAN network, receives video data in real time, and loads local AI models to complete preliminary analysis. AI models include: YOLOv7-PRN for detecting pig behavior, DeepSort for multi-target tracking, and ViT model for detecting abnormal behavior in the area;

[0150] (3-4) The processed results are uploaded to the group's AI algorithm platform or digital space platform to support further analysis and decision-making.

[0151] (3-5) The regional digital space platform module performs video stitching, algorithm scheduling and equipment management, and stitches multi-view videos to generate panoramic digital twin scenes; the processing results include behavior trajectories, heat maps, behavior anomaly counts, etc., and are packaged into JSON format for uploading.

[0152] Furthermore, step (4) specifically involves:

[0153] (4-1) Receive video streams from multiple cameras, stitch and process the video streams to generate a complete scene picture, receive video streams transmitted by the regional server, obtain real-time pictures of multiple cameras through point-to-point data communication, collect video streams from edge servers or NVRs, and ensure smooth transmission of raw data.

[0154] (4-2) Perform timestamp calibration on the video stream to ensure that the video times of multiple cameras are consistent, and use spatial mapping technology to adjust the edge position of the camera screen to achieve geometric alignment;

[0155] (4-3) Using auxiliary calibration software, the camera is subjected to distortion correction, perspective transformation, and cropping, and the fused effect is converted into a corresponding configuration file and sent to the camera;

[0156] (4-4) The NVR receives the video streams from the camera group and merges multiple video streams into one output; a multi-view stitching algorithm (based on SIFT and RANSAC matching key points) is used to synthesize a complete scene; the video frames are synchronized: the timestamps are synchronized using an NTP server with an error of no more than 50ms; video transformation, geometric alignment and distortion correction are implemented using OpenCV and FFmpeg;

[0157] (4-5) The terminal user module provides real-time preview of the panoramic image and the linked independent camera stream.

[0158] (5) The group-level platform module integrates data through the AI ​​algorithm center and the IoT platform, and provides standardized interface services. After generating a digital twin scene, it is presented on the front end in WebGL form and bound to the physical device ID to realize remote control and status feedback in three-dimensional space.

[0159] Furthermore, step (5) specifically involves:

[0160] (5-1) Data acquisition and transmission: Collect multi-source data from the regional digital space platform and perform preliminary summarization. The multi-source data includes video, audio, and sensor data within the target area.

[0161] (5-2) Real-time collection of IoT sensor data from regional servers via an IoT platform, the sensor data including environmental parameters and device status;

[0162] (5-3) Receive videos and splicing results uploaded from the regional digital space platform, integrate data from different sources, and complete preliminary data standardization and format conversion;

[0163] (5-4) Unified scheduling and computation of algorithms: Through the AI ​​algorithm center, intelligent analysis of multiple types of data is performed to complete unified computation tasks. The AI ​​algorithm center receives analysis requests sent by business systems or regional platforms, selects appropriate algorithm models according to task types, and dynamically allocates computing resources.

[0164] (5-5) Standardized interface service: Provide a unified and standardized data service interface to external systems. Based on HTTP / REST or gRPC protocol, design a general interface and provide interface documentation, covering data query, analysis result acquisition and status update functions. Business systems can obtain analysis results and environmental status information in real time through API interface, and support users to submit new tasks through standard interface.

[0165] (5-6) System status monitoring and feedback: Real-time monitoring of data transmission links to ensure stable platform operation, automatic detection of algorithm scheduling load, dynamic adjustment of task distribution, and sending alarms and status update information to business systems or regional platforms via API.

[0166] like Figures 4-8 As shown, applying the technical solution of this invention to a pig farm, in a typical smart pig farm, the pig farm system architecture and deployment method based on a full-domain digital space proposed in this invention can realize intelligent perception and management of the entire process of pig behavior and health status. Figure 4 As shown, the pig farm has formed a sensing network covering key areas inside and outside the pigsty by deploying audio and video acquisition terminals in different areas. The system uses the modular deployment architecture proposed in this invention to upload the audio and video data collected at the front end to edge computing nodes or a cloud platform for real-time processing.

[0167] like Figure 5 As shown, the system completes functions such as individual pig identification, behavior recognition, and movement trajectory tracking at the edge through intelligent algorithm modules, effectively supporting the estimation of pig density in the shed, monitoring of feeding and drinking behavior, and detection of abnormal movement. Figure 6 The demonstration showcased the platform's centralized analysis process in the cloud, including health status reasoning and behavioral anomaly early warning based on large models and knowledge graphs, effectively improving the early warning capabilities of pig farms.

[0168] Figure 7 and Figure 8 The platform's human-computer interaction and data visualization capabilities were further demonstrated. Farmers can view the real-time operational status of pigsties, individual behavior statistics, and alarm information through a digital twin interface, enabling a shift from traditional experience-based management to data-driven intelligent management. The overall system streamlines the entire process of perception, analysis, decision-making, and feedback, significantly improving pig farm management efficiency and pig health.

[0169] The application of the technical solution of this invention in pig farms enables real-time perception of pig behavior, intelligent diagnosis of health status, and digital transformation of breeding management, providing a practical and feasible intelligent upgrade path for modern pig farms.

[0170] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0171] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A system architecture based on a full-domain digital space project, characterized in that: This includes end-user modules, edge device modules, regional server modules, regional digital space platform modules, and group-level platform modules; The terminal user module is used to support multi-terminal access and enable real-time video viewing, device management, and AI analysis result display; The edge device module is used for video data acquisition and preprocessing, and uploads the preprocessed data to the regional digital space platform module. The edge device module includes cameras, switches and network video recorders (NVRs) deployed at the acquisition end. The regional server module adopts a distributed deployment approach to configure a regional server for each target region in order to realize point-to-point video interaction, local AI analysis and dynamic resource scheduling. The regional digital space platform module integrates video stitching, device management, user management, and algorithm scheduling functions to generate panoramic monitoring images of the target area. The group-level platform module includes an AI algorithm center, an IoT data platform, an algorithm classification unit, and a priority and business classification management unit. The AI ​​algorithm center is used for intelligent analysis of multiple types of data to complete unified computing tasks. The algorithm classification unit is used to dynamically classify the intelligent algorithms within the AI ​​algorithm center, based on the average computing resource consumption of each intelligent algorithm during operation, thereby determining the optimal deployment location for each intelligent algorithm. The priority and business classification management unit is used to assign running priority scores to each intelligent algorithm and attach corresponding business tag groups to the intelligent algorithms. The IoT data platform collects IoT sensor data from regional servers in real time.

2. The system architecture based on the global digital space project according to claim 1, characterized in that, In the edge device module, cameras are deployed above the target area to capture dynamic data and environmental status data within the target area. The captured data is sent to the NVR for local storage via a switch. The cameras consist of multiple fisheye cameras deployed at the top of the target area to capture high-resolution video streams from a top-down angle. The switch is a gigabit industrial-grade switch, and the bandwidth for video stream transmission is guaranteed through VLAN segmentation; the NVR encodes and compresses the original video, stores historical data locally, and pushes the real-time stream to the regional server through the RTSP protocol.

3. The system architecture based on the global digital space project according to claim 1, characterized in that, The regional server module deploys local AI algorithms to perform frame extraction, resolution adaptation, and Gaussian filtering noise reduction on the video stream transmitted from the edge device module. Based on the deployed local AI algorithms, it detects abnormal behavior and identifies features in the target area in real time, and adjusts computing resources and data flow according to the real-time needs of the target area.

4. The system architecture based on the global digital space project according to claim 1, characterized in that, The regional digital space platform module receives video streams from multiple cameras, performs correction and stitching processing on the multiple video streams, and generates a complete scene image. The correction and stitching processing specifically includes: A preset distortion correction algorithm is provided, which includes a checkerboard correction algorithm and a latitude and longitude projection correction algorithm. When the acquired video stream is a planar image, a distortion correction algorithm is used for processing. The original image is corrected by using an intrinsic parameter matrix and a set of distortion coefficients and an anti-distortion formula to generate a rectangular planar image. When the captured video stream is a fisheye image, a latitude and longitude projection correction algorithm is used to convert the fisheye image to a latitude and longitude plane based on the projection mapping relationship between fisheye pixels and focal length. On standard test set or online calibration board images, correction experiments are performed on each correction algorithm and a comprehensive score is calculated according to preset quality indicators and weights. The distortion correction algorithm with the highest score is selected as the current optimal correction scheme. For any two corrected images, extract the grayscale vectors of the first and second pixels in their overlapping area, calculate the similarity between the first and second pixel grayscale vectors, and when the similarity is greater than a preset similarity threshold, perform image registration and fusion based on minimizing the pixel error in the overlapping area, and finally generate a seamless panoramic monitoring video stream covering the entire pig farm.

5. The system architecture based on the global digital space project according to claim 1, characterized in that, In the aforementioned group-level platform module, the dynamic classification specifically refers to: When the computational resource consumption of a certain intelligent algorithm is lower than the preset lightweight resource threshold, the intelligent algorithm will be sent to the edge device module to run, so as to reduce data transmission loss and improve real-time processing efficiency. When the computational resource consumption of the intelligent algorithm exceeds the lightweight resource threshold but does not exceed the preset regional resource threshold, the intelligent algorithm will be deployed on the regional server module to adapt to the medium-scale data processing needs within the region. When the computational resource consumption of the intelligent algorithm exceeds the regional-level resource threshold, the intelligent algorithm will be retained within the group-level platform module. The priority and business classification management unit is specifically as follows: The business tag group includes eight categories: supply chain management, production operation, quality management, safety management, equipment maintenance, environmental monitoring, energy consumption optimization, and risk warning. Based on the relationship between each tag and the preset pig farm production monitoring task table, the intelligent algorithm that needs to be evaluated periodically is assigned a call cycle and the algorithm process is dynamically triggered according to the cycle, realizing the business-oriented scheduling mechanism of the intelligent algorithm. For each target region, the AI ​​algorithm center will first generate a business weight configuration based on the business priorities of that region, clarifying the relative importance of each type of business tag in that region; Set a tag association status for each intelligent algorithm. If the function of the intelligent algorithm matches a certain type of business tag, it is marked as "associated"; otherwise, it is marked as "not associated". The priority score of the intelligent algorithm in a certain target area is obtained by multiplying the importance of each type of business tag in that area with the association status between the intelligent algorithm and the corresponding tag and then summing the results. This ensures that intelligent algorithms with higher priority get computing resources first and adapt to the core business needs of the area.

6. The deployment method of the system architecture based on the global digital space project according to any one of claims 1-5, characterized in that, Includes the following steps: (1) Multi-terminal access is achieved through the terminal user module, supporting real-time video viewing, equipment scheduling and abnormal warning; (2) Deploy cameras, switches and NVRs at the acquisition end to complete video acquisition, aggregation and local storage; (3) The regional server module establishes a data link with the NVR, receives video streams and performs preliminary processing, and loads local AI algorithms for real-time behavior analysis; (4) The regional digital space platform module performs video stitching, algorithm scheduling and equipment management, and generates panoramic digital twin scenes by stitching together multi-view videos; (5) The group-level platform module integrates data through the AI ​​algorithm center and the Internet of Things platform and provides standardized interface services.

7. The deployment method of the system architecture based on the full-domain digital space project according to claim 6, characterized in that, Step (2) specifically involves: (2-1) Install multiple cameras above the target area to collect real-time monitoring video data of the target area in order to capture animal dynamics and environmental status information in the farm, wherein the environmental status includes the density and behavior of animals in the target area; (2-2) The switch is responsible for the aggregation and forwarding of data collected by multiple cameras, ensuring efficient data transmission to the NVR; (2-3) The NVR stores the video data collected by the camera locally for review or analysis at any time; (2-4) Transmit video data to the regional server module in real time to provide data support for edge and cloud processing.

8. The deployment method of the system architecture based on the global digital space project according to claim 6, characterized in that, The specific steps (3) are as follows: (3-1) The regional servers adopt a distributed deployment approach, configuring a regional server for each farm to reduce network load; (3-2) Establish a data link with the NVR through the regional server to receive video streams in real time; (3-3) Deploy AI algorithms to perform local analysis on the received video data, and adjust computing resources and data flow according to the real-time needs of the farm to ensure efficient operation; (3-4) The processed results are uploaded to the group's AI algorithm platform or digital space platform to support further analysis and decision-making.

9. The deployment method of the system architecture based on the global digital space project according to claim 6, characterized in that, Step (4) specifically involves: (4-1) Receive video streams from multiple cameras, stitch and process the video streams to generate a complete scene picture, receive video streams transmitted by the regional server, obtain real-time pictures of multiple cameras through point-to-point data communication, collect video streams from edge servers or NVRs, and ensure smooth transmission of raw data. (4-2) Perform timestamp calibration on the video stream to ensure that the video times of multiple cameras are consistent, and use spatial mapping technology to adjust the edge position of the camera screen to achieve geometric alignment; (4-3) Using auxiliary calibration software, the camera is subjected to distortion correction, perspective transformation, and cropping, and the fused effect is converted into a corresponding configuration file and sent to the camera; (4-4) The NVR receives the video streams from the camera group and merges multiple video streams into one output; (4-5) The terminal user module provides real-time preview of the panoramic image and the linked independent camera stream.

10. The deployment method of the system architecture based on the global digital space project according to claim 6, characterized in that, Step (5) specifically involves: (5-1) Data acquisition and transmission: Collect multi-source data from the regional digital space platform and perform preliminary summarization. The multi-source data includes video, audio, and sensor data within the target area. (5-2) Real-time collection of IoT sensor data from regional servers via an IoT platform, the sensor data including environmental parameters and device status; (5-3) Receive videos and splicing results uploaded from the regional digital space platform, integrate data from different sources, and complete preliminary data standardization and format conversion; (5-4) Unified scheduling and computation of algorithms: Through the AI ​​algorithm center, intelligent analysis of multiple types of data is performed to complete unified computation tasks. The AI ​​algorithm center receives analysis requests sent by business systems or regional platforms, selects appropriate algorithm models according to task types, and dynamically allocates computing resources. (5-5) Standardized interface service: Provide a unified and standardized data service interface to external systems. Based on HTTP / REST or gRPC protocol, design a general interface and provide interface documentation, covering data query, analysis result acquisition and status update functions. Business systems can obtain analysis results and environmental status information in real time through API interface, and support users to submit new tasks through standard interface. (5-6) System status monitoring and feedback: Real-time monitoring of data transmission links to ensure stable platform operation, automatic detection of algorithm scheduling load, dynamic adjustment of task distribution, and sending alarms and status update information to business systems or regional platforms via API.

Images