Intelligent management system for comprehensive sports ground

Abstract

This application provides an intelligent management system for integrated sports fields. By organically combining user terminals, sensing and execution devices, edge computing servers, and a cloud management platform, it constructs a full-link intelligent management system covering "user interaction - data collection - local real-time processing - cloud-based in-depth analysis." Addressing pain points in traditional sports field management such as cumbersome reservations, delayed status perception, and insufficient data support, this system utilizes edge computing and local AI model deployment to achieve real-time response to local events, avoiding response delays caused by cloud transmission latency. Simultaneously, it leverages the big data analytics capabilities of the cloud platform to provide data support for operational decisions. This combination satisfies both the immediate needs of on-site management and the decision-making requirements of macro-operation, thereby comprehensively improving the management efficiency and service quality of integrated sports fields and realizing intelligent collaboration among "people, equipment, venues, and data" in sports scenarios.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intelligent management system for a comprehensive sports field. Background Technology

[0002] With the increasing awareness of fitness among the public and the rapid development of the sports industry, multi-purpose sports stadiums, as public or commercial sports facilities integrating multiple sports, are experiencing increasingly higher usage frequency and management complexity. Traditional sports stadium management models rely heavily on manual inspections and paper-based registration, which have several drawbacks: First, the venue reservation process is cumbersome, requiring users to confirm reservations by phone or in person, easily leading to reservation conflicts or information delays; second, the lack of real-time monitoring of venue usage status makes it difficult for managers to promptly grasp the population density and facility condition in different areas, resulting in uneven resource allocation or safety hazards; third, weak data collection and analysis capabilities during sports activities, failing to provide users with feedback on their performance and hindering accurate decision-making by venue operators; fourth, low emergency response efficiency, making it difficult to trigger timely rescue or repair mechanisms in the event of injuries or facility malfunctions. Furthermore, the significant differences in venue requirements across different sports make it difficult for traditional management systems to achieve unified scheduling and intelligent adaptation across sports, thus restricting the service quality and operational efficiency of multi-purpose sports stadiums. Summary of the Invention

[0003] The purpose of this application is to address at least one of the aforementioned technical deficiencies, particularly the insufficient level of intelligence in existing integrated sports stadiums, which restricts the service quality and operational efficiency of integrated sports stadiums.

[0004] This application provides an intelligent management system for a comprehensive sports field. The system includes a user terminal, a sensing and execution device layer deployed on the sports field, an edge computing server, and a cloud management platform.

[0005] The user terminal is used to provide users with venue reservation, status query, data reception and interactive interface;

[0006] The sensing and execution device layer includes various sensors for collecting data on the sports field environment, usage status, and personnel activities, as well as controlled devices for executing control commands.

[0007] The edge computing server, deployed locally in the sports venue, is used to receive and process sensor data from the sensing and execution device layer in real time, call a pre-deployed local artificial intelligence algorithm model to identify specific events or states related to sports venue management from the sensor data, generate corresponding control commands based on the specific events or states and send them to the controlled devices, and / or generate early warning information.

[0008] The cloud management platform is communicatively connected to the edge computing server and the user terminal. It is used to collect and store processing result data related to the specific event or state from the edge computing server, as well as business and interaction data from the user terminal. It integrates, deeply analyzes and statistically processes all the collected data to generate comprehensive management reports and provides administrators with a decision support interface based on the comprehensive management reports.

[0009] The local artificial intelligence algorithm model deployed on the edge computing server includes at least a visual analysis model for real-time analysis of video streams to identify specific events or states.

[0010] Optionally, the specific events or states identified by the edge computing server include athlete injury events;

[0011] The local artificial intelligence algorithm model includes a posture analysis model based on skeletal key point detection, which is used to identify action postures that conform to preset injury determination rules from the video stream;

[0012] The preset injury determination rules include detecting a rapid fall caused by a sudden drop in a person's center of gravity, and the fall posture lasting for more than a preset time threshold without returning to a standing or walking posture.

[0013] Optionally, the process by which the edge computing server generates corresponding control commands and sends them to the controlled device, and / or generates early warning information, includes:

[0014] Trigger the site broadcasting system to play a preset warning message;

[0015] Push notifications containing the location information of the event to the terminals of administrators, pre-set medical staff, and corresponding appointment users;

[0016] Keep the access control equipment at the site entrance open to allow rescue personnel to enter.

[0017] Optionally, the specific events or states identified by the edge computing server include basketball-related events;

[0018] The local artificial intelligence algorithm model includes a basketball motion analysis model, used to identify goal events from video streams, distinguish goal types, and / or track player trajectories to collect individual motion data.

[0019] Optionally, the specific events or states identified by the edge computing server include football-related events;

[0020] The sensing and actuation device layer includes a goal line detection device deployed on the football goal line;

[0021] The local artificial intelligence algorithm model includes a football motion analysis model, used to process the data from the goal line detection device to determine whether a goal is valid, and / or to track player trajectories from the video stream to collect running and tactical data.

[0022] Optionally, the specific events or states identified by the edge computing server include tennis-related events;

[0023] The local artificial intelligence algorithm model includes a tennis motion analysis model, which is used to reconstruct the three-dimensional trajectory and calculate the landing point of the tennis ball based on multi-angle video streams, in order to determine out-of-bounds events and / or statistically analyze serve speed and shot type data.

[0024] Optionally, the specific events or states identified by the edge computing server include competition violation events;

[0025] The local artificial intelligence algorithm model includes a violation action recognition model, which is used to identify preset types of foul actions or violations in basketball or football games from video streams;

[0026] In response to the competition violation, the edge computing server sends auxiliary judgment information, including the event time and video clip, to the designated referee terminal.

[0027] Optionally, the system supports a video playback challenge mechanism;

[0028] The user terminal is configured to receive a penalty challenge request initiated by the user;

[0029] The edge computing server or the cloud management platform is configured to automatically retrieve multi-angle video from the relevant time period and generate replay clips in response to the penalty challenge request, for the referee's terminal to retrieve and view.

[0030] Optionally, the specific events or states identified by the edge computing server include the state of exceeding the limit for the number of people in the venue;

[0031] The local artificial intelligence algorithm model includes a real-time people counting model based on target detection and tracking, which is used to identify from the video stream that the current number of people exceeds the preset capacity limit of the corresponding venue;

[0032] In response to the overcrowding status of the venue, the edge computing server sends a locking command to the entrance access control device and / or sends a full-occupancy message to the venue display screen and pushes an overcrowding alarm to the administrator terminal.

[0033] Optionally, the sensing and actuation device layer includes a light intensity sensor and a controllable lighting device;

[0034] The edge computing server is configured as follows:

[0035] The ambient light level is determined based on the data collected by the light intensity sensor.

[0036] Based on the ambient light level and the current usage status of the site, control commands are generated to adjust the brightness or on / off status of the controllable lighting equipment.

[0037] Optionally, the specific events or states identified by the edge computing server include abnormal states of site facilities;

[0038] The sensing data includes images or sensor data that reflect the condition of the site surface;

[0039] The local artificial intelligence algorithm model or analysis program is used to identify at least one abnormal state from the sensor data, including ground cracks, potholes, fading, or lawn health below a preset health threshold.

[0040] In response to the abnormal status of the site facilities, the cloud management platform automatically generates a maintenance work order that includes the location and type of the abnormality.

[0041] Optionally, the comprehensive management reports generated by the cloud management platform include at least one of the following: site utilization heat map, user behavior analysis profile, facility health report, and revenue analysis report;

[0042] The decision support interface provides operational optimization suggestions based on the comprehensive management reports.

[0043] Optionally, the user terminal is also used to respond to the user's venue reservation operation and generate a unique electronic voucher containing reservation information;

[0044] The sensing and execution device layer also includes a credential verification device located at the site entrance;

[0045] The edge computing server or cloud management platform is also used to: after the credential verification device reads the unique electronic credential, verify the validity of the unique electronic credential, and control the controlled equipment set at the entrance of the venue to be turned on or kept off based on the verification result;

[0046] The verification of the validity of the unique electronic voucher includes at least checking whether the reservation period is within the validity period and whether the current number of people using the venue does not exceed the number registered in the reservation information.

[0047] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0048] This application provides an intelligent management system for integrated sports fields. By organically combining user terminals, sensing and execution devices, edge computing servers, and a cloud management platform, it constructs a full-link intelligent management system covering "user interaction - data collection - local real-time processing - cloud-based in-depth analysis." Addressing pain points in traditional sports field management such as cumbersome reservations, delayed status perception, and insufficient data support, this system utilizes edge computing and local AI model deployment to achieve real-time response to local events, avoiding response delays caused by cloud transmission latency. Simultaneously, it leverages the big data analytics capabilities of the cloud platform to provide data support for operational decisions. This combination satisfies both the immediate needs of on-site management and the decision-making requirements of macro-operation, thereby comprehensively improving the management efficiency and service quality of integrated sports fields. It breaks down information silos in traditional management and achieves intelligent collaboration between "people, equipment, venues, and data" in sports scenarios. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0050] Figure 1 A schematic diagram of the structure of an intelligent management system for a comprehensive sports field provided in an embodiment of this application;

[0051] Figure 2 This is a schematic diagram illustrating the interaction relationships between various modules in the intelligent management system provided in this application embodiment. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] In one embodiment, such as Figure 1 As shown, Figure 1 This is a schematic diagram of the structure of an intelligent management system for a comprehensive sports field provided in an embodiment of this application. The application provides an intelligent management system for a comprehensive sports field, which includes a user terminal, a sensing and execution device layer deployed on the sports field, an edge computing server, and a cloud management platform.

[0054] The user terminal is used to provide users with venue reservation, status inquiry, data reception and interactive interface.

[0055] The sensing and execution device layer includes various sensors for collecting data on the sports field environment, usage status, and personnel activities, as well as controlled devices for executing control commands.

[0056] The edge computing server, deployed locally in the sports venue, is used to receive and process sensor data from the sensing and execution device layer in real time, call a pre-deployed local artificial intelligence algorithm model to identify specific events or states related to sports venue management from the sensor data, generate corresponding control commands based on the specific events or states and send them to the controlled devices, and / or generate early warning information.

[0057] The cloud management platform is communicatively connected to the edge computing server and the user terminal. It is used to collect and store processing result data related to the specific event or state from the edge computing server, as well as business and interaction data from the user terminal. It integrates, deeply analyzes and statistically processes all the collected data to generate comprehensive management reports and provides administrators with a decision support interface based on the comprehensive management reports.

[0058] The local artificial intelligence algorithm model deployed on the edge computing server includes at least a visual analysis model for real-time analysis of video streams to identify specific events or states.

[0059] In this embodiment, the user terminal can implement the functions through lightweight application carriers such as mini-programs and apps, supporting users to make time-sharing reservations for multiple venue types (such as basketball courts, football fields, and tennis courts) on their mobile devices. During reservation, users can simultaneously view real-time data on the target venue, including the number of users, ambient temperature and humidity, and light intensity. For example, the system of this application supports multiple venue types such as basketball courts (full-court / half-court), football fields (11-a-side / 5-a-side), tennis courts, and badminton courts. Each venue has an electronic profile, including information such as venue size, facility status, maximum capacity, and pricing. When making reservations on their mobile devices, users can view the electronic profiles of each venue to select one that meets their needs. Furthermore, users can view the real-time status of the venue through the app: green indicates available for reservation, blue indicates reserved and awaiting use, and red indicates in use. The system of this application integrates a GIS map, allowing users to filter by distance, price, and venue type at multiple levels.

[0060] Furthermore, this application can also include a no-show mechanism when booking venues. For example, users can pay the full amount or a deposit when booking; cancellations within 2 hours of the reservation date will incur a 50% fee deduction, and cancellations within 1 hour will incur a full fee deduction. Three no-shows will result in a 7-day suspension of booking privileges. The specific details of the no-show mechanism can be flexibly adjusted according to the venue's operational strategy. For example, the cancellation deadline could be extended to 1 hour in advance for members, or a flexible scheme allowing users to use points to offset the penalty could be offered.

[0061] Indicatively, such as Figure 2 As shown, Figure 2 This diagram illustrates the interaction between modules in the intelligent management system provided in this application embodiment. The sensor types in the sensing and execution device layer of this application cover multi-dimensional data acquisition needs. Environmental sensors include temperature and humidity sensors, PM2.5 detectors, and illuminance sensors, deployed around the perimeter and on the ceiling, collecting data at preset frequencies and transmitting it to the edge computing server. Personnel and motion status sensors include high-definition network cameras (at least three deployed per venue, covering the entire area without blind spots), millimeter-wave radar (aiding in accurate personnel location identification in low-light environments), access control card readers, and specialized devices for specific projects, such as pressure sensor arrays for soccer goal lines and ground vibration sensors for tennis courts. Facility status sensors include vibration sensors and current monitoring modules deployed on key facilities such as basketball hoops, nets, and lighting equipment, monitoring the structural stability and operational status of the facilities in real time. Controlled devices include venue access control systems, public address systems, intelligent lighting fixtures, electric sunshades, and electronic displays and environmental control equipment around the venue; all devices support remote control and status feedback from the edge computing server.

[0062] like Figure 2As shown, the edge computing server in this application can adopt an industrial-grade hardware architecture and be equipped with a high-performance GPU accelerator card to meet the real-time computing needs of local AI models. Its processing flow follows a closed-loop logic of "data reception - preprocessing - model inference - event judgment - instruction generation". The events processed include, but are not limited to, access control events, security management events, sports injury events, facility status events, and sports data events. Taking the recognition of sports injury events as an example, the visual analysis model can first extract frames from the video stream transmitted by the camera, processing 25 frames per second. Through the skeletal key point detection model, it locates 17 key nodes of the human body, such as the head, torso, and limbs, and calculates the rate of change of spatial coordinates of each node in real time. When the vertical velocity of the center of gravity node (such as the hip) exceeds 3m / s (corresponding to a rapid fall due to a sudden drop in the center of gravity), and the human body remains in a falling state in the subsequent 10 consecutive frames (such as the angle between the torso and the ground being less than 30 degrees), the system determines it as a suspected injury event. At this moment, the edge computing server immediately triggers a three-level response mechanism: First, it controls the venue's broadcast system to continuously broadcast a warning message: "Someone is injured in area A3 of the venue. Please keep your distance if you are not a person in the area." Simultaneously, it pushes an alert message containing the event location, time of fall, and a link to real-time video footage to the administrator's terminal, sends a rescue notification with navigation routes to the mobile terminals of pre-set medical personnel, and pushes event progress reminders to the terminals of users who have made reservations for the venue. At the same time, it keeps the access control equipment at the venue entrance in a constantly open state until the administrator reports that the rescue has been completed, and automatically restores normal access control. It can also be optionally configured to automatically dial 120 emergency services to treat the injured person as quickly as possible.

[0063] The cloud management platform in this application is based on a distributed cloud storage architecture and can use the Hadoop and Spark frameworks for data processing. It automatically integrates and analyzes all data within a preset time period using a preset collection frequency. Taking a venue utilization heatmap as an example, the system combines data such as reservation time slots, actual usage duration, and number of attendees for each venue with a GIS map to generate a heatmap distribution layer in hourly units. The red area represents peak periods with utilization exceeding 80%, and the blue area represents off-peak periods with utilization below 30%. Administrators can view the heatmap change trends for different dates and venue types through a decision support interface, and then adjust venue opening hours or formulate off-peak charging strategies. User behavior analysis profiles integrate data such as users' reservation preferences, exercise duration, activity selection, and consumption records to build user segmentation models with tags such as "high-frequency exercisers," "leisure experience users," and "team organizers." For example, if the system finds that the "high-frequency exercisers" group prefers basketball half-court courts from 7-9 pm on weekdays, it can push exclusive member discounts for that time period. For the "team organizers" group, the system automatically records information such as the number of people in their usual teams and the average exercise duration, and prioritizes recommending court combinations that meet their needs when they make a reservation. The facility health report integrates vibration and current data collected by sensors with regular manual inspection records, and uses a weighted scoring method to calculate the facility's health index (out of 100). When a basketball hoop's vibration sensor continuously detects impact vibrations exceeding 0.5g (corresponding to a severe collision) and the current monitoring module shows unstable power supply, the health index will drop below 70 points. The system automatically generates a maintenance work order, which includes the facility's location, details of abnormal data, recommended repair solutions, and spare parts inventory information. This work order is directly pushed to the terminal of the cooperating maintenance team. After the repair is completed, the engineer uploads the repair certificate via mobile terminal, and the system synchronously updates the facility health index to above 95 points, ensuring the traceability of the facility's status throughout its entire lifecycle.

[0064] In the above embodiments, by organically combining user terminals, sensing and execution device layers, edge computing servers, and cloud management platforms, a full-link intelligent management system covering "user interaction - data collection - local real-time processing - cloud-based in-depth analysis" is constructed. This system addresses pain points in traditional sports field management such as cumbersome reservations, delayed status perception, and insufficient data support. It achieves real-time response to local events through the deployment of edge computing and local AI models, avoiding response delays caused by cloud transmission latency. Simultaneously, it leverages the big data analysis capabilities of the cloud platform to provide data support for operational decisions. This combination satisfies both the immediate needs of on-site management and the decision-making needs of macro-operations, thereby comprehensively improving the management efficiency and service quality of integrated sports fields. It breaks down information silos in traditional management and realizes intelligent collaboration between "people, equipment, venues, and data" in sports scenarios.

[0065] In one embodiment, the specific events or states identified by the edge computing server may include athlete injury events.

[0066] The local artificial intelligence algorithm model includes a posture analysis model based on skeletal key point detection, which is used to identify action postures that conform to preset injury determination rules from the video stream.

[0067] The preset injury determination rules include detecting a rapid fall caused by a sudden drop in a person's center of gravity, and the fall posture lasting for more than a preset time threshold without returning to a standing or walking posture.

[0068] In this embodiment, when the edge computing server calls a local artificial intelligence algorithm model to identify specific events or states, including athlete injury events, this local artificial intelligence algorithm model can be a posture analysis model based on skeletal keypoint detection. This model first performs real-time frame extraction on the video stream transmitted from a high-definition network camera, processing 25 frames per second to ensure timely analysis. In the image preprocessing stage, the model enhances image contrast through adaptive histogram equalization and removes environmental noise using Gaussian blur, ensuring the accuracy of skeletal keypoint detection. Subsequently, the model employs a lightweight HRNet (High Resolution Network) architecture to locate 17 key nodes in the human body, including the head, neck, shoulders, elbows, wrists, hips, knees, and ankles, outputting the three-dimensional spatial coordinates (x, y, z) and confidence score for each node (the threshold is set to 0.7; nodes below this value will be re-detected).

[0069] To identify "rapid falling movements with a sudden drop in center of gravity," the model can calculate the vertical velocity of the hip node (an approximate location of the body's center of gravity) in real time: the instantaneous vertical velocity is obtained by dividing the change in the hip's z-coordinate across 5 consecutive frames by the time interval (0.2 seconds). When the absolute value of this velocity exceeds 3 m / s (corresponding to the typical velocity of a rapid fall from a standing position), an initial warning is triggered. Next, the model further analyzes the overall posture of the body: it calculates the angle between the torso (the line connecting the neck to the hip node) and the ground (by converting pixel coordinates to actual angles using camera calibration parameters). If this angle is less than 30 degrees across 10 consecutive frames, and there is no significant spatial movement (less than 5 cm) at the knee and ankle nodes, the model is judged as "continuous falling posture." At this point, the model combines the satisfaction of the above two conditions and outputs a judgment result of "suspected injury event." When the confidence level is higher than 0.85, the edge computing server's response mechanism is directly triggered.

[0070] It should be noted that the preset injury judgment rules in this application are not a fixed, single threshold standard, but can be dynamically optimized and adjusted according to the characteristics of the sport, the differences in the venue environment, and feedback from misjudgments in actual operation. For example, for highly competitive sports such as basketball and football, since athletes may engage in vigorous movements such as quick saves and diving to grab the ball, which are not injury-related, the system allows administrators to adjust the trigger threshold for hip vertical velocity to 3.5 m / s in the background, while widening the judgment threshold for the angle between the torso and the ground to 25 degrees, in order to reduce misjudgments caused by normal movement. For sports that rely heavily on agility, such as badminton and table tennis, where athletes often fall with obvious body imbalance, the hip vertical velocity threshold can be maintained at 3 m / s, while shortening the continuous judgment frame count for the falling posture to 8 frames, in order to more quickly identify real injuries.

[0071] Furthermore, the system supports self-learning optimization based on historical data: the edge computing server records video clips, sensor data, and manual review results (such as administrator confirmation of whether it is a real injury) for each judgment event, and periodically synchronizes this labeled data to the cloud management platform. The cloud platform updates the parameters of the posture analysis model through offline training. For example, when the system finds that the misjudgment rate of a certain type of venue (such as an indoor basketball court) is higher than 15%, it will automatically adjust the model threshold corresponding to that venue, or add new auxiliary judgment conditions, such as combining human respiratory rate data collected by millimeter-wave radar (if the respiratory rate is lower than 12 breaths / minute or higher than 30 breaths / minute, the confidence of injury judgment is increased), thereby continuously improving the accuracy and adaptability of the rules. This dynamic optimization mechanism ensures that the preset injury judgment rules can meet the actual needs of different scenarios, avoid missed judgments or misjudgments due to rigid rules, and further protect the safety of athletes and the reliable operation of the system.

[0072] Building upon this foundation, the posture analysis model in this application can also incorporate auxiliary data from millimeter-wave radar to address low-light scenarios: when the camera's light sensor detects an ambient brightness below 50 lux, the model automatically calls upon human motion trajectory data collected by the millimeter-wave radar and corrects the positions of key skeletal points using a Kalman filter algorithm, avoiding misjudgments caused by insufficient light. For example, in a nighttime soccer field scenario, millimeter-wave radar can accurately capture the trajectory of a person falling, cross-validating it with the skeletal analysis results of the visual model, thus improving the accuracy of injury event recognition to over 98%. The model also supports dynamic updates: the edge computing server periodically uploads local misjudgment cases (such as an athlete's normal fall to save the ball being misidentified as an injury) to the cloud management platform. The cloud then fine-tunes the model through transfer learning before distributing the optimized model parameters back to the edge, achieving continuous iteration of model performance.

[0073] In one embodiment, the process by which the edge computing server generates corresponding control commands and sends them to the controlled device, and / or generates early warning information, may include:

[0074] Trigger the site broadcasting system to play a preset warning message.

[0075] A notification containing the location information of the event is pushed to the terminals of the administrator, the pre-set medical staff, and the corresponding appointment personnel.

[0076] Keep the access control equipment at the site entrance open to allow rescue personnel to enter.

[0077] In this embodiment, when the edge computing server determines that an athlete is injured, the process of generating control commands and early warning information follows the principle of "tiered response and precise reach." First, the system immediately sends a command to the venue broadcasting system, triggering a preset warning voice to be broadcast in a loop, such as "Someone is injured in area A3 of the venue. Please keep your distance. Staff will be on site soon." The voice can be a clear male / female voice, and the volume can be set to 85 decibels to cover the entire venue. At the same time, the background music volume in the venue is automatically reduced to below 10 decibels to ensure that the warning information is clearly received by the people present.

[0078] Secondly, edge computing servers can push customized notifications to three types of terminals via the MQTT protocol: To administrator terminals (such as the venue operations manager's mobile app), a warning message containing the location of the incident (accurate to the venue number and area coordinates, such as "southwest corner of the 5-a-side football field"), a link to real-time video footage, and the time of the suspected injury; the notification appears as a red pop-up window accompanied by continuous vibration. To pre-set medical personnel terminals (such as the venue's resident doctor or a tablet device from a partner emergency station), a rescue notification with GIS navigation routes is pushed. The route is automatically planned as the shortest path from the medical station to the incident location, along with the venue's personnel density data (such as "there are currently 12 athletes on the venue") and the location of the nearest emergency equipment (such as "there is an AED defibrillator in the equipment cabinet on the east side of the venue"). To terminals of those who have made reservations for the venue, a progress reminder is pushed, stating, "A temporary incident has occurred at your reserved 5-a-side football field (14:00-16:00 session). Venue usage may be delayed. We will update the progress in real time. Thank you for your understanding," thus avoiding dissatisfaction among reservation users due to a lack of transparency.

[0079] Finally, the system can also send commands to the site entrance access control system to keep the dedicated rescue passage access control (such as the emergency door on the west side of the site) permanently open, while simultaneously locking external access to other entrance access control systems (such as the main entrance), allowing only internal personnel to leave and preventing unauthorized personnel from entering and interfering with the rescue. Upon receiving the command, the access control system immediately provides feedback via LED indicators: a solid green indicator above the emergency door indicates it is open, while a flashing red indicator at the main entrance indicates external entry is prohibited. The system also transmits access control status data back to the edge computing server in real time to ensure the commands are executed correctly. If other auxiliary passages need to be opened during the rescue, the administrator can manually send a secondary command via the terminal, and the system will respond immediately and adjust the access control status accordingly.

[0080] Through the aforementioned tiered response mechanism, the system can complete the entire process from event assessment to command issuance within 30 seconds, ensuring rapid mobilization of rescue resources and effective maintenance of on-site order. Furthermore, the system supports dynamic adjustment of the response level based on the severity of the event: if subsequent video frame analysis shows the injured person exhibiting limb convulsions or loss of consciousness (determined by irregular shaking of key skeletal points or lack of voluntary head movement), the edge computing server automatically escalates the response, triggering a fourth-level action—sending an automatic dialing request to the 120 emergency center containing the detailed address of the location, a description of the injured person's condition, and a real-time location sharing link, further shortening the emergency response time. This refined response design avoids excessive resource allocation while achieving "second-level response and efficient rescue" in emergencies, maximizing the safety of athletes.

[0081] In the goal event recognition module, the model can first adopt YOLOv8. In one embodiment, the specific events or states identified by the edge computing server include basketball-related events.

[0082] The local artificial intelligence algorithm model includes a basketball motion analysis model, used to identify goal events from video streams, distinguish goal types, and / or track player trajectories to collect individual motion data.

[0083] In this embodiment, when the edge computing server identifies a specific event or state as a basketball-related event through a local artificial intelligence algorithm model, the local artificial intelligence algorithm model can be a basketball sports analysis model optimized for basketball scenarios. This model integrates three core functional modules: goal event recognition, goal type differentiation, and player trajectory tracking. It achieves parallel processing of video stream data through a multi-task learning framework, and can complete full-function analysis of 15 frames of images per second, meeting the real-time requirements of basketball games or training scenarios.

[0084] The target detection algorithm performs real-time localization of the basketball, rim, and backboard in video frames, outputting their boundingbox coordinates and confidence scores (threshold set to 0.8). For the basketball's trajectory, the model generates its path through inter-frame target matching (using the Hungarian algorithm combined with IoU similarity). When the model detects that the basketball's center coordinates enter the inner area of ​​the rim (based on the rim's dimensions, such as a circular area with a diameter of 45cm), and in subsequent frames the basketball completely passes through the rim plane (verified by a depth sensor behind the backboard; when the basketball's z-axis coordinate exceeds the backboard plane by 20cm, it's considered "passing through"), a preliminary goal is determined. Simultaneously, the model performs secondary verification using the shooter's motion characteristics: the skeletal keypoint detection module locates the shooter's arm and wrist nodes. If an upward wrist rotation is detected within 0.5 seconds before the goal (the angle between the wrist and elbow nodes changes from 120° to over 180°), the goal confidence score is increased to over 0.95, avoiding misjudgments due to the basketball accidentally hitting the rim or bouncing.

[0085] The distinction between different types of baskets is based on a multi-dimensional assessment of the shooting position, action characteristics, and basketball trajectory before the basket is scored. For example, if the shooter's foot is outside the three-point line (the position of the three-point line is determined by the semantic segmentation of the court's landmarks), and the basketball's trajectory is parabolic (the curvature is calculated by fitting the basketball's path; if the curvature is greater than a preset threshold, it is considered a parabola), then it is classified as a "three-pointer." If the shooter's hand is below the shoulder node (the elbow node's y-coordinate is higher than the wrist node), and the basketball's trajectory is linear (the curvature is less than a preset threshold), and a collision between the basketball and the backboard is detected (the backboard vibration sensor collects a vibration signal greater than 0.1g), then it is classified as a "bank shot." For free throws, the model identifies whether there is only the shooter in the free throw line area (semantic segmentation result), and if the basketball's trajectory does not collide with the backboard, it is directly classified as a "free throw." In addition, the model also supports distinguishing between "dunk" types: when the key point of the shooter's hand is detected to be higher than the rim node (the y-coordinate of the hand is greater than the y-coordinate of the rim), and the trajectory of the basketball is vertically downward (the vertical speed is greater than 2m / s), it is marked as a "dunk" and the player ID of the dunker is automatically recorded.

[0086] The player trajectory tracking and individual motion data statistics module is implemented by combining ReID (Person Re-identification) technology with skeletal keypoint detection. The model first extracts features from each player, using a Transformer-based ReID model to generate an appearance feature vector (including jersey color, number, body type, etc.). This is combined with the posture feature vector of skeletal keypoints to achieve continuous player tracking across cameras and frames. Even with physical contact or brief occlusion, the tracking accuracy remains above 92%. Based on the tracked player trajectories, the model statistically analyzes individual motion data: "Running distance" is obtained by calculating the distance traveled by the player's skeletal keypoints (converted to actual meters using the court scale); "Ball holding time" is calculated by detecting player interactions with the basketball (e.g., the distance between the hand keypoint and the basketball is less than 30cm and lasts for more than 1 second); and core data such as "Number of shots," "Number of baskets," and "Shooting percentage" are calculated by associating the shot ID with the goal event recognition results. For example, if the system finds that a player with the ID "P012" ran a distance of 4.2 kilometers and took 18 shots in a game, including 5 three-pointers, 3 layups, and 4 free throws, the system will automatically generate the player's "scoring details" and "movement heat map". The heat map uses the court as a background and marks the player's activity area with different colors (red represents areas where the player stayed for more than 5 minutes, and blue represents areas where the player stayed for less than 1 minute), which intuitively shows the player's on-court position preference.

[0087] It's worth noting that this basketball analysis model supports real-time interaction with user terminals: when a user selects "View Real-Time Data" in the mobile app, the model can synchronously push statistical personal data, goal types, and player trajectories to the terminal via the WebSocket protocol, with a latency of less than 1 second. For training scenarios, the model can also generate "action optimization suggestions." For example, if it detects that a player's elbow is lower than their shoulder when shooting a three-pointer (which can easily lead to insufficient shooting power), the system automatically extracts a video clip of the shooting motion, marks the key skeletal points of the incorrect motion on the clip, and pushes it to the "Training Guidance" module on the coach's terminal to assist the coach in targeted instruction. Furthermore, the model's parameters can be adjusted according to scenario requirements. For instance, in amateur game scenarios, the position threshold for three-point shot judgment can be lowered (expanding the judgment range outside the three-point line by 5cm) to accommodate the shooting habits of amateur players; in professional training scenarios, the rigor of motion verification can be increased (e.g., raising the wrist rotation angle threshold to 190°) to ensure the accuracy of data statistics.

[0088] In one embodiment, the specific events or states identified by the edge computing server include football-related events.

[0089] The sensing and actuation device layer includes a goal line detection device deployed on the football goal line.

[0090] The local artificial intelligence algorithm model includes a football motion analysis model, used to process the data from the goal line detection device to determine whether a goal is valid, and / or to track player trajectories from the video stream to collect running and tactical data.

[0091] In this embodiment, when the edge computing server identifies football-related events through a local artificial intelligence algorithm model, this model is a football analysis model customized for football scenarios. It can work with the goal line detection device to determine the validity of a goal and combine video streams to track player trajectories and perform tactical data statistics, providing accurate support for match decisions and training analysis.

[0092] The goal-line detection device in this application can adopt a dual-mode fusion design of "optical sensor + pressure sensor". For example, eight sets of optical sensors (each set includes an infrared transmitter and receiver) are deployed equidistantly on the inside of the goal line. When the football completely blocks the optical path of any set of sensors (for a duration of ≥20ms), a preliminary goal-line judgment is triggered. At the same time, four pressure sensing strips (covering the entire goal line) are embedded at the bottom of the goal. When the pressure generated by the football on the sensing strips exceeds 5N, a pressure trigger signal is output. After receiving the dual-mode data, the football motion analysis model can confirm the validity of the goal through logical AND operation. For example, if the optical sensor and the pressure sensor both trigger signals within 100ms, and the video stream detects the entire football crossing the goal line in three consecutive frames (by extracting the football's outline through semantic segmentation and calculating the relative position of the outline's center to the goal line), it is determined to be a "valid goal." The system immediately pushes a judgment assistance message to the referee's terminal, including the goal time, the goal scorer's ID, and a visualization of goal line data (such as an optical occlusion timeline diagram and a pressure distribution heatmap). If only a single sensor triggers, the model marks it as a "suspected goal," automatically extracting a 5-second video clip (including multi-angle camera footage) before and after the event for the referee to review, avoiding misjudgments due to a single sensor malfunction. For instance, in an attack, the football grazed the edge of the goal line, the optical sensor did not completely block it, but the pressure sensor detected 2.3N of pressure. The model determined it to be a "suspected goal" and simultaneously pushed videos from the linesman's perspective and the goal's top perspective. The referee confirmed through slow-motion replay that the football did not completely cross the goal line, ultimately ruling the goal invalid.

[0093] The player trajectory tracking and tactical data statistics module can be implemented based on multi-camera collaboration and ReID technology. Multiple 4K high-definition cameras can be deployed around the field (covering the entire field without blind spots), and the edge computing server stitches the video streams from each camera in real time to generate a panoramic view of the field. The model can use a CNN-based ReID algorithm to extract player features (including jersey numbers, team logos, hairstyles, etc.), combined with the posture features of skeletal key points, to achieve continuous player identification tracking across cameras. Even in the midst of dense player-dominated penalty area battles, the tracking accuracy can still reach over 90%. Based on trajectory data, the model statistically analyzes individual running data: by calculating the difference in position coordinates of players in each frame, the "total running distance" is accumulated, and the running type is divided according to speed thresholds (≥7m / s for sprinting, 3-7m / s for jogging, <3m / s for walking). For example, player ID "F007" ran a total distance of 11.2 kilometers in a game, including 1.8 kilometers of sprinting and 6.5 kilometers of jogging. At the same time, the model analyzes the player's "activity area heat map", using red, yellow, and blue to mark areas where the player stays for more than 10 minutes, 5-10 minutes, and less than 5 minutes, respectively, to intuitively show the player's on-field position preference. For example, the heat map of midfielder "F015" is concentrated near the center circle, indicating that he is mainly responsible for organization and scheduling.

[0094] In terms of tactical data statistics, the model also supports functions such as "passing path analysis" and "formation maintenance calculation". Passing path analysis detects the interaction between the ball and the player (e.g., the distance between the foot's key point and the ball is <20cm and accompanied by a kicking motion), records information such as the initiator, receiver, passing distance, and passing speed, and generates a visual passing network diagram (using lines of different thicknesses to represent passing frequency, with red lines representing successful passes and gray lines representing failed passes). For example, a team's passing network diagram shows that the frequency of crosses from the left wing is 35%, but the success rate is only 22%, allowing the coach to adjust tactics accordingly. Formation maintenance calculation extracts the position coordinates of all team players in real time and compares them with the standard positions of preset formations (e.g., 4-3-3, 5-4-1), calculating the root mean square of the positional deviation. A deviation value <0.5 indicates "good formation maintenance," 0.5-1.0 indicates "slight disjointedness," and >1.0 indicates "disorganized formation." The model updates the formation score every 30 seconds and pushes warnings to the coach's terminal, such as "In the 25th minute, formation maintenance is 0.8, the distance between the right back and the midfielder exceeds 15 meters, indicating a defensive vulnerability."

[0095] Furthermore, the model supports integration with the VAR (Video Assistant Referee) system: when a referee initiates a VAR review request, the edge computing server immediately retrieves relevant goal-line data, player trajectories, and video clips, transmitting them to the VAR terminal via a dedicated channel with a transmission latency of ≤2 seconds. For training scenarios, the model can generate "energy allocation suggestions." For example, based on player running distance and heart rate data (collected via a heart rate sensor built into the smart jersey), when a player's heart rate consistently exceeds 180 beats per minute and sprinting accounts for more than 20% of their playing time, a reminder to "reduce training intensity and replenish fluids" is pushed to the coach's terminal. Model parameters can be adjusted according to the scenario. For instance, in youth matches, the sprinting speed threshold can be lowered to 6 m / s to suit the physical characteristics of teenagers; in professional leagues, the accuracy requirements for goal-line judgments can be increased by shortening the optical sensor occlusion duration threshold to 15 ms to ensure the rigor of the rulings.

[0096] In one embodiment, the specific events or states identified by the edge computing server include tennis-related events.

[0097] The local artificial intelligence algorithm model includes a tennis motion analysis model, which is used to reconstruct the three-dimensional trajectory and calculate the landing point of the tennis ball based on multi-angle video streams, in order to determine out-of-bounds events and / or statistically analyze serve speed and shot type data.

[0098] In this embodiment, when the edge computing server identifies tennis-related events through a local artificial intelligence algorithm model, the model is a tennis motion analysis model optimized for tennis scenarios. It integrates four core functions: three-dimensional trajectory reconstruction, landing point determination, serve speed statistics, and shot type recognition. Through multi-camera synchronous acquisition and spatiotemporal fusion algorithms, it achieves high-precision analysis of tennis motion data, meeting the dual needs of match judging and training improvement.

[0099] The 3D trajectory reconstruction and landing point calculation module relies on six high-speed cameras (240fps, 1080p resolution) deployed around the court. Each camera is pre-calibrated (using the Zhang Zhengyou calibration method to obtain intrinsic and extrinsic parameters). The model first performs tennis ball target detection on the video frames from each camera (e.g., using the YOLOv8-tiny algorithm to optimize the anchor frame size for small tennis balls), outputting the two-dimensional coordinates of the tennis ball in each frame. Then, it calculates the three-dimensional spatial coordinates of the tennis ball using triangulation (utilizing the parallax information from multiple cameras and combining the positional relationship between the cameras), and uses Kalman filtering to smooth the three-dimensional coordinates of consecutive frames, generating the tennis ball's trajectory curve. Landing point determination is based on the positional relationship between the endpoint of the trajectory curve and the court markings: the model extracts the court markings such as sidelines, baselines, and service lines using semantic segmentation algorithms (e.g., using the DeepLabv3+ model to optimize court color and texture features). When the x / y axis of the tennis ball's three-dimensional coordinate endpoint is outside the markings (z-axis coordinate ≤ 0, i.e., in contact with the ground), and the vertical velocity component of the trajectory curve changes from positive to negative (determined as the moment of landing), it is determined to be "out of bounds"; if the endpoint is inside the markings, it is determined to be "in bounds". To avoid errors caused by camera angle deviations, the model introduces calibration marks at the corners of the court (such as black dots with a diameter of 10cm) for real-time calibration, updating the camera's extrinsic parameters every 5 minutes to ensure that the landing point determination error is ≤ 3cm. For example, in a certain serve, the model reconstructs the tennis ball trajectory, and the x-coordinate of the end point is 2.31m (2.1m beyond the singles sideline) and the z-coordinate is -0.02m (already landed). It is immediately judged as "serve out of bounds" and pushes a visual information including the landing point coordinates, trajectory curve and court line overlay to the referee's terminal. The referee can make a quick ruling without relying on the linesman.

[0100] The serve speed statistics module calculates based on the trajectory reconstruction results. The model can extract the three-dimensional coordinate sequence of the tennis ball in the air above the service area (the area between the service line and the net), calculate the distance difference between two consecutive frames, and combine it with the frame rate (240fps, i.e., each frame interval is about 4.17ms) to obtain the instantaneous speed. The maximum value is taken as the "serve speed". For example, in a serve by player ID "T011", the model detected that the tennis ball moved 0.35m between frames 120 and 121, and calculated the instantaneous speed to be 84m / s (about 302km / h). The system automatically records this data and synchronizes it to the player's training file. The shot type recognition is achieved through multi-dimensional judgment by analyzing the player's action characteristics and tennis ball trajectory characteristics at the moment of impact: The model uses the MediaPipe skeleton detection algorithm to locate the player's arm, wrist, and waist nodes. At the moment of impact (when the tennis ball contacts the racket, the racket's vibration sensor assists in the judgment, vibration signal ≥0.2g), if the player's waist node rotates in the direction of impact (waist yaw angle change ≥30°), and the horizontal velocity component of the tennis ball trajectory is greater than the vertical velocity component (e.g., ratio >2), it is judged as a "flat shot"; if the bending angle of the wrist node at the moment of impact is ≤90° (the angle between the wrist and elbow), and the vertical velocity component of the tennis ball trajectory is greater than the horizontal velocity component (ratio >1.5), and the curvature of the trajectory curve is greater than a preset threshold (judged as a high parabola), it is judged as a "topspin shot"; for the "slice shot" type, if the model detects that the wrist node's rotation direction is downward at the moment of impact (the wrist node's rotation angle around the elbow is negative), and the horizontal velocity component of the tennis ball trajectory has a lateral component (absolute value of x-axis velocity ≥10m / s), it is marked as a "slice shot". For example, in a certain shot by player ID "T018", the model detected that his waist rotation angle was 45°, the horizontal speed of the tennis ball was 25m / s and the vertical speed was 10m / s, and judged it as a "flat shot". The model also recorded the landing position and speed of the shot, providing data support for coaches to analyze the technical characteristics of players.

[0101] In addition, the tennis analysis model also supports interactive functions with the player's terminal: when a player selects "view training data" in the mobile app, the model can push the distribution of serve speed (such as average speed, fastest speed, speed standard deviation), the percentage of hit types (such as flat hits 40%, topspin hits 50%, and slice hits 10%), and a heat map of the landing point (the red area is the high-frequency landing point) for the player's most recent 10 training sessions. For technical problems in training, the model can generate optimization suggestions. For example, when it is detected that the player's serve landing point is frequently concentrated near the baseline (out-of-bounds rate of 30%), the system automatically extracts the serve action video, marks the height of the ball toss (calculated by the maximum y-coordinate of the tossing hand through skeletal nodes) and the timing of the hit (the height at which the tennis ball contacts the racket), and prompts "The ball toss height is too high (average 2.1m, it is recommended to adjust to 1.8m), which causes a delay in the timing of the hit and makes the landing point easy to go out of bounds", to help the player make targeted improvements. The model parameters can be adjusted according to the scenario. For example, in amateur competitions, the error tolerance for landing point judgment can be increased to 5cm to adapt to the marking accuracy of non-professional venues; in professional competitions, the frame rate for serving speed statistics can be increased to 360fps to ensure that the accuracy of speed measurement is ≤0.5m / s.

[0102] Through the aforementioned multi-scenario adaptable algorithm design, the integrated sports field intelligent management system can achieve accurate data collection and analysis for different sports such as football and tennis. In practical applications, the system can automatically switch the corresponding analysis model according to the type of sport, eliminating the need for manual reconfiguration of hardware equipment and significantly improving the utilization efficiency and management convenience of the field. For example, when the integrated sports field switches from football training to tennis matches, the edge computing server can automatically retrieve the tennis motion analysis model, simultaneously activate the calibration process of the high-speed camera, and connect to the sensing equipment of the tennis court. The entire switching process takes no more than 1 minute. When the field is used for youth football introductory training, the system automatically reduces the strictness of action judgment, such as expanding the positional error tolerance for offside judgment to 10cm, while simplifying the presentation of tactical data by displaying the running range in the form of cartoonish heat maps and animated trajectories, which is more in line with the cognitive characteristics of teenagers.

[0103] In one embodiment, the specific events or states identified by the edge computing server include competition violation events.

[0104] The local artificial intelligence algorithm model includes a violation action recognition model, which is used to identify preset types of foul actions or violations in basketball or football games from video streams.

[0105] In response to the competition violation, the edge computing server sends auxiliary judgment information, including the event time and video clip, to the designated referee terminal.

[0106] In this embodiment, when the edge computing server identifies violations in the game through a local artificial intelligence algorithm model, the model is a violation action recognition model optimized for basketball and football scenarios. It integrates a multimodal feature extraction and rule matching engine, which can detect a variety of preset fouls and violations in real time, covering common scenarios such as physical contact fouls and technical violations, helping referees to make accurate judgments quickly.

[0107] Taking a football match as an example, the model's identification logic for the "pulling jersey" foul is as follows: First, the player's jersey area (marked as red or blue pixel blocks) is extracted through semantic segmentation, and key points of the player's arm (such as the wrist and elbow) are located using MediaPipe skeletal detection. When the pixel distance between player A's wrist key point and player B's jersey area is less than 15cm, accompanied by stretching and deformation of the jersey area (the pixel displacement of the jersey edge is calculated to be ≥5cm using optical flow), and player B's body posture shows an involuntary shift (such as the acceleration vector of the skeletal node being opposite to the direction of movement), it is initially determined to be a suspected "pulling jersey". Subsequently, the rule matching engine retrieves the clauses in the football competition rules regarding pulling fouls (such as "intentionally pulling an opponent's player to obstruct their movement"), and performs secondary verification in combination with the scene of the event (such as whether it is in an advantageous attacking position). If all conditions are met, it is marked as a "confirmed foul", and the time of the event (accurate to 0.1 seconds), the ID of the player involved, a slow-motion video clip of the foul action (including close-up frames of jersey stretching), and a description of the rule clause are pushed to the referee's terminal. For example, during a breakthrough on the wing in the 65th minute, the model detected that the distance between player "F023's" wrist and player "F031's" blue jersey was only 8cm, the pixel displacement of the right edge of the jersey reached 7cm, and "F031's" running speed dropped sharply from 5m / s to 2m / s. The system immediately judged it as a "jail-pulling foul". After reviewing the close-up video on the terminal, the referee quickly made the decision to award a free kick.

[0108] For identifying "traveling violations" in basketball games, the model can adopt a two-dimensional judgment logic of "foot sequence analysis + rule matching": First, the complete foot trajectory of the player is obtained by stitching together videos from multiple cameras, and the landing frame and take-off frame of each step are marked; then, based on the basketball rule that "when carrying the ball, only one foot can be the pivot foot, and the other foot can move", the model analyzes the player's foot movements after receiving the ball: when it is detected that after the player receives the ball (judged by the distance between the key point of the hand and the basketball is <10cm), if the pivot foot (the foot that lands first) has not left the ground, and the other foot lands ≥2 times consecutively, or if the pivot foot leaves the ground but does not complete the shooting / passing action before landing again, it is judged as a "traveling violation". For example, if player "B009" lands with his left foot as the pivot foot after receiving the ball at the top of the key, then lands with his right foot and lifts his left foot, and lands with his left foot again without shooting or passing, the model immediately marks it as a "traveling violation" and pushes auxiliary information including a sequence diagram of foot landings (timestamps and position coordinates of each step), so that the referee can confirm the violation without repeatedly replaying the footage.

[0109] For identifying "technical fouls," the model focuses on analyzing non-contact violations by players, such as "insulting the referee" and "deliberately wasting time." Taking "deliberately wasting time" as an example, the model detects player actions after a game interruption: when the referee blows the whistle to signal a pause (by extracting whistle characteristics through audio recognition), if a player exhibits behaviors such as "standing still with the ball for more than 10 seconds without passing it to the referee" or "slowly walking towards the inbounds point (speed < 1 m / s and deviation from the shortest path ≥ 2 m)," combined with the time difference on the court timer (the pause duration exceeding the rule-allowed 30 seconds), it is judged as a "deliberately wasting time" technical foul. For example, in the last two minutes of a basketball game, if player "B017" stands still with the ball for 12 seconds without shooting during inbounds, the model simultaneously detects that the timer shows the pause duration has reached 35 seconds, immediately sends the event information to the referee, who then issues a technical foul and awards the opposing team free throws.

[0110] Furthermore, the violation detection model supports a "custom rule base" function, allowing parameter adjustments based on the specific rules of different leagues. For example, in youth basketball leagues, the model can reduce the strictness of the "three-second rule" (extending the threshold for time spent in the restricted area from 3 seconds to 5 seconds); in professional football leagues, the model can add a dimension for "diving" detection, comprehensively determining whether a player has dunked by analyzing the intensity of physical contact before falling (impact force detected by pressure sensors <10N), the posture after falling (e.g., unsupported arm movement and time of complete fall <0.5 seconds), and facial features (exaggerated pain expressions detected through facial recognition), thus further improving the fairness of the ruling. The model also supports integration with the VAR system. When a referee requests a review of a violation, the system can automatically retrieve multi-angle video clips (including different perspectives of the foul) from 10 seconds before and after the incident, quantitative data on the movement characteristics (e.g., pulling force, number of steps), and generate a "violation probability score" (e.g., 95% probability of a pulling foul) to assist the referee in making a final decision.

[0111] In one embodiment, the system supports a video playback challenge mechanism.

[0112] The user terminal is configured to receive penalty challenge requests initiated by users.

[0113] The edge computing server or the cloud management platform is configured to automatically retrieve multi-angle video from the relevant time period and generate replay clips in response to the penalty challenge request, for the referee's terminal to retrieve and view.

[0114] In this embodiment, the system also supports a video replay challenge mechanism. The core process of this mechanism is as follows: when a player or coach disagrees with a ruling, they can initiate a challenge request through a bound user terminal (such as a tablet on the coach's bench or a player's smart bracelet). The system first verifies the validity of the request, for example, if it is submitted within 30 seconds of the ruling, and if the number of challenges per team in that match does not exceed the rule limit (e.g., two challenges per team in professional matches). After successful verification, the edge computing server immediately retrieves 15 seconds of multi-angle video before and after the ruling (covering high-speed cameras around the field and close-up camera footage near the goal / basket), stitches the multi-view videos into a 360° panoramic replay segment using a spatiotemporal fusion algorithm, and automatically marks keyframes: such as the moment of contact when pulling a jersey in a football match, or the landing frame of a foot in a traveling violation in a basketball match. At the same time, the system will generate a "challenge analysis report", which includes the timestamp of the event, information of the players involved, algorithm data on which the original ruling was based (such as trajectory coordinates and action feature quantification values), and time node annotations of the replay segment (such as "00:02 seconds is the moment of physical contact").

[0115] For example, in a professional tennis match, player "T025" challenged the referee's decision that the serve was out of bounds. The system verified the challenge was valid within 20 seconds (the team had one challenge opportunity remaining). It then retrieved footage from six high-speed cameras at the moment of the serve, stitching together a complete trajectory of the tennis ball from its toss to its landing. A comparison of the 3D trajectory curve and the court markings was overlaid on the footage. The original model determined the landing point's x-coordinate to be 2.31m (outside the sideline), but the replay, in slow motion (1 / 10 frame rate), showed the actual landing x-coordinate to be 2.29m (the sideline position was 2.30m, an error of 0.01m). The system immediately updated the landing point determination to "in bounds" and pushed the corrected trajectory data and the replay footage to the referee's terminal. After reviewing the footage, the referee immediately changed the decision to "serve valid." The entire challenge process took less than 2 minutes, avoiding the impact of human error on the fairness of the match.

[0116] In addition, the system supports a "challenge result feedback" function: if the challenge is successful, the system will automatically adjust the parameters of the corresponding algorithm model (e.g., temporarily reducing the error threshold for judging the tennis ball's landing point to 2cm); if the challenge fails, one challenge opportunity will be deducted from the team, and the reason for the failure will be pushed to the user's terminal (e.g., "The replay shows that the landing point was indeed 0.05m beyond the sideline, which is consistent with the original ruling"). For major events, the cloud management platform will archive video clips, algorithm data, and ruling results of all challenge events into the event database for post-event technical review and rule optimization reference. For example, in a certain season of football league, the success rate of "diving" challenges reached 62%. Based on the diving case data archived by the system, the event organizing committee optimized the "diving" judgment dimension in the illegal action recognition model, adding "the rate of change of acceleration before falling" (e.g., if the acceleration suddenly drops from 5m / s² to -10m / s², it is judged as a suspected dive), further improving the accuracy of the ruling.

[0117] In one embodiment, the specific events or states identified by the edge computing server include an over-limit situation of personnel on site.

[0118] The local artificial intelligence algorithm model includes a real-time people counting model based on target detection and tracking, which is used to identify from the video stream that the current number of people exceeds the preset capacity limit of the corresponding venue.

[0119] In response to the overcrowding status of the venue, the edge computing server sends a locking command to the entrance access control device and / or sends a full-occupancy message to the venue display screen and pushes an overcrowding alarm to the administrator terminal.

[0120] In this embodiment, the real-time people counting model based on object detection and tracking can adopt the YOLOv8 object detection algorithm combined with the DeepSORT multi-object tracking technology to accurately identify and count the people in the venue. First, the model can extract human features (such as human body contours and head key points) from video stream frames, and use the high frame rate detection ability of YOLOv8 (processing more than 30 frames per second) to quickly locate each person present; then, the DeepSORT algorithm is used to track the trajectories of the detected people, and a unique ID is assigned to each person to avoid double counting or missing counting caused by people walking and occlusion. For example, when the preset capacity of a badminton court in a comprehensive sports ground is 6 people (at most 2 pairs of doubles per court), the model will count the number of people entering the court in real time: if the 7th person is detected entering the court (confirmed by the linkage of the personnel identification of the access control system and the tracking data of the cameras in the court), it is determined as the "venue overcrowding status".

[0121] After determining the overcrowding, the edge computing server will immediately trigger a multi-terminal linkage response: on the one hand, it will send a locking instruction to the access control device at the entrance of the venue, temporarily prohibiting new people from entering the venue, and display a prompt on the access control display screen saying "The current venue is full. Please go to other venues"; on the other hand, it will push a full venue prompt message to the LED display screen in the venue to remind the people present of the venue capacity in the form of a red scrolling subtitle; at the same time, it will push an overcrowding alarm to the administrator terminal (such as the computer in the venue management center and the administrator's mobile App), including the overcrowded venue number, the current number of people (such as 7 people), the overcrowding time (accurate to seconds), and a screenshot of the real-time video image in the venue, so as to facilitate the administrator to go to the scene for guidance in a timely manner. For example, the preset capacity of a tennis court in a comprehensive sports ground is 4 people (singles or doubles). After the model detects that 5 people have entered the court, the system completes the determination and triggers the response within 1 second: the access control device automatically locks, the venue display screen shows a full venue prompt, and after the administrator terminal receives the alarm, the administrator views the real-time image of the venue through the App and finds that a spectator has strayed into the training venue, and then goes to guide him to leave. The whole processing process takes less than 5 minutes, effectively avoiding the safety hazards caused by venue congestion.

[0122] Furthermore, the real-time crowd counting model also supports dynamic capacity adjustment. For example, when a large-scale event (such as a school sports meet) is held at the sports stadium, the administrator can temporarily adjust the stadium's capacity limit through the cloud management platform (e.g., increasing the capacity of a football field from 22 to 50 people to accommodate the gathering needs of the opening ceremony). The model will automatically synchronize the new capacity parameters and determine if the limit is exceeded based on the adjusted threshold. Simultaneously, the model can distinguish between "participants" and "temporary visitors" within the stadium, such as by identifying their movement characteristics (whether they are carrying sports equipment or engaging in continuous physical activity). If two people merely stand at the sideline for more than 10 minutes without participating in any activities, they will not be counted as "participants," avoiding misjudgments of exceeding the limit due to temporary spectators. Through this precise crowd counting and intelligent response mechanism, the intelligent management system for sports stadiums can effectively ensure the safety and order of stadium use and improve the precision of stadium management.

[0123] In one embodiment, the sensing and actuation device layer includes a light intensity sensor and a controllable lighting device.

[0124] The edge computing server is configured as follows:

[0125] The ambient light level is determined based on the data collected by the light intensity sensor.

[0126] Based on the ambient light level and the current usage status of the site, control commands are generated to adjust the brightness or on / off status of the controllable lighting equipment.

[0127] In this embodiment, the illuminance sensor can collect illuminance data of different areas within the venue in real time (e.g., every 5 seconds, with an accuracy of 1 lux) and transmit the data to the edge computing server. The server's built-in illuminance level determination model will classify the ambient illuminance level based on the collected data and the illuminance requirement thresholds for different sports: for example, the minimum illuminance requirement for a football match is 500 lux (training scenario) and 1000 lux (official match scenario), while the minimum illuminance requirement for a basketball match is 750 lux (training) and 1500 lux (official match). When the illuminance sensor detects that the illuminance of the football training field drops to 420 lux, the model determines it as "insufficient illuminance"; if it detects that the illuminance of the basketball official match field reaches 1600 lux, it determines it as "suitable illuminance".

[0128] After determining the light level, the edge computing server generates lighting control instructions based on the current usage status of the venue (such as whether it is in a match, training, or idle state, confirmed through the linkage of reservation system data and personnel activity detection by cameras). For example, when the outdoor soccer field of the multi-purpose sports stadium is in "official match" status, and the average light intensity collected by the light sensor is 850 lux (below the match requirement threshold of 1000 lux), the server will send an instruction to the controllable lighting equipment (such as the LED floodlights around the field) to "gradually increase the brightness to 1000 lux". The floodlights will adjust at a rate of 100 lux every 10 seconds to avoid sudden changes in light affecting the players' vision. If it is halftime (confirmed through the reservation system time and the resting posture of the personnel on the field), the server will reduce the light intensity to the training level of 500 lux to save energy.

[0129] For indoor venues (such as badminton courts), the system also supports a "zoned lighting adjustment" function. Illuminance sensors are deployed in zones (e.g., one sensor at each of the four corners of each badminton court). When players are training on a particular court (detected by cameras through racket swings and player movements), and adjacent courts are idle, the server will only increase the lighting brightness of that training court to 500 lux (the requirement for badminton training), while the idle courts maintain basic lighting of 200 lux, avoiding energy waste caused by overall lighting. For example, if four players are practicing doubles on court 3 in an indoor badminton court, and the illuminance sensor detects an illuminance of 380 lux in that area, the server will immediately send a command to the lighting equipment on court 3 to "increase brightness to 500 lux," while the lighting in idle courts 1, 2, and 4 remains unchanged, saving approximately 20% of lighting energy consumption per hour.

[0130] In addition, the system also possesses "lighting adaptive learning" capabilities. The edge computing server records the patterns of light changes under different weather conditions (such as sunny, cloudy, and rainy days), and combines this with historical usage data (such as venue reservation rates and types of sports during a certain time period) to adjust lighting strategies in advance. For example, based on historical data, the reservation rate for the football field reaches 90% every Saturday afternoon from 2 pm to 4 pm, and the probability of cloudy weather during this time period is 60%. The system will automatically adjust the football field's lighting equipment to "prepared boost mode" at 1:50 pm on Saturday. Five minutes before the light intensity drops to the threshold, it will initiate brightness boost in advance to ensure that the venue's lighting always meets the sports needs. At the same time, if the light sensor malfunctions (such as data continuously being 0 or exceeding the normal range), the server will use image brightness analysis from the camera (such as comparing the grayscale values ​​of the image with a preset light-grayscale correspondence table) to assist in the judgment and push sensor fault alarms to the administrator terminal to avoid hardware problems affecting the accuracy of lighting adjustments.

[0131] In one embodiment, the specific events or states identified by the edge computing server include abnormal conditions of site facilities.

[0132] The sensing data includes images or sensor data that reflect the condition of the site surface.

[0133] The local artificial intelligence algorithm model or analysis program is used to identify at least one abnormal state from the sensor data, including ground cracks, potholes, fading, or lawn health below a preset health threshold.

[0134] In response to the abnormal status of the site facilities, the cloud management platform automatically generates a maintenance work order that includes the location and type of the abnormality.

[0135] In this embodiment, sensor data reflecting the condition of the field surface can be collected through two types of devices: First, high-definition inspection cameras (4K resolution, 25 frames per second) installed around the field perform a panoramic scan of the field surface daily (e.g., when the field is idle at 3 AM), generating an image sequence covering the entire field; second, a pressure sensor array embedded in the field surface (one sensor per square meter, sampling frequency 10Hz) collects ground force data in real time to help identify structural anomalies such as potholes. Taking an outdoor soccer field as an example, the high-definition inspection cameras will take pictures area by area along the path "from the sideline to the center line, from the goal area to the corner kick area," generating a close-up image block every 10 centimeters to ensure no blind spots; the pressure sensor array captures changes in the pressure distribution of the ground during player training or matches—if the pressure data in a certain area is consistently lower than the surrounding area (e.g., the ground force is dispersed in potholes), it is marked as a suspected abnormal area.

[0136] Local artificial intelligence algorithm models can use "image semantic segmentation + multimodal data fusion" technology to identify facility anomalies. For cracks in the field, the model first uses the U-Net semantic segmentation algorithm to perform pixel-level annotation on the inspection images, extracting features such as the length, width, and direction of the cracks: if the crack length exceeds 5 cm and the width exceeds 2 mm, it is judged as "cracks that need repair"; for potholes, the model combines the grayscale difference of the inspection images (potholes show a deeper grayscale value) with the force data from the pressure sensor (the average pressure in the pothole area is 30% lower than that in the normal area) to comprehensively determine the depth and area of ​​the potholes—if the depth exceeds 3 cm and the area exceeds 0.1 square meters, it is marked as "severe potholes"; for lawn health, the model extracts the color features of the lawn (e.g., if the green component ratio is less than 60%, it is judged as fading) and texture features (e.g., if the grass blade density is less than 50 plants / square centimeter, it is judged as sparse) from the inspection images, and combines the data from the soil moisture sensor (humidity less than 15% affects lawn growth) to calculate the lawn health score—if the score is less than 70 points (out of 100), it is judged as "unhealthy".

[0137] Once the model identifies an abnormal condition, the cloud management platform automatically triggers the maintenance work order generation process. The work order includes: a unique identifier for the abnormal site (e.g., "Football Field 03"), the precise coordinates of the abnormal location (based on the site's GIS map, e.g., "10 meters to the left of the center line, 5 meters from the sideline"), the type and severity of the abnormality (e.g., "Severe pothole, 4 cm deep, 0.15 square meters in area"), visual evidence of the abnormality (including close-up images of cracks and 3D depth reconstruction images of the pothole), and a suggested repair plan (e.g., "The pothole needs to be filled with asphalt mixture and compacted; a pressure test is required after repair"). After the work order is generated, the system automatically assigns it to the appropriate personnel based on their skill tags (e.g., "Proficient in lawn maintenance" or "Expert in site hardening repair") and notifies them to accept the order via SMS, app push notifications, etc. For example, after an inspection of basketball court 02 in a multi-purpose sports stadium, the model identified "two 8-centimeter-long cracks in the three-point line area." The cloud platform immediately generated a work order and assigned it to Mr. Li, a maintenance worker specializing in hardening and repairing sports fields. After viewing the abnormal image through the App, Mr. Li took the crack filler to the site for repair. After the repair was completed, he uploaded the repaired image and pressure test data. The system automatically re-inspected the repair effect. If the gray value of the crack filler area was consistent with the surrounding ground and the pressure data returned to normal, the work order was marked as "completed." If the re-inspection failed, a second repair reminder was sent.

[0138] Furthermore, the system supports trend prediction for abnormal conditions. The cloud management platform stores historical abnormal data and uses time series analysis models to predict the development trend of abnormalities. For example, if the health score of the turf on football field 03 drops from 72 points last week to 68 points this week, and the soil moisture remains below 15%, the model predicts that if no measures are taken, the health score will drop below 60 points next week. The system will then generate a "preventive maintenance work order" in advance, recommending that maintenance personnel water and fertilize the field promptly. Through this closed-loop management of "real-time detection - automatic work order dispatch - effect review - trend prediction," the integrated sports field intelligent management system can shorten the fault repair time of field facilities to within 24 hours, effectively extending the service life of facilities and ensuring the safety of athletes.

[0139] In one embodiment, the comprehensive management reports generated by the cloud management platform include at least one of the following: site utilization heat map, user behavior analysis profile, facility health report, and revenue analysis report.

[0140] The decision support interface provides operational optimization suggestions based on the comprehensive management reports.

[0141] In this embodiment, the comprehensive management reports generated by the cloud management platform include at least one of the following: site utilization heat map, user behavior analysis profile, facility health report, and revenue analysis report.

[0142] The heatmap for venue usage uses a clear color gradient to show the frequency of use of different venues at different times. For example, the 12 badminton courts, 3 tennis courts, and 2 football fields in the multi-purpose sports stadium are divided into independent areas, with hourly data representing usage rates below 30%, yellow representing 30%-70%, and red representing above 70%. The system automatically summarizes usage data from the past 7 days to generate a weekly heatmap. If it is found that from 9:00 AM to 11:00 AM every Monday to Friday, badminton courts 1-4 are consistently red (usage rate above 90%), while courts 5-8 are blue (usage rate below 20%), then "peak area" and "idle area" are marked on the heatmap to help administrators adjust venue opening strategies.

[0143] User behavior analysis profiles are generated based on user reservation records, exercise duration, and activity preferences. For example, the system uses facial recognition data from the access control system to link user IDs, identifying user "Zhang" who reserves badminton courts 12 times per month, averaging 1.5 hours per session, and prefers air-conditioned indoor courts; user "Li" reserves tennis courts 3 times per week, primarily on weekend afternoons, and often brings family members. The system uses these profiles to provide personalized services, such as offering discounted indoor badminton court packages to Zhang and family workout packages to Li. Furthermore, the profiles can identify unusual behavior; if a user fails to show up or cancel their reservations three times consecutively, the system sends a reminder SMS and adds an "attendance confirmation" step to subsequent reservations to prevent wasting court resources.

[0144] The facility health report integrates all anomaly records and maintenance status of the facilities, presenting the health status of each facility in a scoring format: for example, the health score of football field 01 is 92 points (no major anomalies, only slight fading at the edge of the grass, which has been repaired), and the health score of basketball field 03 is 85 points (two small cracks exist, maintenance work orders have been assigned). The report also lists the facility's "health trend curve," such as the football field 01's health score increasing from 88 points at the beginning of the year to the current 92 points, indicating that maintenance measures are effective. The revenue analysis report statistics on venue rental income, membership fee income, advertising revenue, etc., and calculates "revenue per unit area" based on utilization rate data: for example, the revenue per unit area for badminton courts is 120 yuan / square meter per month, and for tennis courts it is 180 yuan / square meter per month. The system will suggest that administrators increase the rental price of tennis courts during peak hours or increase the number of badminton courts to improve overall revenue.

[0145] The decision support interface can automatically generate operational optimization suggestions based on these reports. For example, regarding the heatmap showing that "badminton courts 1-4 are fully booked from 9-11 AM Monday to Friday," it is recommended that administrators open courts 5-8 as backup during this period and adjust the allocation rules of the reservation system to guide some users to reserve available courts. Regarding the issue of "high recurrence rate of cracks after repair in basketball courts" in the facility health report, it is recommended to replace the surface material with a more durable one and increase the frequency of quarterly in-depth inspections. Regarding the situation in the revenue analysis report where "advertising revenue from football fields accounts for only 5%," it is recommended to increase advertising space around the fields and collaborate with local sports brands to place advertisements. Administrators can directly click the "Adopt Suggestion" button on the decision support interface, and the system will automatically adjust relevant settings, such as updating the allocation rules of the reservation system and generating material procurement requisition forms, enabling rapid implementation from data analysis to operational decisions.

[0146] In one embodiment, the user terminal is also used to respond to a user's venue reservation operation and generate a unique electronic voucher containing reservation information.

[0147] The sensing and execution device layer also includes a credential verification device located at the site entrance.

[0148] The edge computing server or cloud management platform is also used to: after the credential verification device reads the unique electronic credential, verify the validity of the unique electronic credential, and control the controlled equipment set at the entrance of the venue to be turned on or kept off based on the verification result.

[0149] The verification of the validity of the unique electronic voucher includes at least checking whether the reservation period is within the validity period and whether the current number of people using the venue does not exceed the number registered in the reservation information.

[0150] In this embodiment, when a user initiates a venue reservation through the integrated sports field management app on their mobile phone, they need to select the sport, venue number, reservation time slot, and number of accompanying persons. After submission, the system generates a unique electronic voucher containing a QR code. The voucher encrypts and stores the reservation ID, venue information, valid time slot (accurate to the minute), and maximum number of people allowed in. For example, if Mr. Wang reserves badminton court number 5 from 2 PM to 4 PM on Saturday, with 3 people accompanying him, the electronic voucher generated by the system will be marked "Court number 5, 2024-06-15 14:00-16:00, limited to 4 people," and the voucher will be synchronized to the reservation database on the cloud management platform.

[0151] The voucher verification device at the venue entrance employs a dual verification mode of "QR code scanning + facial recognition": the high-definition scanning module on the top of the device (scanning speed ≤0.5 seconds) can read the electronic voucher QR code displayed on the user's mobile phone, while the facial recognition camera on the side (supporting liveness detection) captures the user's facial features. When Mr. Wang and his party arrived at the entrance of badminton court No. 5, they first scanned the electronic voucher QR code. The device immediately sent the reservation information parsed from the QR code to the edge computing server. The server checked whether the reservation time was within the current time ±15 minutes buffer period (to prevent users from missing verification due to lateness), and at the same time, it confirmed that Mr. Wang was the initiator of the reservation by associating his user ID with facial recognition. If the verification is successful, the server sends an "open" command to the controlled gate at the entrance, and the gate opens within 2 seconds. If Mr. Wang is accompanied by 4 people (more than the 3 people registered in the reservation), the system will display a message on the verification device's screen that "The current number of people entering has exceeded the reservation limit. Please check and verify again." The system will also send an over-limit reminder to Mr. Wang's App. He can only enter again after he confirms that he has reduced the number of people entering or made a new reservation.

[0152] If the electronic voucher is lost, the user can enter the mobile phone number reserved during the reservation on the touch screen of the verification device. The system will send an SMS containing a 6-digit verification code to the smart terminal corresponding to that mobile phone number. The user can then enter the verification code to complete the verification. For elderly users or other groups unfamiliar with smart devices, the administrator can enter the user's name and ID number through the backend system to retrieve the corresponding reservation record to assist in verification. In addition, the system supports the "multi-person reservation shared voucher" function: if Mr. Wang sets Ms. Li, who is traveling with him, as a "shared verifier" in the App in advance, Ms. Li can directly enter by facial recognition verification without having to scan the QR code again.

[0153] When an electronic voucher is invalidated (e.g., the reservation time slot has expired, the voucher has been forged, or the number of attendees exceeds the limit), the verification device will display the specific reason for the error and upload the abnormal verification record to the cloud management platform. For example, if a user attempts to enter the venue using a QR code modified with Photoshop, the verification device's QR code parsing module will detect a mismatch in the encrypted information, immediately triggering a "forged voucher" alarm. The platform will then push a real-time alert to the administrator's terminal and simultaneously activate the surveillance camera at the entrance to record video, preserving evidence of the violation. Through this "electronic voucher + dual verification" mechanism, the system can accurately control venue entry permissions, prevent unauthorized personnel from occupying resources, and ensure venue usage order and user rights.

[0154] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0155] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0156] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent management system for a comprehensive sports field, characterized in that, The system includes a user terminal, a sensing and execution device layer deployed on the sports field, an edge computing server, and a cloud management platform; The user terminal is used to provide users with venue reservation, status query, data reception and interactive interface; The sensing and execution device layer includes various sensors for collecting data on the sports field environment, usage status, and personnel activities, as well as controlled devices for executing control commands. The edge computing server, deployed locally in the sports venue, is used to receive and process sensor data from the sensing and execution device layer in real time, call a pre-deployed local artificial intelligence algorithm model to identify specific events or states related to sports venue management from the sensor data, generate corresponding control commands based on the specific events or states and send them to the controlled devices, and / or generate early warning information. The cloud management platform is communicatively connected to the edge computing server and the user terminal. It is used to collect and store processing result data related to the specific event or state from the edge computing server, as well as business and interaction data from the user terminal. It integrates, deeply analyzes and statistically processes all the collected data to generate comprehensive management reports and provides administrators with a decision support interface based on the comprehensive management reports. The local artificial intelligence algorithm model deployed on the edge computing server includes at least a visual analysis model for real-time analysis of video streams to identify specific events or states.

2. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include athlete injury events; The local artificial intelligence algorithm model includes a posture analysis model based on skeletal key point detection, which is used to identify action postures that conform to preset injury determination rules from the video stream; The preset injury determination rules include detecting a rapid fall caused by a sudden drop in a person's center of gravity, and the fall posture lasting for more than a preset time threshold without returning to a standing or walking posture.

3. The intelligent management system according to claim 2, characterized in that, The process by which the edge computing server generates corresponding control commands and sends them to the controlled device, and / or generates early warning information, includes: Trigger the site broadcasting system to play a preset warning message; Push notifications containing the location information of the event to the terminals of administrators, pre-set medical staff, and corresponding appointment users; Keep the access control equipment at the site entrance open to allow rescue personnel to enter.

4. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include basketball-related events; The local artificial intelligence algorithm model includes a basketball motion analysis model, used to identify goal events from video streams, distinguish goal types, and / or track player trajectories to collect individual motion data.

5. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include football-related events; The sensing and actuation device layer includes a goal line detection device deployed on the football goal line; The local artificial intelligence algorithm model includes a football motion analysis model, used to process the data from the goal line detection device to determine whether a goal is valid, and / or to track player trajectories from the video stream to collect running and tactical data.

6. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include tennis-related events; The local artificial intelligence algorithm model includes a tennis motion analysis model, which is used to reconstruct the three-dimensional trajectory and calculate the landing point of the tennis ball based on multi-angle video streams, in order to determine out-of-bounds events and / or statistically analyze serve speed and shot type data.

7. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include competition violation events; The local artificial intelligence algorithm model includes a violation action recognition model, which is used to identify preset types of foul actions or violations in basketball or football games from video streams; In response to the competition violation, the edge computing server sends auxiliary judgment information, including the event time and video clip, to the designated referee terminal.

8. The intelligent management system according to claim 7, characterized in that, The system supports a video playback challenge mechanism; The user terminal is configured to receive a penalty challenge request initiated by the user; The edge computing server or the cloud management platform is configured to automatically retrieve multi-angle video from the relevant time period and generate replay clips in response to the penalty challenge request, for the referee's terminal to retrieve and view.

9. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include the state of exceeding the limit for the number of people in the venue; The local artificial intelligence algorithm model includes a real-time people counting model based on target detection and tracking, which is used to identify from the video stream that the current number of people exceeds the preset capacity limit of the corresponding venue; In response to the overcrowding status of the venue, the edge computing server sends a locking command to the entrance access control device and / or sends a full-occupancy message to the venue display screen and pushes an overcrowding alarm to the administrator terminal.

10. The intelligent management system according to claim 1, characterized in that, The sensing and actuation device layer includes a light intensity sensor and a controllable lighting device; The edge computing server is configured as follows: The ambient light level is determined based on the data collected by the light intensity sensor. Based on the ambient light level and the current usage status of the site, control commands are generated to adjust the brightness or on / off status of the controllable lighting equipment.

11. The intelligent management system according to claim 1, characterized in that, The specific events or states identified by the edge computing server include abnormal states of site facilities; The sensing data includes images or sensor data that reflect the condition of the site surface; The local artificial intelligence algorithm model or analysis program is used to identify at least one abnormal state from the sensor data, including ground cracks, potholes, fading, or lawn health below a preset health threshold. In response to the abnormal status of the site facilities, the cloud management platform automatically generates a maintenance work order that includes the location and type of the abnormality.

12. The intelligent management system according to claim 1, characterized in that, The comprehensive management reports generated by the cloud management platform include at least one of the following: site utilization heat map, user behavior analysis profile, facility health report, and revenue analysis report; The decision support interface provides operational optimization suggestions based on the comprehensive management reports.

13. The intelligent management system according to claim 1, characterized in that, The user terminal is also used to respond to the user's venue reservation operation and generate a unique electronic voucher containing reservation information. The sensing and execution device layer also includes a credential verification device located at the site entrance; The edge computing server or cloud management platform is also used to: after the credential verification device reads the unique electronic credential, verify the validity of the unique electronic credential, and control the controlled equipment set at the entrance of the venue to be turned on or kept off based on the verification result; The verification of the validity of the unique electronic voucher includes at least checking whether the reservation period is within the validity period and whether the current number of people using the venue does not exceed the number registered in the reservation information.

Images