Intelligent sports testing system for physical education
The intelligent sports testing system, which integrates deep learning for human pose estimation and multimodal sensing, solves the problems of low efficiency, large errors, cheating, and data silos in traditional sports testing methods, and achieves efficient and accurate sports testing and personalized exercise suggestions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGXI INST OF FASHION TECH
- Filing Date
- 2026-03-06
- Publication Date
- 2026-05-29
AI Technical Summary
Traditional sports testing methods are inefficient, costly in terms of manpower, prone to errors in human judgment, susceptible to cheating, and suffer from severe data silos. Existing electronic devices are difficult to popularize in large-scale campus settings.
An intelligent sports testing system based on deep learning human pose estimation and multimodal sensing fusion is adopted. Through the "end-edge-cloud" collaborative architecture, it integrates data acquisition, edge computing, data transmission and cloud management modules to achieve high-precision and cheat-proof intelligent testing.
It improves testing efficiency, eliminates human error, prevents cheating, creates continuous student physical health records, and supports big data analysis and personalized exercise programs.
Smart Images
Figure CN122116479A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sports information technology, specifically to an intelligent sports testing system for sports teaching. Background Technology
[0002] With the advancement of education evaluation system reform, improving the physical health of teenagers has increasingly become an important educational evaluation indicator. The weight of physical education test scores in middle school entrance examinations, college entrance examinations, and even higher education is increasing year by year. Traditional physical education teaching and testing models mainly rely on manual organization, using simple tools such as stopwatches, measuring tapes, and mechanical counters for measurement. This traditional model has exposed many problems when facing the demands of large-scale, high-frequency, and routine physical education testing.
[0003] First, low testing efficiency and high labor costs are the main bottlenecks hindering the routine monitoring of school physical education. During the annual physical fitness tests or physical education entrance examinations, schools often need to allocate a large number of teachers to act as referees, spending several weeks to complete the testing of all students. This not only encroaches on normal teaching time but also makes it difficult to implement routine, process-based evaluation.
[0004] Secondly, the subjective errors and inconsistent standards of human judging seriously affect the fairness of the examination. In exercise-based events such as pull-ups and sit-ups, the judgment of whether the chin is over the bar, whether the arms are straight, and whether the shoulders and back touch the mat are easily affected by the judge's visual fatigue, subjective differences in standards, and stance. In long-distance running events, manual timing not only has limited accuracy but also makes it difficult to keep track of the lap counts of multiple students simultaneously, easily leading to omissions or errors in recording.
[0005] Furthermore, cheating methods are constantly evolving, and traditional prevention measures are inadequate. In physical education exams, which involve college admissions, cheating behaviors such as proxy test-taking, cutting corners in long-distance running (taking shortcuts), and using improper movements to fraudulently increase the score occur frequently. Traditional identity verification relies solely on manual checks before the exam, which cannot monitor personnel substitutions during the testing process in real time.
[0006] Finally, the problem of data silos is severe. Data generated by traditional tests is mostly recorded in paper forms, and the subsequent digitization process is enormous and prone to errors. This data is often fragmented, making it difficult to form continuous student physical health records, to discover patterns and weaknesses in students' physical development through big data analysis, and to provide precise personalized exercise plans for physical education.
[0007] Although some electronic sports testing devices have appeared on the market, they all have varying degrees of technical limitations: Infrared blocking technology: Existing standing long jump or sit-up testing devices mostly use infrared photoelectric switches. While the principle is simple, its anti-interference capability is poor. For example, in the standing long jump, if the corner of clothing or an arm lands before the heel, blocking the beam, it can lead to misjudgment of the result; in sit-ups, it is impossible to determine whether a student violated regulations by using momentum (such as not covering their head).
[0008] Wearable sensor technology: Although smart bracelets or straps based on accelerometers and gyroscopes can provide high motion capture accuracy, the frequent distribution, recycling, charging, disinfection, and wearing comfort issues make them difficult to popularize and promote in daily teaching in large-scale campus scenarios.
[0009] Traditional computer vision technologies: Early vision solutions were mostly based on color recognition or simple background subtraction methods, which are extremely sensitive to changes in ambient lighting. Recognition rates drop sharply in bright outdoor light, shadows, or backlighting indoors. Furthermore, traditional algorithms are prone to losing targets in scenarios with multiple people occluding objects, failing to meet the needs of simultaneous testing by multiple users.
[0010] To address the aforementioned issues, this invention proposes an intelligent sports testing system based on deep learning human pose estimation and multimodal sensing fusion. By constructing an "edge-cloud" collaborative architecture, it achieves high-precision, cheat-proof intelligent sports testing. Summary of the Invention
[0011] This invention discloses an intelligent sports testing system for physical education teaching, comprising a data acquisition module, an edge computing module, a data transmission module, a cloud management module, and a monitoring module; The data acquisition module is used to collect multimodal data from the test site, specifically: acquiring motion video streams from the test site through a high-definition camera, acquiring depth information from the test site through a depth sensor, and acquiring the identity information of the test participants through a radio frequency identification reader / writer. The edge computing module is connected to the data acquisition module and includes an inference unit and an action logic analysis unit; The inference unit is used to process the acquired video stream data and extract the coordinates of key points of the human skeleton; The action logic analysis unit is used to determine the action type, count, and standard degree based on the temporal changes of key points according to a preset finite state machine model. The data transmission module is used to upload the test data, violation keyframe images, and identity verification information processed by the edge computing module to the cloud. The cloud management module is used to store students' physical health records, perform statistical analysis on the uploaded test data, and generate personalized exercise plans based on the analysis results. The monitoring module is integrated into the edge computing module and the cloud management module, and is used to monitor in real time for proxy test-taking, violation of action rules, interference from non-test subjects, and abnormal behavior of equipment.
[0012] Preferably, the action logic analysis unit performs the following steps when conducting a pull-up test: Construct a finite state machine that includes the hanging state, rising state, over-bar state, falling state, and reset state; Real-time calculation of the subject's elbow joint angle and the relative position of the key point of the jaw to the horizontal bar; A valid movement is counted only when the subject starts from the hanging position, goes through the rising position, reaches the over-bar position, and then goes through the descending position back to the hanging position. If the player starts the next movement before reaching the over-bar position or returning to the hanging position during the movement, the movement will not be counted and a voice correction prompt will be triggered.
[0013] Preferably, in the standing long jump test, the data acquisition module is equipped with a stereo camera, and the edge computing module performs the following ranging algorithm based on the data acquired by the stereo camera: Image pixel coordinates are mapped to physical plane coordinates based on the perspective transformation matrix; Human foreground is extracted using a background subtraction and depth thresholding segmentation algorithm; Track key points on the human foot to identify the moment of takeoff and landing; At the moment of landing, the vertical distance between the point closest to the takeoff line and the takeoff line among all valid ground contact points is calculated as the test score.
[0014] Preferably, the monitoring module employs a dual verification mechanism of facial recognition and RFID during middle- and long-distance running tests. RFID carpet antennas are set at the start and end points to record the time when the chip passes through. Facial recognition devices are installed at key nodes along the runway path to record the facial features of people passing by. The system compares the identity information recorded by RFID with the facial information captured at key points. If the two are inconsistent or the key point record is missing, it is determined to be a case of impersonation or corner cutting violation, and the score is invalid.
[0015] Preferably, the monitoring module further includes a liveness detection unit and a multi-target attribution analysis unit; The liveness detection unit distinguishes between real people and photo / video attacks by analyzing facial texture and micro-expression features. The multi-target attribution analysis unit uses affinity field technology to classify the limbs in the image into different human body instances. When a non-subject's limb is detected entering the test area and applying external force to the subject, it is determined to be illegal assistance.
[0016] Preferably, the cloud management module performs statistical analysis on the uploaded test data and generates personalized exercise prescriptions based on the analysis results, specifically including: Receive students' historical test data, body shape data, and physiological function data; Use regression analysis or machine learning clustering algorithms to assess students' physical weaknesses; The system matches relevant training movements, intensities, and frequencies from a pre-set exercise knowledge base to generate personalized after-school workout plans.
[0017] A smart physical education testing method includes the following steps: Step S1: Complete the test subject's identity login through facial recognition or RFID scanning; Step S2: Load the corresponding model and anti-cheating rules according to the test project; Step S3: Real-time acquisition of motion video stream, and extraction of human key points and state machine logic determination at the edge; Step S4: Real-time feedback of test results to the on-site display terminal, and voice announcement when violations occur; Step S5: After the test is completed, upload the final score, process data and evidence of violations to the cloud database; Step S6: The cloud platform updates the student's health record and pushes the academic report and exercise suggestions.
[0018] Compared with the prior art, the advantages of this invention are: Using non-contact machine vision technology, students do not need to wear any equipment and can be tested anytime, greatly improving the pass rate. It is suitable for large breaks or school-wide physical fitness tests.
[0019] The quantitative judgment algorithm based on skeletal key points eliminates the subjective errors of human judges. The system not only records "how much was done", but also judges "whether it was done correctly", such as accurately identifying and eliminating "half-range movements" in pull-ups.
[0020] By integrating facial recognition, RFID, and behavioral analysis technologies, the system monitors the entire process from identity verification to action execution, effectively preventing cheating behaviors such as impersonation, taking shortcuts, and using others' help, thus ensuring the fairness of the examination. Attached Figure Description
[0021] Figure 1 This is a system architecture diagram of an intelligent sports testing system for physical education proposed in this invention. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0023] This solution is an intelligent sports testing system for physical education teaching, including a data acquisition module, an edge computing module, a data transmission module, a cloud management module, and a monitoring module.
[0024] The data acquisition module collects multimodal data from the test site using a data acquisition device. Different data acquisition devices may be used for different test items and environments. For the standing long jump, a binocular stereo camera or a depth camera equipped with a ToF sensor will be deployed in the testing area as the data acquisition device. The installation height should be set to 2.5 to 3 meters, covering the area from the take-off zone to the landing zone (approximately 3.5 meters long and 1.5 meters wide). The binocular baseline distance should be set to 10-20 cm to ensure a depth accuracy of ±2 mm within the measurement range (0.5m-3.5m). The camera should also have a wide dynamic range to compensate for shadow interference under direct sunlight outdoors.
[0025] For pull-ups / sit-ups, a high frame rate camera should be placed in the testing area, with a resolution of at least 1920×1080 and a frame rate of 60fps or higher, to capture details of the movement. A low-distortion fixed-focus lens should be used, mounted at a 45-degree angle to the side or directly to the side to ensure complete imaging of the subject's head, arms, and torso.
[0026] For middle- and long-distance running tests, cameras and RFID readers are deployed at the start, finish, and curves of the track. The RFID readers are connected to ground-mounted antennas or side-mounted panel antennas to form a radio frequency sensing zone.
[0027] Edge computing modules are deployed in each test area (such as a horizontal bar area or a running track). The specific hardware carrier can be an edge computing box. This module connects to various sensors and is responsible for: Video stream decoding and preprocessing (denoising, distortion correction).
[0028] Loading and inference of deep learning models.
[0029] Local caching of test results ensures that data is not lost when the network is disconnected.
[0030] The system drives the on-site display screen via an HDMI interface to show real-time images and results.
[0031] The cloud management module is deployed on the school-level server or cloud server, and receives and aggregates data from various edge computing modules through the data transmission module. The database adopts a combination of "time series database (InfluxDB) + relational database (MySQL)" to store high-frequency sensor / video stream metadata and students' basic profile data, respectively.
[0032] The monitoring module is integrated into the edge computing module and the cloud management module to monitor in real time for proxy test-taking, violation of action rules, interference from non-test subjects, and abnormal equipment behavior.
[0033] Example 1: This embodiment specifically describes the method of using the system for standing long jump testing. The main challenge in standing long jump testing lies in accurately capturing the landing point and eliminating interference. This embodiment employs machine vision-based trajectory tracking and depth ranging technology, specifically: Before the test begins, the system performs perspective transformation and background modeling and calibration on the test area.
[0034] Perspective Transformation: Due to the camera's mounting angle, the image exhibits perspective distortion, with objects appearing larger than distant objects. Four reference points are set on the test mat. The system identifies the pixel coordinates of these four points and, combined with their known physical coordinates, calculates the homography matrix. All subsequent pixel coordinates are then mapped to their actual physical distances using this homography matrix.
[0035] Background modeling: A background model is built using Vibe (Visual Background Extractor) or Gaussian mixture model algorithm for subsequent foreground segmentation.
[0036] When a student enters the testing area, the system uses object detection algorithms (such as YOLOv8) to locate the human body area. Subsequently, a human pose estimation algorithm is used to extract key points of the feet: heel, toes, and ankle.
[0037] The system monitors the vertical velocity vector and horizontal position of key points on both feet in real time. When the vertical velocity vector exceeds a set threshold and both feet leave the ground simultaneously (abrupt change in depth information), it is determined to be a jump, and the position of the takeoff line is recorded.
[0038] During the aerial phase, the DeepSORT algorithm is used to continuously track the body's center of mass and key points on the feet to prevent ID loss caused by limb swinging.
[0039] After tracking key points on the human foot and identifying the moment of landing, the system needs to distinguish between "valid landing points" (usually the heel) and "invalid ground contact" (such as hands supporting the ground or sitting on the buttocks).
[0040] According to sports testing rules, the score is based on the closest point of contact with the ground from any part of the body. The system calculates in real time the distance from all ground-touching pixels (based on the difference between the depth map and the background) to the take-off line, and takes the minimum value as the landing point.
[0041] If the landing point is detected to have shifted significantly forward in a very short time (e.g., within 100ms) (which may be due to sensor noise or insect interference), it will be removed by time-series smoothing filtering.
[0042] The final score is the straight-line distance between the landing point and the take-off line. The system automatically deducts for take-off line violations (stepping on the line). If the toes of the foot exceed the take-off line threshold before take-off, a voice announcement will sound saying "Stepping on the line is a foul," and the score will be invalid.
[0043] Example 2: This embodiment specifically illustrates the method of using this system for pull-up / sit-up testing. This embodiment mainly explains how to quantify the accuracy of movement by constructing a finite state machine using skeletal key points.
[0044] For the pull-up test: The system defines the following states: Hanging preparation: Check if both arms are straight. Calculate the angle formed by the shoulder, elbow, and wrist. If this angle is greater than 160° and the body swing amplitude is less than the threshold, enter the hanging preparation state. The system voice prompts "Please begin".
[0045] Ascending phase: A decrease in the elbow joint angle and an increase in the vertical coordinate of the head are detected.
[0046] Passing the bar test: Detecting the relative position of the jaw key point and the bar. Specifically: Identifying the equation of the bar's straight line during image initialization. Real-time determination of whether the vertical coordinate of the jaw key point is greater than the vertical coordinate of the bar. Simultaneously, head posture needs to be detected to prevent students from cheating by excessively tilting their heads back.
[0047] Descent phase: Body descent is detected, and the elbow joint angle increases.
[0048] Reset complete: Return to the hanging position, and the angle formed by the shoulder, elbow, and wrist is greater than 160°.
[0049] Only after completing the entire closed loop of the above-mentioned preparation, ascent, clearance, descent, and return to the starting position will the pull-up count be incremented by 1. If the pull-up fails to clear the bar during the clearance phase and descends directly, it is judged as "failure to clear the bar"; if the pull-up is performed without fully extending the arms during the return to the starting position, it is judged as "not straight arms," and neither will be counted, and the specific reason for the violation will be announced.
[0050] For the sit-up test: The system defines the following states: Lying flat: The angle between the line connecting the shoulder and hip and the horizontal plane is less than 20°, and the shoulder is in contact with the mat.
[0051] Sit-up: Curl your torso upwards, increasing the angle of the shoulder-hip line.
[0052] Knee Touch / Vertical Plane: Detect whether the elbow key point touches or exceeds the vertical line of the knee key point, or whether the torso angle is greater than 80°.
[0053] Only after completing all the above stages will the sit-up count increment by 1. If the system detects that the wrist key point is significantly away from the ear during the sit-up, it is judged as a violation of "not holding the head"; if it detects a large vertical displacement of the hip at the moment of sit-up, it is judged as a violation of "hips off the ground".
[0054] Example 3: For middle- and long-distance running tests of 800 / 1000 meters, a single technology is insufficient to simultaneously meet the requirements of timing accuracy and anti-cheating measures. This embodiment employs a data acquisition solution that integrates RFID and vision, including: Students wear bib numbers with built-in UHF RFID chips. At the check-in area, students undergo facial recognition, and the system automatically reads the RFID information from the bib number, binding the "face ID" with the "RFID EPC code".
[0055] A carpet of antennas is laid at the start and finish lines to record crossing times down to the millisecond level. RFID has group reading capability, allowing it to handle dozens of students crossing the finish line simultaneously.
[0056] A "visual check-in device" is installed on the inside of the track curve. When students pass by, the camera performs a non-intrusive facial capture and re-identification.
[0057] Based on the above data collection scheme, each lap must include: RFID record at the start, facial recognition data at the curves, and RFID record at the finish line. Missing any one of these steps will result in a cut-off angle or a missed run.
[0058] Meanwhile, facial recognition is performed again at the destination. If the student information corresponding to the RFID tag read at the destination does not match the face captured by the camera, the system will issue an alarm indicating "identity abnormality".
[0059] In addition, the system calculates the pace for each lap in real time. If the time for a lap is far below the physical limit for that age group (e.g., 400 meters in less than 40 seconds), or deviates from the student's historical average pace by more than 3 standard deviations, the system automatically marks the result as "questionable" and prompts for manual review of the video.
[0060] Example 4: In addition to the anti-cheating scheme mentioned in Example 3, this scheme also includes other methods to address potential cheating in high-stakes exams, specifically including: Liveness detection: In response to attacks that use photos or videos to replace real people for testing, this method utilizes depth information from binocular or structured light cameras to directly filter out attacks on planar media (photos / screens) because they lack depth undulations.
[0061] Biometric consistency verification: During the test, the system continuously extracts Re-ID features from the subject in the image. If, during the test (such as during a pull-up break), the test subject leaves the screen and another person enters (i.e., the Re-ID feature vector distance is greater than the feature vector threshold), the system will immediately interrupt the test and issue an alarm to prevent "relay testing".
[0062] Non-intrusive monitoring: Utilizing PAFs (Part Affinity Fields) technology in the skeletal keypoint algorithm, all limbs in the image are categorized into different human instances. If, during a sit-up test, a "third hand" is detected reaching into the edge of the image to push the subject's back, or during a pull-up test, someone is seen lifting the legs, the system will identify that the limb does not belong to the subject, deem it "unauthorized assistance," interrupt the test, and issue an alarm.
[0063] Example 5: The cloud management module uses a time-series database (InfluxDB) to store unstructured motion process data (such as waveforms of each jump and joint angle curves) and a relational database (MySQL) to store student records.
[0064] Based on a large amount of historical data, a regression model or expert system is trained in the cloud management module.
[0065] The model takes data such as students' height, weight (BMI), vital capacity, standing long jump performance, and pull-up performance as input to identify students' weaknesses. For example, if a student has a normal BMI but a pull-up score of 0, it indicates insufficient upper body strength. The model then generates a personalized training plan. For example, it suggests "3 kneeling push-up sessions per week, 4 sets each time, 12 repetitions per set; supplemented by resistance band pull-up exercises."
[0066] Based on an intelligent sports testing system for physical education teaching, this invention also provides an intelligent sports testing method for physical education teaching, comprising the following steps: Step S1: Complete the test subject's identity login through facial recognition or RFID scanning.
[0067] Step S2: Load the corresponding model and anti-cheating rules according to the test project.
[0068] Step S3: Real-time acquisition of motion video streams, and extraction of human key points and state machine logic determination at the edge.
[0069] Step S4: Feedback the test results to the on-site display terminal in real time, and issue a voice broadcast when a violation occurs.
[0070] Step S5: After the test is completed, upload the final results, process data and evidence of violations to the cloud database.
[0071] Step S6: The cloud platform updates the student's health record and pushes the academic report and exercise suggestions.
[0072] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0073] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A smart physical education testing system for physical education teaching, characterized in that, It includes a data acquisition module, an edge computing module, a data transmission module, a cloud management module, and a monitoring module; The data acquisition module is used to collect multimodal data from the test site, specifically: acquiring motion video streams from the test site through a high-definition camera, acquiring depth information from the test site through a depth sensor, and acquiring the identity information of the test participants through a radio frequency identification reader / writer. The edge computing module is connected to the data acquisition module and includes an inference unit and an action logic analysis unit; The inference unit is used to process the acquired video stream data and extract the coordinates of key points of the human skeleton; The action logic analysis unit is used to determine the action type, count, and standard degree based on the temporal changes of key points according to a preset finite state machine model. The data transmission module is used to upload the test data, violation keyframe images, and identity verification information processed by the edge computing module to the cloud. The cloud management module is used to store students' physical health records, perform statistical analysis on the uploaded test data, and generate personalized exercise plans based on the analysis results. The monitoring module is integrated into the edge computing module and the cloud management module, and is used to monitor in real time for proxy test-taking, violation of action rules, interference from non-test subjects, and abnormal behavior of equipment.
2. The intelligent physical education testing system for physical education teaching according to claim 1, characterized in that, When performing a pull-up test, the action logic analysis unit executes the following steps: Construct a finite state machine that includes the hanging state, rising state, over-bar state, falling state, and reset state; Real-time calculation of the subject's elbow joint angle and the relative position of the key point of the jaw to the horizontal bar; A valid movement is counted only when the subject starts from the hanging position, goes through the rising position, reaches the over-bar position, and then goes through the descending position back to the hanging position. If the player starts the next movement before reaching the over-bar position or returning to the hanging position during the movement, it will not be counted and a voice correction prompt will be triggered.
3. The intelligent physical education testing system for physical education teaching according to claim 1, characterized in that, In the standing long jump test, the data acquisition module is equipped with a stereo camera, and the edge computing module executes the following ranging algorithm based on the data acquired by the stereo camera: Image pixel coordinates are mapped to physical plane coordinates based on the perspective transformation matrix; Human foreground is extracted using a background subtraction and depth thresholding segmentation algorithm; Track key points on the human foot to identify the moment of takeoff and landing; At the moment of landing, the vertical distance between the point closest to the takeoff line and the takeoff line among all valid ground contact points is calculated as the test score.
4. The intelligent physical education testing system for physical education teaching according to claim 1, characterized in that, The monitoring module employs a dual verification mechanism of facial recognition and RFID during middle- and long-distance running tests. RFID carpet antennas are set at the start and end points to record the time when the chip passes through. Facial recognition devices are installed at key nodes along the runway path to record the facial features of people passing by. The system compares the identity information recorded by RFID with the facial information captured at key points. If the two are inconsistent or the key point record is missing, it is determined to be a case of impersonation or corner cutting violation, and the score is invalid.
5. The intelligent physical education testing system for physical education teaching according to claim 1, characterized in that, The monitoring module also includes a liveness detection unit and a multi-target attribution analysis unit; The liveness detection unit distinguishes between real people and photo / video attacks by analyzing facial texture and micro-expression features. The multi-target attribution analysis unit uses affinity field technology to classify the limbs in the image into different human body instances. When a non-subject's limb is detected entering the test area and applying external force to the subject, it is determined to be illegal assistance.
6. The intelligent physical education testing system for physical education teaching according to claim 1, characterized in that, The cloud management module performs statistical analysis on the uploaded test data and generates personalized exercise prescriptions based on the analysis results, specifically including: Receive students' historical test data, body shape data, and physiological function data; Use regression analysis or machine learning clustering algorithms to assess students' physical weaknesses; The system matches relevant training movements, intensities, and frequencies from a pre-set exercise knowledge base to generate personalized after-school workout plans.