AI-based exercise evaluation system
By integrating multimodal sensors and AI assessments into a flexible band worn on the thigh, the problem of unstable wrist measurements was solved, enabling accurate motion assessment and personalized feedback, thus improving motion quality and device lifespan.
Patent Information
- Application Number
- CN202511360750.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-23
- Publication Date
- 2026-02-13
AI Technical Summary
Existing motion bands primarily measure at the wrist, resulting in unstable signal strength, large data fluctuations, and an inability to provide accurate motion assessment and personalized feedback. Furthermore, they lack mature AI assessment algorithms.
Design an AI-based motion assessment system that uses a flexible band worn on the thigh, integrates multimodal sensors and a core processing module, incorporates self-healing elastomer materials, uses an aluminum alloy shell, and integrates optical sensors, MEMS inertial sensors, impedance detection electrodes, etc. It builds a dynamic knowledge base through deep learning to provide personalized motion suggestions.
It achieves improved signal strength and accuracy when worn on the thigh, extends device lifespan, provides multi-dimensional dynamic training curves and personalized assessments, and improves exercise quality and safety.
Smart Images
Figure CN121512487A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of smart wearable device technology, and particularly relates to an AI-based exercise assessment system. It is suitable for real-time training monitoring and personalized feedback for people who engage in lower limb exercises such as running and cycling. Background Technology
[0002] Current development of motion bands primarily focuses on measuring parameters at the wrist. However, the wrist has relatively low blood flow, and arm swing during movement causes significant positional shifts, affecting sensor performance and resulting in highly volatile data. There is still room for improvement. Considering the higher blood flow in the thigh, which significantly improves signal strength, and the stability of the thigh muscles during movement, measurement is easier. To address the issue of data accuracy and facilitate more scientific and rational AI evaluation of relevant data, a flexible motion band for thigh measurement is designed. Its function is to detect the post-training status and assess injuries in athletes / fitness enthusiasts (including runners and cyclists), enabling personalized and rational exercise plans.
[0003] Currently, exercise rings are widely used in daily exercise and professional training, providing basic measurement of fundamental human movement parameters such as heart rate, blood oxygen saturation, cadence, and stride length. However, they lack assessment and feedback functions, meaning users cannot evaluate the effectiveness of their training based on these parameters and their actual physical condition. In contrast, exercise rings with AI assessment capabilities can provide users with optimal training plan references, fundamentally helping to improve exercise quality. Therefore, exercise systems with assessment functions have significant market value and are expected to surpass similar products currently on the market.
[0004] Current AI technologies are relatively mature, capable of objectively evaluating specific situations through database construction and large models, and automating data processing through deep exploration. However, mainstream AI models, including Deepseek and Chatgpt, lack readily available databases of exercise training effectiveness for the general public, as well as mature exercise evaluation algorithms. Therefore, AI has blind spots in exercise evaluation. Summary of the Invention
[0005] Purpose of the Invention: The purpose of this invention is to provide an AI-based exercise assessment system. Empowered by modern AI, it can provide personalized and targeted opinions and suggestions for exercise training; building upon existing fitness trackers, it offers more professional and accurate measurement methods, thereby providing references for exercise training for runners, cyclists, and other sports enthusiasts.
[0006] Technical Solution: The present invention provides an AI-based motion assessment system, comprising a flexible ring and a shell; the shell contains a multimodal sensing module, a core processing module, and an energy module; the multimodal sensing module collects human physiological signals and motion data and transmits them to the core processing module, which interacts with an external AI module via wireless communication; the energy module supplies power to the system; the surface of the shell is provided with an interactive module; the interactive module includes a display screen and a haptic feedback unit.
[0007] Furthermore, the flexible loop adopts a curvature gradient structure that conforms to the human leg, with a curvature radius R satisfying: 40mm <= R <= 50mm; and the flexible loop is made of a stretchable hybrid flexible material, with a stretchable dimension L satisfying: 40cm <= L <= 55cm, and is designed with a Velcro fastening method. For the main loop of the system, considering the frequent muscle relaxation and contraction of the legs during exercise, the loop material is easily stretched. Preferably, a self-healing elastomer is used as the flexible material, which can self-heal at certain temperatures, greatly extending the material's service life, reducing strap replacements, and promoting product environmental friendliness. A nylon Velcro fastening method is designed at the end of the loop (10cm-12cm), which can improve signal strength and monitoring accuracy while ensuring the device conforms to the user's leg.
[0008] Furthermore, the flexible ring is an integral design with a material of the same size that fits the shell at its center for embedding into the shell. The base has a circular hole with a radius of 4.8-5.2m to allow the multimodal sensing module to fit the skin. The material can withstand tensile forces of more than 6N.
[0009] Preferably, the outer shell is made of aluminum alloy, which is relatively inexpensive and strong enough to withstand the damage caused by daily bumps and knocks. It adopts a rectangular shape of 2cm*4cm*1.5cm with rounded corners for a more natural transition.
[0010] Furthermore, the multimodal sensing module includes an optical sensor array, a MEMS inertial sensor, an impedance detection electrode, a blood lactate sensor, a blood oxygen sensor, and a heart rate sensor; the multimodal sensing module jointly monitors relevant human data to form a multidimensional dynamic training curve.
[0011] Furthermore, the optical sensor array emits light at a frequency between 530nm and 850nm, with a light source-detector spacing of 1-3mm. It employs a tilted lens design to receive reflected light signals from capillary blood flow, with an accuracy of ±1pbm.
[0012] Preferably, the optical sensor array emits light at a frequency between 530nm (green light) and 920nm (infrared light), primarily for identifying hemoglobin (strong absorption peak at approximately 530nm) and blood oxygen saturation (strong absorption peak at approximately 910nm). The distance between the light source and the detector is 1-3mm to avoid excessive signal noise due to insufficient distance. A tilted lens design is employed to amplify the light signal and assist the parameter sensors in precise operation.
[0013] Furthermore, the MEMS inertial sensor includes a triaxial accelerometer with a range of ±20g and a triaxial gyroscope with a range of ±2000° / s, and is equipped with a high-frequency sampling frequency of >=150Hz to capture motion characteristics, and is equipped with a temperature compensation circuit to maintain the sensor's operating temperature at 40°-50°.
[0014] Furthermore, the impedance of the impedance detection electrode is 8kΩ. <R<10kΩ。
[0015] Furthermore, the blood lactate sensor, blood oxygen sensor, and heart rate sensor determine the corresponding substance levels by monitoring the wavelength and intensity of absorbed and reflected light, with an accuracy of ±2%.
[0016] Furthermore, the core processing module uses the nRF52840 system-on-a-chip, which has a built-in 12-bit / 200ksps ADC, adopts Bluetooth communication, and achieves a maximum RF output power of +8dBm by loading a Bluetooth 5.0 compatible RF and S140 protocol stack.
[0017] Furthermore, the AI module adopts Deepseek-style interactive and network-connected AI, acquires open-source datasets, establishes corresponding databases, determines the user's exercise level through changes in the user's personal sensor data, and provides personalized solutions through an expert knowledge base.
[0018] Preferably, the interactive module adopts an external touch structure made of a 2cm*4cm OLED screen and a linear resonant motor, and has a simple UI interface that supports human-computer interaction.
[0019] Preferably, the energy module uses a lithium polymer battery with a PCB copper layer for heat dissipation on its surface. The battery is located on the side of the motherboard, occupying a small storage space. It is wired charged via a Type-C interface, and an NTC thermistor is placed near the battery to control the battery temperature at 40℃-60℃, the charging voltage at 3V-5V, and the charging current at 50mA-100mA to ensure safety.
[0020] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:
[0021] This invention achieves a revolutionary wearable fitness device design, worn on the thigh rather than the wrist, effectively reducing movement displacement, enhancing signal strength, and achieving signal stability and accuracy. Simultaneously, the strap is designed with a self-healing elastomer material, extending its lifespan and aligning with current environmental themes.
[0022] This invention enables multimodal sensor collaboration, achieving exercise physiological monitoring that surpasses consumer-grade wristbands, and can jointly output multidimensional dynamic training curves, integrating and analyzing parameters such as minimum blood oxygen value, lactate change slope, heart rate recovery speed, and sweating rate.
[0023] This invention proposes combining AI technology with motion detection equipment to construct a dynamic knowledge base and provide users with four levels of quantitative assessment for reference. It also enables personalized and targeted assessments of users' exercise status and recovery suggestions.
[0024] This invention uses the nRF52840 chip as the core chip, which can realize Bluetooth 5.0 anti-interference data transmission and 12-bit data operation, greatly ensuring the integrity and stability of the data. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the structure of the present invention;
[0026] Figure 2 This is a schematic diagram of the internal modules of the outer casing;
[0027] Figure 3 This is a side view of the outer casing;
[0028] Figure 4 This is a flowchart illustrating the overall system framework and the overall human-computer interaction process.
[0029] Figure 5 This is a hardware structure block diagram;
[0030] Figure 6 This is a block diagram of the terminal software structure.
[0031] The components include a flexible ring belt 101, a housing 102, a multimodal sensing module 201, a core processing module 202, an energy module 205, a display screen 204a, a haptic feedback unit 204b, an optical sensor array 201a, a MEMS inertial sensor 201b, an impedance detection electrode 201c, a blood lactate sensor 201d, a blood oxygen sensor 201e, and a heart rate sensor 201f. Detailed Implementation
[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings.
[0033] System hardware implementation:
[0034] 1. Make a ring mold that conforms to the curvature radius of the human thigh (R=45±5mm), and after flattening, its three dimensions meet the geometric rectangle of 1.5cm*18cm*0.2cm with an error of no more than 1mm. Preheat the mold to 50℃, put in the self-healing elastomer preform and pressurize it to 10MPa to make it quickly form. Hold the pressure for at least 5 minutes and then demold with cold water.
[0035] 2. Coat one end of the self-healing elastomer with PU adhesive and heat-press nylon hook and loop fastener material. The nylon material has been cut into a standard geometric shape of 1.8cm*10cm*0.5cm, and a hole of 1cm*1.6cm*0.3cm is chiseled at one end to embed the self-healing elastomer, increasing its contact area and obtaining a stronger material adhesion effect.
[0036] 3. Referring to step 1, use a mold to make a 2cm*4cm*1cm self-healing elastomer hollow rounded rectangle with a reserved thickness of 0.1±0.05mm. Open a 2cm*4cm area at the top for displaying screen information and human-computer interaction, and open a circular area with a radius of 0.5cm at the bottom to expose the Fresnel lens.
[0037] 4. Produce aluminum alloy shells of the same shape according to the parameters of rounded rectangles, with a thickness not exceeding 1mm to ensure reasonable material usage and lightweight design.
[0038] 5. Etch the corresponding circuits on the motherboard that matches the size of the casing. Solder a linear resonant motor on one side of the bottom of the motherboard to realize tactile feedback. Next to the motor, there is a rechargeable lithium battery with a plastic sleeve, which is connected to the motherboard by hot melt adhesive. At the same time, an NTC thermistor is soldered to the bottom. The resistance can be adjusted by sensing the charging temperature of the lithium battery to ensure temperature control.
[0039] 6. The nRF52840 system-on-a-chip is mounted on the top center of the motherboard described in step 5. The chip itself contains powerful Bluetooth functionality and large flash memory, and a MEMS inertial sensor is set on the top for monitoring user activities.
[0040] 7. A circular opening (5mm radius) is made at the bottom of the housing as described in step 4. A Fresnel lens of the same size is placed using a supporting step, and the edge of the opening is covered with a flexible pressure strip. The strip is then tightened diagonally. A light source and an optical sensor array are positioned directly above the lens. Next to the optical sensors are impedance detection electrodes, a blood lactate sensor, a blood oxygen sensor, and a heart rate sensor, all integrated into a small module next to the light source to receive signals. All of these sensors are connected to the nRF52840 system-on-a-chip via flexible cables.
[0041] 8. The connection between each sensor and the chip is as follows: Since the optical sensor array outputs analog signals, it is connected to the ADC interface, and an on-chip OPAMP is used to construct a TIA (transimpedance amplifier) to amplify the optical signal; the MEMS inertial sensor is configured in FIFO mode for batch reading, reducing CPU interrupt frequency, and is connected to the chip's SPI interface; the impedance detection electrode generates a 50kHz-100kHz sine wave through the nRF52840 DAC, and the on-chip ADC and software phase-locked loop (PLL) are used to calculate the phase / amplitude amplified signal; the blood lactate sensor, which generates a weak current analog signal, is also connected to the nRF52840 ADC, meaning the nRF52840 needs to implement multi-channel ADC communication, and an LMP91000AFE low-noise current amplifier is added to amplify the signal; the blood oxygen sensor is integrated with the heart rate sensor and directly connected to the chip via IIC communication, using the nRF52840's FPU to accelerate the heart rate variability (HRV) algorithm to optimize data.
[0042] 9. A thin copper foil is provided on the upper and lower sides of the motherboard described in step 5 to protect and isolate different components.
[0043] Select an OLED screen with a rounded rectangle size of approximately 2cm*4cm and a resolution of 128*32 pixels. Connect its pins to VCC, GND, and SPI respectively, and then apply Dow Corning 732 waterproof adhesive and SMT adhesive to the joints for fixation.
[0044] System software implementation:
[0045] 1. By collecting data from sports databases of the General Administration of Sport of China or other institutions, data cleaning is performed to reduce invalid information, forming a new sports type database; by collecting data from national health system databases, national health and medical databases, and other health and medical databases, data cleaning is performed to reduce invalid information, forming a new expert knowledge database; and by using GPU clusters, data parallelism, and other methods for machine learning and precision training, a large AI model is formed.
[0046] 2. Receive sensor data via Bluetooth and use API integration and RAG technology to search for user and database data.
[0047] 3. AI quantifies physiological and technical indicators and employs a weighted comprehensive scoring mechanism to provide a comprehensive score. The specific scheme is as follows: exercise intensity (heart rate) accounts for 0.3, metabolic stress (blood lactate) accounts for 0.3, oxygen supply stress (blood oxygen) accounts for 0.2, step length and cadence (pace) account for 0.1, and heat stress (sweating) accounts for 0.1. The "normal reference value" and "maximum deviation value" of each indicator are defined through standardization (based on physiological common sense or individual baseline): for example, heart rate: normal reference value = resting heart rate (e.g., 60-80 bpm), maximum deviation value = maximum heart rate (e.g., 220 - age). Then, the actual difference of each indicator is standardized to the range of 0-100 using the formula:
[0048]
[0049] After standardization, each indicator is set as follows: 0 represents no deviation (normal state), and 100 represents the maximum deviation (extreme stress state). The overall difference range is then calculated based on the weights of each item, ranging from 0 to 100 (0 = completely normal, 100 = maximum deviation). If any data is abnormal, an immediate warning should be issued to the user to stop exercising.
[0050] 4. The AI model provides corresponding motion ratings based on interval values, specifically set as follows:
[0051] Range values:
[0052] Grade A: Overall difference ≤ 25 (0-25 range);
[0053] Category B: 25 < overall difference ≤ 50 (range of 25-50);
[0054] Grade C: 50 < overall difference ≤ 75 (50-75 range);
[0055] Category D: Overall difference > 75 (75-100 range);
[0056] Sports rating:
[0057] Grade A (Low Intensity / Recovery): Small overall difference, easy exercise, mainly used for recovery or basic aerobic training.
[0058] Grade B (Moderate Intensity / Aerobic): Overall, the difference is moderate, suitable for aerobic endurance training.
[0059] Grade C (High Intensity / Threshold): The overall difference is large, corresponding to lactate threshold or high-intensity interval training.
[0060] Grade D (Extremely High Intensity / Anaerobic): The largest overall difference, indicating anaerobic or maximum effort training / emergency situations requiring training to be stopped.
[0061] 5. The AI big data model uses NLP model text encoding to generate corresponding data (individual user data, standardized data), uses diffusion model to generate curves, and provides corresponding exercise training suggestions, real-time suggestions, and medical suggestions based on expert knowledge database.
[0062] The data analyzed by the AI model is transmitted to the motion system via Bluetooth. Users can view the corresponding data and images by controlling the "exercise suggestions" and "exercise curves" on the UI interface.
Claims
1. An AI-based motion evaluation system, characterized by, It comprises a flexible ring belt (101) and a shell (102); the shell (102) is internally provided with a multi-modal sensing module (201), a core processing module (202) and an energy module (205); the multi-modal sensing module (201) collects human physiological signals and motion data and transmits them to the core processing module (202), the core processing module (202) interacts with the external AI module (203) through wireless communication; the energy module (205) supplies power for the system; the surface of the shell (102) is provided with an interactive module (204); the interactive module (204) comprises a display screen (204a) and a touch feedback unit (204b).
2. The AI-based motion assessment system of claim 1, wherein, The flexible ring belt (101) adopts a curvature gradient structure that fits the human leg, and the curvature radius R satisfies: 40mm <= R <= 50mm; and the flexible ring belt (101) is made of a stretchable hybrid flexible material, and the stretchable size L satisfies: 40cm <= L <= 55cm, and is designed in a magic tape buckle tightening mode.
3. The AI-based motion assessment system of claim 1, wherein, The flexible ring belt (101) is designed in one piece, and a same material as the size of the shell (102) is arranged at the length center of the flexible ring belt (101) for embedding the shell (102), and a circular hole with a radius of 4.8-5.2m is opened in the base to make the multi-modal sensing module (201) fit the skin, and the material can withstand a tensile force of more than 6N.
4. The AI-based motion assessment system of claim 1, wherein, The multi-modal sensing module (201) comprises an optical sensor array (201a), a MEMS inertial sensor (201b), an impedance detection electrode (201c), a blood lactic acid sensor (201d), a blood oxygen sensor (201e) and a heart rate sensor (201f); the multi-modal sensing module (201) jointly monitors the corresponding data of the human body to form a multi-dimensional dynamic training curve.
5. The AI-based motion assessment system of claim 4, wherein, The optical sensor array (201a) emits light with a frequency of 530nm-850nm, the light source-detector distance is 1-3mm, an inclined lens design is adopted to receive the reflected light signal of capillary blood flow, and the accuracy is ±1pbm.
6. The AI-based motion assessment system of claim 4, wherein, The MEMS inertial sensor (201b) comprises a three-axis accelerometer with a range of +-20g and a three-axis gyroscope with a range of ±2000° / s, and is provided with a high-frequency sampling of >=150hz to capture the action characteristics, and is matched with a temperature compensation circuit to maintain the working temperature of the sensor at 40°-50°.
7. The AI-based motion assessment system of claim 4, wherein, The impedance of the impedance detection electrode (201c) is 8kΩ<R<10kΩ.
8. The AI-based motion assessment system of claim 4, wherein, The blood lactic acid sensor (201d), the blood oxygen sensor (201e) and the heart rate sensor (201f) determine the corresponding substance level by monitoring the absorption light, reflected light wavelength and intensity, and the accuracy is +-2%.
9. The AI-based motion assessment system of claim 1, wherein, The core processing module (202) selects an nRF52840 system chip, which is provided with a 12-bit / 200ksps ADC, adopts a Bluetooth communication mode, loads a Bluetooth 5.0 compatible radio frequency, and realizes a maximum radio frequency output power of +8dBm through an s140 protocol stack.
10. The AI-based motion assessment system of claim 1, wherein, The AI module (203) adopts deepseek type interactive and networkable AI, acquires an open source data set, establishes a corresponding database, judges a user motion level through changes in user personal sensor data, and gives a personalized solution through an expert knowledge base.