Multi-node collaborative recognition and dynamic anti-cheating exercise and running data management system

Abstract

According to the multi-node collaborative recognition and dynamic anti-cheating exercise running data management system provided by the invention, multi-node distributed camera deployment is combined with a face recognition technology, so that the technical problems of low congestion recognition rate, insufficient night performance, poor data accuracy and the like in a traditional running counting system are solved. The system comprises a camera array, a data asynchronous processing module, a dynamic error correction algorithm and a security isolation strategy, and the functions of millisecond-level error calculation, multi-site adaptation, high-concurrency data processing and the like are achieved. By optimizing the asynchronous processing flow of the face equipment and combining the expansion capability of the super-brain equipment, the recognition accuracy and stability of the system in a complex scene are remarkably improved, the effective score warehousing rate reaches 98% or above, the night recognition rate exceeds 91%, and the method has the capability of automatically cleaning junk data.

Description

Technical Field

[0001] The present invention relates to the field of sports data management technology, and specifically to a sports running data management system and method based on multi-node collaborative identification and dynamic anti-cheating technology, which is suitable for running data collection and management in scenarios such as schools and fitness venues. Background Art

[0002] With the growing awareness of national fitness and the development of information-based physical education, sports running data management systems are increasingly being used in various scenarios. However, existing sports running data management systems have numerous shortcomings. Firstly, during data collection and processing, image recognition accuracy in complex environments is low, prone to errors, and it is difficult to effectively combat cheating, resulting in a lack of assurance of data authenticity and reliability. For example, in low-light nighttime scenes, camera image quality is poor, affecting facial recognition and accurate recording of running data. Some users may also falsify running results through illicit means such as skipping nodes and falsifying pacing. Furthermore, existing systems also have shortcomings in device management, data interaction, and security management. Device management lacks remote operation and maintenance capabilities and real-time status monitoring. Data interaction and visualization are insufficient, and data security measures are inadequate, failing to meet the tiered data access and manipulation requirements of different users. Therefore, there is an urgent need for a sports running data management system that can achieve multi-node collaborative recognition, dynamic anti-cheating measures, efficient device management, real-time interaction, and secure data management. Summary of the Invention

[0003] In response to the above problems, the present invention aims to provide a sports running data management system and method with multi-node collaborative identification and dynamic anti-cheating, so as to solve the problems of existing sports running data management systems such as large recognition errors, weak anti-cheating capabilities, inconvenient equipment management, untimely data interaction, and poor data security performance.

[0004] The present invention provides a multi-node collaborative identification and dynamic anti-cheating sports running data management system, which includes a data acquisition module, an AI motion processing module, a real-time interaction module, and a data service module; wherein,

[0005] The data acquisition module consists of a distributed node image acquisition unit and an image processing unit. The distributed node image acquisition unit utilizes a redundant deployment architecture, with monitoring units deployed at key runway nodes. Each monitoring unit contains three cameras with a horizontal angle of 15°, covering a runway area ≥20 meters wide and 10 meters long. Its edge devices feature a built-in GPU acceleration module, supporting local facial feature extraction and synchronizing data with the central server via Gigabit Ethernet (latency ≤ 50 milliseconds). The image processing unit utilizes a multi-level asynchronous processing architecture. The data acquisition layer uses an independent thread to pull RTSP streams, invoke the facial recognition device for feature extraction, and encapsulates the raw feature values ​​as DataPacket objects, storing them in a 1024-blocking queue. The processing layer uses a fixed thread pool of 8 threads to retrieve data from the queue, perform feature comparison (processing ≤15 facial data in a single batch), and asynchronously notify the business module of the results via a Kafka message queue. A load balancing mechanism automatically prioritizes recognition tasks during peak hours, ensuring a response time of ≤300 milliseconds in high-concurrency scenarios. Device failures automatically switch to a backup channel.

[0006] The AI ​​motion processing module includes a dynamic error correction unit, a dynamic anti-cheating unit, and a mileage calculation unit. The dynamic error correction unit implements simultaneous compensation of multi-track data based on the structural characteristics of the track. Using the node distance of the reference track as a reference, it compensates for the offset distance d caused by differences in camera position on each track. The compensation value △S for each track is calculated using the formula △S=2πd (where d is the camera position offset), resulting in an equivalent recognition length of L'=L±△S for each track. The system also supports technicians who can adjust the offset parameter d and the compensation distance △S through on-site measurement and analysis of camera images to ensure compensation accuracy and recognition consistency. The dynamic anti-cheating unit detects pace anomalies in real time and calculates the user's real-time pace between adjacent nodes: Vi = Li′ / △ti (Li′ is the segment distance after dynamic error correction, and △ti is the time difference between the two nodes). If Vi > 12m / s (the maximum speed for human sprinting) or Vi < 0.5m / s (the lower limit of walking speed), cheating is determined and the result is invalid. A node jump logic check is performed to construct a node topology sequence. If the user directly appears at a non-adjacent node Nj+k (k>1) from node Nj, an anomaly is triggered and the result is invalid. The maximum number of jump nodes, Kmax, is set. If ΔK> Kmax (ΔK is the actual number of jump nodes), the result is invalid. The system also includes a night mode enhancement strategy. When the ambient light is ≤ 10 lux, infrared fill light is automatically turned on and black and white recognition mode is switched. The face comparison threshold is dynamically adjusted from 0.75 to 0.6 to compensate for the impact of image noise. The mileage calculation unit calculates the segmented mileage. Based on the sequence of adjacent nodes passed by the user and the standard distance L between nodes, combined with the equivalent identification length L' after runway offset compensation, the actual mileage of each segment is calculated as L'=L±△S; the total mileage is accumulated. When the user completes the entire running path (i.e., passes the final identification node), all valid segmented mileage is accumulated, and the total mileage S=ΣL'_i; abnormal data processing is performed. If the dynamic anti-cheating unit determines that a certain segment of data is invalid, detailed information will be prompted in the application, such as substandard pace, substandard nodes, and skipped nodes, and detailed cheating judgment photos and time nodes will be displayed.

[0007] The real-time interaction module enables real-time data exchange and information sharing between system modules and with users. This module deploys a large LED screen next to the playground, displaying two areas: total mileage ranking: displays the top 20 male and female runners (name + class + mileage + total exercise time) in descending order of accumulated mileage by semester, updated every five minutes; and real-time distance ranking: ranks the top 50 runners in the current run, with real-time updates.

[0008] The data service module includes a data open upload unit, a data security isolation unit, a venue management unit, and a device management unit. The data open upload unit provides a standardized data interface, namely a RESTful API, supporting data upload and query in JSON format, with a response time of ≤100ms. Open data types include user running tracks (timestamp, node location, pace), total mileage, and cheating records (including abnormal photos and reason codes). The unit allows the school's academic affairs system to batch export encrypted data packets (AES-256 encryption) for use in physical education performance calculation. The data security isolation unit adopts a hierarchical permission policy. Teachers can view data for their classes and manually correct abnormal scores (requires secondary authentication by scanning a QR code). Administrators have full data access and their operations are recorded in the blockchain log. Facial feature values ​​and running data are transmitted using AES-256 encryption, and the transport layer uses the TLS1.3 protocol to prevent man-in-the-middle attacks. The venue management unit is used to manage basic information and usage status of the sports venue, set activity requirements, and manage sports data. It includes functions such as activity parameter configuration, data statistics and analysis, and user permission management. It also displays device online status, recognition rate, and abnormality warnings. The device management unit has a device status monitoring mechanism. Through the heartbeat monitoring function of the edge computing device and the system's active polling (cycle of 30 minutes), it can realize real-time monitoring of the device's online status and automatically trigger an alarm when the device is offline. It has remote operation and maintenance capabilities, supports remote adjustment of the camera's focal length and fill light intensity on the web and mobile terminals, and can also view video streams and recognition logs in real time. It has facial data management functions. The edge computing device uses customized firmware to generate 128-dimensional feature vectors through batch import or user submission (reviewed by the administrator) and store them in the face database.

[0009] The present invention also provides a Sunshine Long-Distance Running Counting Anti-Cheating Method, which is applied to the above-mentioned system and includes:

[0010] Multi-node cameras are deployed at key nodes on the runway, with three cameras per node configured for redundancy and a horizontal angle of 15°. Edge computing is used to collect and pre-process facial images and motion data in real time.

[0011] Adopting an asynchronous processing flow, decoupling the data collection layer and the business logic layer through a two-layer architecture, using a multi-threaded task queue to manage facial data, and switching to a backup channel when a device anomaly occurs;

[0012] Filter invalid data based on node jump threshold, time and space logic verification and night infrared fill light strategy, and automatically clean up unfinished node garbage data;

[0013] Error compensation calculation is performed based on the physical parameters of the inner and outer lanes of the runway and the formula L′=L±ΔS;

[0014] Secure isolation is achieved through hierarchical permissions and AES-256 encrypted transmission and storage of data.

[0015] The anti-cheating judgment rules mentioned include:

[0016] 1) Node sequence logic check: Record the time sequence of the student passing the nodes. If the current node ID - the previous node ID > the maximum number of allowed skip nodes, the score is considered invalid;

[0017] 2) Single-node residence time threshold judgment: If the interval between two recognitions of the same node is less than 10 seconds and the node does not change, it is judged as "repeated face recognition". If it is judged as "repeated face recognition", the data will be discarded;

[0018] 3) Dynamic recognition priority adjustment: Through real-time monitoring of CPU load and queue length, 70% of computing resources are automatically allocated to recognition tasks during peak hours (>200 people detected simultaneously), ensuring a response time of ≤300ms.

[0019] The beneficial effects of the present invention are: improving data accuracy, through multi-node collaborative identification and dynamic error correction, the distance error of each runway is less than 0.5%, and the speed calculation error is less than 1%; enhancing anti-cheating capabilities, adopting a multi-level anti-cheating mechanism, including node sequence verification, time and space logic verification, etc., to effectively prevent cheating behaviors such as running on behalf of others and taking shortcuts; improving equipment management, equipment heartbeat monitoring and automatic restart mechanism to ensure stable operation of the equipment; backup channels quickly take over to ensure the continuity of data collection; ensuring data security, complying with GDPR data protection requirements, blockchain logs to ensure traceability of operations, no risk of over-authorization, hierarchical authority strategies and encrypted transmission and storage, effectively ensuring the security of sensitive data such as students' faces. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Flowchart of the system architecture algorithm

[0021] Figure 2 Installation diagram for playground equipment DETAILED DESCRIPTION

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] See also Figure 1 As shown in the examples, the system architecture of the present invention includes a data acquisition module, an AI motion processing module, a real-time interaction module, and a data service module. These modules work together to collect, process, store, and manage running data.

[0024] See also Figure 2As shown in the example, in the data collection and processing process, the distributed node image acquisition unit is deployed, for example, at key nodes along the runway (e.g., every 100 meters). Each node is equipped with three cameras, forming a triangular coverage area with a horizontal angle of 15°, ensuring comprehensive coverage of a 20-meter wide and 10-meter long runway. The cameras are powered by PoE and connected to edge computing devices. They feature a built-in GPU acceleration module for local facial feature extraction and communicate with the central server via Gigabit Ethernet, ensuring end-to-end latency of less than 50 milliseconds.

[0025] Edge computing devices use multi-threaded parallel processing:

[0026] Thread 1: Pulls RTSP stream and performs image preprocessing (noise reduction, distortion correction);

[0027] Thread 2-3: Calls the GPU-accelerated face detection algorithm to extract facial feature values ​​and encapsulates them as DataPacket objects;

[0028] Thread 4: Synchronizes the feature values ​​to the central server via Gigabit Ethernet, and locally caches the last 5 minutes of data for anomaly backtracking.

[0029] In the image processing unit, a two-layer asynchronous architecture is used to decouple data acquisition and business logic:

[0030] 1) Data collection layer: An independent thread pool (4 threads) pulls the RTSP stream, calls the facial recognition device for detection, and stores the original feature values ​​in a blocking queue with a capacity of 1024;

[0031] 2) Processing Layer: A fixed thread pool of 8 threads retrieves data from the queue, processing 15 facial data sets in a single batch. The business module is asynchronously notified of the processing results via the Kafka message queue. During peak hours, recognition task priority is automatically increased, and the thread pool dynamically adjusts its policy to ensure a response time of ≤300 milliseconds. In the event of a device anomaly, a backup channel automatically takes over data processing using a load balancing algorithm.

[0032] Edge computing devices feature heartbeat monitoring, automatically uploading data to the system if the device loses connection. The system proactively sends monitoring requests to the device every half hour to determine its online status. Remote control is enabled through a RESTful API on both the web and mobile platforms, supporting real-time adjustment of camera focus and fill light intensity, as well as viewing of video streams and recognition logs.

[0033] 4K-resolution LED screens (size tailored to the venue's scale, such as 10m x 3m) were deployed in prominent locations at the main entrance and along the runway. These screens are connected to the central server using a distributed architecture. Each screen has a built-in edge computing terminal, supporting local data caching and real-time rendering. They communicate with the system via gigabit Wi-Fi or a wired network, with data latency ≤ 200 milliseconds.

[0034] To build a facial database, edge computing devices integrate a facial recognition engine. After batch importing student photos, the system uses a deep learning model to generate 128-dimensional feature vectors for storage in the facial database. Students can submit facial update requests via mobile devices. After administrator review, these requests are synchronized to all nodes via a message queue, ensuring real-time consistency in the facial database.

[0035] Dynamic anti-cheating mechanism implementation, including:

[0036] 1) Node sequence verification: The system assigns a unique index to each track node (e.g., 0 for the first scan and N for the last scan). A timestamp is generated each time a student passes a node. If the difference between the current node index and the previous node index exceeds 1 (e.g., going directly from 0 to 2), the node is considered a skipped node, and the score is invalid.

[0037] 2) Single-node dwell detection: Use a concurrently safe HashMap to store each student's most recent face scan time, with a timeout of 10,000 milliseconds. If the same node detects the same face multiple times within 10 seconds and the node remains unchanged, it is considered a duplicate scan and the data is discarded.

[0038] 3) Enhanced night mode: When ambient light is ≤10 lux, the system automatically activates the camera's infrared fill light, switches to black and white recognition mode, and reduces the feature comparison threshold from 0.7 to 0.6. A head posture detection module has also been added, using a deep learning model to identify unusual postures such as tilting the head and facing sideways, further improving nighttime recognition accuracy.

[0039] Dynamic error correction implementation, including:

[0040] 1) Taking a standard 400-meter track and field stadium as an example, the benchmark runway is set as lane 3, with a curve radius of r meters and a circumference of L = 400 meters. Distributed camera nodes are deployed at key runway nodes (every 100 meters), with each node equipped with three cameras arranged in a sequence to monitor the runway, achieving full runway monitoring without blind spots.

[0041] 2) Determine the offset distance d. Use a laser rangefinder to measure the actual monitoring range of each lane's camera and determine the time difference caused by differences in shooting angles. For Lane 1 (the inner lane), the camera's shooting angle is forward, so the athlete enters the frame 0.5 seconds earlier, corresponding to an offset distance d1 = -1.2 meters (a negative sign indicates an inner offset). For Lane 5 (the outer lane), the camera's shooting angle is backward, so the athlete enters the frame 0.8 seconds later, corresponding to an offset distance d5 = +1.5 meters (a positive sign indicates an outer offset).

[0042] 3) Calculation of compensation distance △S: Calculate the compensation distance for each runway according to the formula △S=2πd:

[0043] Track 1: △S1=2π×(-1.2)≈-7.54 meters (total mileage reduced by 7.54 meters);

[0044] Track 5: △S5=2π×1.5≈+9.42 meters (total mileage increased by 9.42 meters);

[0045] The total mileage of the first track is L′=L±△S1=400-7.54=392.46 meters;

[0046] The total mileage of the 5th track is L′=L±△S=400+9.42=409.42 meters.

[0047] Data is open for upload and can be imported through Excel or manually input to mark grades in batches. Abnormal data can be rejected with one click.

[0048] Data security and rights management, including;

[0049] 1) Hierarchical permission control: After logging in through the parent mobile app, students can only view their own running records and facial information. Teachers can view data for their classes and manually correct abnormal scores (such as missed tests due to equipment failure) after scanning the QR code for secondary authentication. Operation records are synchronized to the blockchain log. Administrators have full data permissions, and all operations are recorded on the blockchain with digital signatures to ensure that they cannot be tampered with.

[0050] 2) Encrypted Transmission and Storage: Facial feature values ​​and running data are encrypted at the transport layer using the TLS 1.3 protocol, using the AES-256-GCM algorithm for data encryption and RSA-2048 for key exchange. Storage is performed using a sharded database and table strategy, with facial feature values ​​physically isolated from business data, and encryption keys managed via a hardware security module (HSM).

[0051] Site and equipment management, including:

[0052] View device online status, recognition rate, and abnormal warnings, and support filtering abnormal data by class; the device management unit supports Web remote debugging and adjusts camera parameters (such as PTZ control and exposure compensation) through the ONVIF protocol.

[0053] In one embodiment, the process is:

[0054] 1) Equipment deployment: Four nodes were deployed along the 400-meter runway, with three cameras on each node to cover the entire runway.

[0055] 2) Data Collection: When a student runs through a node, three cameras synchronously capture images. The edge device extracts feature values ​​and queries the local face database (caching the most recent 1,000 feature data entries). Upon successful matching, a "first check-in event" (including the student ID, timestamp, and node ID = 0) is generated.

[0056] 3) Anti-cheating detection: When passing through an intermediate node, the system verifies that the current node ID = the previous node ID + 1, and the time difference is greater than 30 seconds (to prevent cheating by short-term backtracking). If the conditions are met, mileage is accumulated; otherwise, the result is invalid;

[0057] 4) Error correction: The inner lane compensation distance of lane 1 is ΔS = -7.57 meters, reducing the mileage by 7.54 meters to ensure that the actual running distance is consistent with the reference lane;

[0058] 5) Night mode: When the ambient light drops to 8 lux, the system automatically turns on infrared fill light, and the recognition rate increases from 75% to 92%;

[0059] 6) Dynamic adjustment: During peak hours (e.g., 100 people starting at the same time), the system automatically increases the priority of the recognition task, and the response time is stabilized within 280 milliseconds.

[0060] The above implementation methods describe in detail the technical solutions and application effects of the present invention through specific hardware configurations, software processes and actual cases, covering all technical features in the claims and ensuring the integrity and feasibility of the patent application.

Claims

1. A multi-node collaborative identification and dynamic anti-cheating sports running data management system, characterized by: It includes a data acquisition module (1), an AI motion processing module (2), a real-time interaction module (3) and a data management module (4); wherein, The data acquisition module (1) includes a distributed node image acquisition unit (1.1) and an image processing unit (1.2); The AI ​​motion processing module (2) includes a dynamic error correction unit (2.1), a dynamic anti-cheating unit (2.2), and a mileage calculation unit (2.3); The real-time interactive module (3) is to display the total mileage ranking, real-time progress ranking and personal single running data in real time through the LED large screen; The data service module (4) includes a data open upload unit (4.1), a data security isolation unit (4.2), a site management unit (4.3) and an equipment management unit (4.4).

2. The system according to claim 1, wherein: The distributed node image acquisition unit (1.1): A redundant deployment architecture is adopted: monitoring units are deployed at key nodes on the runway. Each monitoring unit contains three cameras with a horizontal angle of 15 degrees, covering a runway area ≥ 20 meters wide and 10 meters long. Edge computing configuration: The edge device has a built-in GPU acceleration module, supports local facial feature extraction, and synchronizes data with the central server via Gigabit Ethernet (delay ≤ 50 milliseconds).

3. The system according to claim 1, wherein: The image processing unit (1.2) adopts a multi-level asynchronous processing architecture: Data collection layer: pulls RTSP streams through independent threads, calls face recognition equipment to extract features, and encapsulates the original feature values ​​into DataPacket objects and stores them in a blocking queue with a capacity of 1024; Processing layer: Retrieve data from the queue through a fixed 8-thread thread pool, perform feature comparison (single batch processing ≤ 15 facial data), and asynchronously notify the business module of the results through the Kafka message queue; Load balancing mechanism: Automatically increase the recognition task priority during peak hours to ensure response time ≤ 300 milliseconds in high-concurrency scenarios, and automatically switch to the backup channel when equipment fails.

4. The system according to claim 1, wherein: The dynamic error correction unit (2.1): Synchronous compensation of multi-runway data is achieved based on the structural characteristics of the runway. Using the node distance of the reference runway as a reference, the offset distance d caused by differences in camera shooting positions on each runway is compensated. This offset distance d is determined by analyzing the actual camera monitoring range. The compensation value △S for each runway is calculated using the formula △S=2πd (where d is the camera position offset distance), ensuring the equivalent recognition length of each runway is L'=L±△S. The system also supports technicians adjusting the offset parameter d and the compensation distance △S through on-site measurement and analysis of camera images to ensure compensation accuracy and recognition consistency.

5. The system according to claim 1, wherein: The dynamic anti-cheating unit (2.2): Real-time pace anomaly detection: Calculate the real-time pace of the user between adjacent nodes, Vi = Li′ / △ti (Li′ is the segment distance after dynamic error correction, △ti is the time difference between the segments). If Vi > 12m / s (the human sprint limit) or Vi < 0.5m / s (the lower limit of walking speed), cheating is considered and the result is invalid. Node jump logic check: Path continuity check, building a node topology sequence. If a user goes directly from node Nj to a non-adjacent node Nj + k (k>1), an exception is triggered and the result is considered invalid. The maximum number of jump nodes Kmax is set. If ΔK>Kmax (ΔK is the actual number of jump nodes), the result is considered invalid. Includes night mode enhancement strategy: when the ambient light is ≤10 lux, it automatically turns on infrared fill light and switches to black and white recognition mode, dynamically adjusting the face comparison threshold from 0.75 to 0.6 to compensate for the impact of image noise.

6. The system according to claim 1, wherein: The mileage calculation unit (2.3): Segment mileage calculation: Based on the sequence of adjacent nodes passed by the user and the standard distance L between nodes, combined with the equivalent identification length L' after runway offset compensation, the actual mileage of each segment is calculated as L'=L±△S; Total mileage accumulation: When the user completes the entire running path (i.e., the final identified node), all valid segmented mileage is accumulated, and the total mileage S = ΣL'_i, where i is the number of nodes passed (S is the total mileage of the user's run; ∑ is the summation symbol, indicating the accumulation of subsequent items); L'_i is the user's compensated equivalent mileage between the i-th node and the i+1-th node, that is, the segmented distance after dynamic error correction; i: the node number passed by the user); Abnormal data processing: If the dynamic anti-cheating unit (2.2) determines that a certain piece of data is invalid, the data will be prompted in the application with detailed information, such as speed failure, node failure, and cross-node, and detailed cheating judgment photos, time nodes, etc.

7. The system according to claim 1, wherein: The real-time interaction module (3) realizes data synchronization through WebSocket long connection. The LED screen updates at a frequency of 2 minutes / time for total mileage ranking, instant updates for real-time progress ranking, and dynamic refresh of personal data.

8. The system according to claim 1, wherein: The data development upload unit (4.1): Standardized data interface: Provides a RESTful API interface, supports JSON format data upload and query, and the interface response time is ≤100ms; Open data types include: user running track (timestamp, node location, pace), total mileage, cheating judgment record (including abnormal photos and reason codes); Allows the school's academic affairs system to batch export encrypted data packages (AES-256 encryption) for sports performance calculation.

9. The system according to claim 1, wherein: The data security isolation unit (4.2): A hierarchical permission strategy is adopted: teachers can check the data of their classes and manually correct abnormal grades (requires scanning a QR code for secondary authentication), while administrators have full data permissions and record operations in the blockchain log; Facial feature values ​​and running data are transmitted via AES-256 encryption, and the transport layer uses the TLS 1.3 protocol to prevent man-in-the-middle attacks.

10. The system according to claim 1, wherein: The site management unit (4.3) is used to set activity requirements and manage sports data, including activity parameter configuration, data statistical analysis, user authority management and other functions, and displays the device online status, recognition rate and abnormal warning.

11. The system according to claim 1, wherein: The device management unit (4.4) has: Device status monitoring mechanism: Through the heartbeat monitoring function of the edge computing device and the system's active polling (cycle of 30 minutes), the device's online status is monitored in real time, and an alarm is automatically triggered when the device goes offline; Remote operation and maintenance capabilities: Supports remote adjustment of camera focus and fill light intensity via web and mobile terminals, and real-time viewing of video streams and identification logs; Facial data management: Edge computing devices use customized firmware to support real-time updates of the facial database and automatic annotation of abnormal data. 128-dimensional feature vectors are generated and stored in the facial database through batch import or user submission (after administrator review).

Images