Intelligent fitness following guidance system and method based on terminal state determination

Abstract

The application discloses an intelligent fitness following guidance system and method based on terminal state judgment, and relates to the technical field of intelligent internet of things and edge computing. The system dynamically adjusts the multi-source positioning fusion weight (including Wi-Fi, UWB, vision, radar, etc.) by detecting the wearing / standing state of the intelligent terminal in real time, solves the problems of positioning drift and identity loss caused by the device leaving the body in the gym scene; relies on local edge computing to complete posture recognition and original data destruction in real time, and guarantees privacy compliance; and according to the terminal state, the guidance information is intelligently routed to the audio terminal or converted into subtitles for display, realizing low-delay, high-robustness and all-process intelligent fitness guidance experience.

Description

Technical Field

[0001] This invention relates to the fields of intelligent Internet of Things, edge computing and artificial intelligence application technology, and more specifically, to an intelligent fitness follow-up guidance system and method based on terminal status determination. Background Technology

[0002] With the deep integration of IoT and AI technologies in the sports and health industry, unmanned and intelligent gyms have become an important trend in industry development. However, existing virtual fitness guidance systems face the following key technical bottlenecks in practical implementation:

[0003] 1. Identity and location binding is easily lost: When users perform strength training, they often place their smartphones or other devices on the ground or next to the equipment. If the system still uses the signal from this stationary device as the core positioning source, it will lead to misjudgment of location; if multi-source data is fused indiscriminately, it will cause serious coordinate drift, making it impossible for the virtual coach to follow, and interrupting the user experience.

[0004] 2. Privacy risk and high latency: Traditional solutions often upload camera video streams to cloud servers for posture recognition and analysis. This not only poses a significant privacy risk of leaking raw video data, but also causes delays in network transmission and cloud processing, resulting in delayed voice correction for safety and making it impossible to prevent sports injuries in a timely manner.

[0005] 3. The intervention methods are disconnected from the user's physical state: When the user's smart terminal is idle, the voice guidance pushed to this terminal by the system cannot be received in a timely manner. The existing audio intervention system lacks an intelligent routing mechanism based on the physical state of the terminal.

[0006] 4. Sensor blind spots: Cameras have difficulty recognizing posture when there is insufficient light or severe obstruction; wireless positioning is easily interfered with in areas with dense metal equipment, and there is a lack of non-contact backup sensors that can provide both positioning and posture perception.

[0007] Unlike existing technologies that rely solely on fixed weights or simple superposition of multi-source data, this invention introduces dynamic metadata such as "terminal physical wearing status" to achieve real-time assessment and weight reconstruction of the reliability of heterogeneous positioning sources, fundamentally solving the systemic failure problem caused by the device being removed from the user.

[0008] In the prior art, patent application CN120871020A discloses a Bluetooth and UWB fusion positioning method based on Kalman filtering, which only fuses Bluetooth and UWB positioning sources, and the weight adjustment is based on static region division, without considering changes in the physical wearing state of the terminal. Patent application CN120563731A discloses a posture recognition method based on millimeter-wave radar, which uses millimeter-wave radar as the core posture perception source, but does not involve automatic switching with a visual camera or privacy protection mechanisms. Patent application CN121281090A discloses an anti-tailgating detection system for an unmanned gym, which only solves the identity verification problem at the entrance access control point, and does not involve continuous identity tracking after entry. Compared with the above-mentioned prior art, the technical solution of the present invention is fundamentally different. In summary, existing technologies either fail to address the issue of devices being removed from the body, or fail to achieve seamless switching and privacy protection for multimodal perception. Furthermore, they do not use the "physical wearing status of the terminal" as a unified control input to collaboratively manage the entire system. Therefore, they cannot provide the end-to-end, robust, low-latency, and privacy-secure fitness guidance experience achieved by this invention. Summary of the Invention

[0009] The purpose of this invention is to overcome the above-mentioned defects of the prior art and provide an intelligent fitness follow-up guidance system and method that can be completed locally in a closed loop, resists device separation interference, has low latency, and complies with privacy.

[0010] The core technical idea of ​​this invention is to introduce "terminal physical wearing status" as the core decision variable, dynamically trigger the weight redistribution of multi-source fusion positioning, and ensure that the identity coordinates are not decoupled; rely on local edge computing to complete posture extraction and instant destruction of raw data to ensure privacy and security; and finally intelligently route intervention guidance information to the user's audio terminal or convert it into subtitle display to form a seamless technical closed loop.

[0011] Special Note: The health data processing involved in this invention is solely for sports safety assistance and risk control, and does not involve disease diagnosis, treatment, or medical advice, falling within the scope of non-medical technical applications. The processing and application of the health physiological indicator data and real-time heart rate data are strictly limited to sports risk control and safety assistance for non-medical purposes, and do not constitute any form of medical diagnosis, treatment advice, or health assessment. The intervention guidance information is limited to adjustment suggestions related to sports safety, such as exercise load and movement posture, and does not contain any medical diagnostic conclusions or treatment recommendations. The AI ​​decision-making mechanism adopted in this invention is entirely based on the user's objective physiological indicators (heart rate value) and movement parameters (joint angles), and does not involve any judgment based on personality attributes such as race, gender, or age, or discriminatory decisions, complying with AI ethical requirements. All decision rules can be interpreted through a preset safety boundary model, and users can view the decision basis at any time. Data collection and processing in this invention are carried out with the explicit authorization of the user. Raw video data is processed only on the local edge computing node and destroyed immediately, without being uploaded to the cloud or persistently stored, thus protecting user privacy at the physical level.

[0012] Compared with the prior art, the present invention has the following beneficial effects: - By dynamically adjusting the weights based on the physical state of the terminal, the problem of location drift and identity loss caused by the device being removed from the user is solved; - Millimeter-wave radar was used as a backup sensing source to fill the sensing blind spots of the camera; - Edge computing processes and destroys raw data locally and instantly, achieving privacy compliance; - Intelligent audio routing and subtitle backup ensure 100% delivery of guidance information. Attached Figure Description

[0013] Statement: "In the accompanying drawings of this application, solid arrows indicate data flow, signal flow, or process execution direction; dashed arrows indicate control flow, wireless communication connections, or optional logical switching relationships. Flowchart ( Figure 2 , Figure 4 All arrows in the diagram indicate the direction of process execution and are consistently drawn with solid lines.

[0014] Figure 1 This is a schematic diagram of the system module architecture of the present invention. In the diagram: 1-communication module, 2-sensor component, 3-memory, 4-processor, 5-local edge computing node.

[0015] Figure 2 The logic block diagram for multi-source fusion positioning and dynamic weight adjustment algorithm.

[0016] Figure 3 This is a schematic diagram of the data flow for local pose recognition and privacy protection at edge computing nodes.

[0017] Figure 4 Flowchart for adapting to and following diverse display terminals. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to embodiments.

[0019] Example 1: Hardware Topology and Terminal Status Determination

[0020] This system is deployed in an unmanned fitness space. Its core hardware includes a ceiling-mounted UWB base station array, area cameras, millimeter-wave radar sensors, interactive displays for each training area, and local edge computing nodes. The processor can be part of a central server deployed in the fitness space or integrated into the local edge computing node. The local edge computing node is a dedicated computing device physically located close to the sensor components, used to perform low-latency, high-privacy data processing tasks.

[0021] The system establishes connections with heterogeneous smart terminals (including but not limited to smartphones, smartwatches, smart bracelets, and Bluetooth headsets) carried by users via a communication module. Sensor components (cameras, millimeter-wave radar, etc.) communicate with local edge computing nodes via wired or dedicated wireless links to transmit collected video streams and point cloud data to the edge computing nodes in real time.

[0022] Terminal status determination logic: The system collects IMU data from the smart terminal and extracts motion rhythm through frequency domain analysis. Specifically, it collects acceleration data within 1 second, performs Fast Fourier Transform (FFT), and calculates the matching degree between the peak frequency and the typical frequency band of human motion (0.5-3Hz). If the terminal's spatial coordinates change very little in a short period of time, and its acceleration frequency domain characteristics have a low matching degree with human motion rhythm, while the spatial distance from key points on the human body exceeds a preset threshold (e.g., 0.5 meters), it is determined to be in a "static state"; otherwise, it is in a "wearing state".

[0023] Example 2: Dynamic Weighted Fusion Positioning Based on Wi-Fi and Millimeter-Wave Radar

[0024] This embodiment employs the Extended Kalman Filter (EKF) algorithm to fuse multi-source positioning data. The state vector of the EKF includes the user's position and velocity, and the observation vector includes at least three positioning source data. When the terminal determines to be in a "stationary state," the processor immediately adjusts the weights: by adjusting the observation noise covariance matrix of the EKF, the weight of the terminal's inertial navigation data is reduced (e.g., reduced to below 30% of the original weight), while the weights of visual positioning, millimeter-wave radar positioning, and other on-board terminal positioning data are increased (e.g., the visual weight is increased to more than 1.8 times the original weight). The above weight adjustment ratios are merely examples; in practical applications, they can be dynamically adjusted according to signal quality and do not constitute a limitation of the present invention.

[0025] This strategy significantly reduces location drift, ensuring that the user's identity and location remain continuously bound as they move, and that the virtual guidance screen switches smoothly without any lag. In a simulated test environment where the smartphone is stationary on the ground and the user moves 5 meters, the traditional fixed-weight fusion solution experiences a location drift of approximately 1.2 meters and an identity loss recovery time of approximately 5 seconds. This solution, through terminal status-aware weight adjustment, reduces the location drift to only approximately 0.2 meters, shortens the identity loss recovery time to 0.8 seconds, and reduces the location interruption rate by more than 90%.

[0026] Example 3: Local zero-privacy gesture recognition and rapid intervention

[0027] Once the triggered area is displayed, the edge computing node receives camera video streams or millimeter-wave radar point cloud data and uses a lightweight pose recognition model (such as human pose recognition algorithms known in the field, such as MediaPipe or its equivalents) to extract skeletal key points.

[0028] Video stream quality assessment method: The video stream quality is determined by comprehensively evaluating at least one of the following indicators: image brightness standard deviation (below 5 is considered too dark), human key point detection confidence (below 0.6 is considered occlusion or blur), and motion blur degree (assessed by gradient variance; below a preset threshold is considered blurry). When any indicator falls below the corresponding preset threshold, automatic switching to point cloud data collected by a millimeter-wave radar sensor for attitude recognition is triggered, and the switching process is imperceptible to the user. The millimeter-wave radar sensor is configured as a backup attitude recognition source and is automatically activated only when the camera video stream quality is insufficient.

[0029] Millimeter-wave radar point cloud data processing: The millimeter-wave radar sensor transmits frequency-modulated continuous waves and receives human body echo signals. Range FFT and Doppler FFT processing are then performed to generate raw point cloud data. The DBSCAN clustering algorithm is used to segment the human body point cloud, and then a lightweight PointNet model (three-layer convolution + global pooling) is used to extract joint coordinates, outputting at least the 3D coordinates of the shoulder, elbow, wrist, hip, knee, and ankle joints. The above algorithm parameters can be adjusted according to the actual hardware environment and do not constitute a limitation of this invention.

[0030] Privacy protection mechanism: After extracting key points, the original video frames and radar point cloud data are immediately overwritten and destroyed in memory (e.g., following the NIST SP 800-88 standard) to ensure that the original data is not persistently stored or uploaded to the cloud.

[0031] Audio intervention logic: Based on the posture recognition results, determine the risk of the action (such as excessive knee valgus angle), and combine it with the terminal status to bypass the stationary terminal and send the intervention command to the user's Bluetooth headset in milliseconds.

[0032] Example 4: Diverse Display Terminals and Subtitle Routing

[0033] This embodiment supports adaptation to various display terminals, including fixed interactive displays, smart glasses display terminals, and holographic projection devices, enabling seamless cross-regional following of virtual guidance screens.

[0034] Subtitle routing mechanism: When no audio terminal is detected being worn by the user, the system automatically converts the guidance information into high-contrast subtitles and overlays them on the user's current display device in real time, ensuring that the guidance information is 100% reached.

[0035] Example 5: A dual intervention mechanism based on real-time heart rate and terminal status

[0036] After user authorization, the system obtains the user's physiological indicators through the health data interface and generates personalized exercise safety parameters (such as heart rate safety threshold and movement safety angle).

[0037] Heart rate data validity verification: Real-time heart rate data is only accepted when the smart bracelet is in "wearing mode"; if the bracelet is idle, the data is ignored to avoid invalid intervention.

[0038] The intervention guidance information is only used for exercise load and risk warning, and does not involve medical diagnosis and treatment, strictly complying with the requirements for non-medical applications.

[0039] Industrial applicability

[0040] This invention can be widely applied to unattended gyms, community fitness centers, hotel gyms, and other similar locations. The UWB base station, camera, Wi-Fi AP, millimeter-wave radar sensor, edge computing server, and other components are all commercially available and mature components. The privacy protection scheme and posture recognition algorithm are common technologies in this field, possessing extremely high industrial practical value and commercial application prospects.

Claims

1. An intelligent fitness follow-up guidance system based on terminal status determination, characterized in that, The system includes: A communication module for establishing a wireless connection with at least one heterogeneous smart terminal carried by the user; The sensor assembly includes at least one of a visual sensor, a wireless positioning base station, and a millimeter-wave radar sensor deployed in the fitness space; the sensor assembly is connected to a local edge computing node via wired or wireless communication for transmitting collected video data and / or point cloud data to the edge computing node; Memory, which stores computer programs; The processor, connected to the communication module, sensor components, and edge computing node via a bus, electrical connection, or wireless communication, is configured to perform the following functions when executing the computer program: - (1) The communication module detects and determines the current connection status and physical wearing status of the heterogeneous smart terminal carried by the target user in real time. The physical wearing status includes the wearing status and the static placement status after being removed from the human body. The processor determines the physical wearing status of the heterogeneous smart terminal by comparing the inertial data rhythm of the heterogeneous smart terminal with the human body movement rhythm and analyzing the relative spatial relationship between the terminal and key points of the human body. - (2) Acquire at least two different types of positioning source data collected by the sensor component, and perform fusion calculation to output the location coordinates of the target user in real time; - (3) In response to the determination result of the physical wearing state, reduce the weight of the positioning source data provided by the heterogeneous smart terminal in the fusion calculation, and correspondingly increase the weight of other effective positioning source data; the dynamic adjustment of weight is based primarily on the determination result of the physical wearing state. - (4) Based on the adjusted location coordinates, maintain the continuous binding between the target user's identity and location coordinates.

2. The system according to claim 1, characterized in that, The processor is also configured to perform the following functions: - (5) Based on the continuously bound location coordinates, dynamically trigger the display device in the current area of ​​the target user so that the virtual guidance screen can seamlessly switch across areas as the user moves; - (6) Receive and process user motion data collected by the sensor components through the edge computing node, and identify the real-time motion posture of the target user; - (7) Generate corresponding guidance information based at least on the recognition result of the real-time motion posture, and according to the physical wearing state, send the guidance information to the target audio terminal for playback or convert it into subtitles for display on the display device.

3. The system according to claim 1, characterized in that, The different types of positioning source data include at least three of the following: ultra-wideband (UWB) positioning data, visual positioning data, Bluetooth positioning data, Wi-Fi positioning data, inertial navigation data, and millimeter-wave radar positioning data; the Wi-Fi positioning data includes signal strength index (RSSI) or round-trip time (RTT) data collected after the user connects to the gym's wireless access point (AP).

4. The system according to claim 1, characterized in that, The fusion calculation employs the extended Kalman filter algorithm; the processor dynamically adjusts the weights by adjusting the observation noise covariance matrix of the extended Kalman filter.

5. The system according to claim 1, characterized in that, The processor determines the physical wearing state by: comparing the inertial data rhythm of the heterogeneous smart terminal with the human motion rhythm and analyzing the relative spatial relationship between the terminal and key points of the human body to determine whether the heterogeneous smart terminal is currently in a wearing state or a static placement state detached from the human body; wherein, the inertial data rhythm includes the frequency domain energy distribution characteristics of the acceleration waveform.

6. The system according to claim 1, characterized in that, The user motion data includes video data collected by a camera fixed in the fitness space, and / or point cloud data collected by the millimeter-wave radar sensor; the edge computing node is also configured to automatically switch to using the point cloud data collected by the millimeter-wave radar sensor for attitude recognition when the quality of the video stream collected by the camera is detected to be lower than a preset threshold, and the switching process is imperceptible to the user; wherein, the millimeter-wave radar sensor is configured as a backup attitude recognition source, and is automatically activated only when the quality of the video stream collected by the camera is lower than the preset threshold.

7. The system according to claim 6, characterized in that, After extracting key point data of the human skeleton, the edge computing node does not persistently store the original video data and original point cloud data, but only performs an overwrite and destruction operation immediately after processing in the memory buffer.

8. The system according to claim 1, characterized in that, The processor is further configured to: first detect whether the audio smart terminal bound to the target user is being worn; if at least one audio smart terminal is detected to be being worn, then the guidance information is directed to that audio smart terminal for playback; if all associated audio smart terminals are not being worn, then the guidance information is converted into subtitle text and displayed in real time through the display device.

9. The system according to claim 1, characterized in that, The system also includes a health data interface, which is used to obtain the target user's health and physiological index data after user authorization, and generate personalized sports safety parameters; the heterogeneous smart terminal includes a smart bracelet or smartwatch, which is used to collect the user's real-time heart rate data; The processor is further configured to: verify the validity of the real-time heart rate data based on the determination result of the physical wearing state; when it is determined that the smart bracelet or smartwatch is in a wearing state, accept the real-time heart rate data and combine it with the personalized sports safety parameters to generate intervention guidance information; the intervention guidance information is limited to adjustment suggestions related to sports safety such as exercise load and movement posture, and does not contain any medical diagnosis conclusions or treatment suggestions.

10. The system according to claim 1, characterized in that, The display device includes at least one of the following: a distributed interactive display screen, a smart glasses display terminal worn by the user, and a holographic projection device deployed in space.

11. An intelligent fitness following guidance method based on terminal status determination, applied to the system described in any one of claims 1-10, characterized in that, Includes the following steps: - S1. Real-time detection of the physical wearing status of heterogeneous smart terminals carried by users, the physical wearing status including the wearing status and the static placement status after being removed from the human body; - S2. Acquire at least three different types of positioning source data for multi-source fusion, dynamically adjust the weight of each positioning source data based on the physical wearing state, and output the user's location coordinates in real time; - S3. Based on the adjusted location coordinates, maintain the continuous binding between the user's identity and location coordinates; - S4. Based on the continuously bound location coordinates, dynamically trigger the display device in the area to achieve cross-area following and switching of the virtual guidance screen; - S5. Process the collected user motion data on the local edge computing node, identify the user's real-time motion posture, and immediately destroy the raw video data and radar point cloud data after feature extraction; - S6. Generate guidance information based at least on real-time motion posture, and send it to the user's audio terminal for playback or convert it into subtitles according to the physical wearing state.

12. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in claim 11.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method of claim 11.

Images