Pull-up motion detection method, device and equipment and storage medium

Abstract

The invention provides a pull-up motion detection method, device and equipment and a storage medium, and the method comprises the steps: collecting a pull-up motion video stream through a front camera; inputting the motion video stream into a posture recognition model to extract an initial joint point coordinate, judging whether jitter noise exists or not through an integration mode of a fixed coordinate system and a human body dynamic coordinate system, and adjusting a Kalman gain according to a judgment result to filter the jitter noise to obtain a corrected joint point coordinate; on the basis of the corrected joint point coordinates, motion counting is carried out according to pull-up periodic characteristics, side camera auxiliary recognition is started when the head bar-passing characteristics are judged, quantitative index analysis is carried out on periodic actions included in counting, and a motion detection result is output; and risk early warning is prompted according to the motion detection result, and the motion detection result is stored. According to the invention, subjective judgment is converted into objective data judgment, so that the counting accuracy is improved; quantitative index analysis supports scientific training guidance.

Description

Technical Field

[0001] This invention relates to the field of motion detection technology, and in particular to a method, apparatus, device, and storage medium for detecting pull-up motion. Background Technology

[0002] Pull-ups are a core physical training and assessment subject for measuring upper limb strength and core muscle stability. They are widely used in military training and school physical education tests. The standard technique requires: gripping the horizontal bar with either an overhand or underhand grip, hanging naturally, and using the upper limb and back muscles to pull the body up until the chin is above the bar, then slowly lowering back to the initial hanging position to complete one full repetition. Scientific and proper pull-up training requires correct form as a foundation; incorrect posture not only leads to poor training of the target muscle groups but can also cause sports injuries.

[0003] In related technologies, current traditional pull-up detection relies on manual observation. Examiners need to manually judge multiple dimensions such as arm extension, chin over the bar, and body stability. However, manual judgment of elbow angle and head height over the bar has a large margin of error, easily leading to missed or incorrect counts. Especially in scenarios where multiple people are being assessed simultaneously, the examiner's attention is diverted, further reducing the accuracy of judgment. Furthermore, the traditional model only focuses on the single result of valid repetitions, failing to record key details such as movement speed, force balance, and postural deviation. Trainees cannot know the root cause of their movement problems, and coaches cannot develop targeted improvement plans based on data, causing training to remain in a stage of blind repetition, making it difficult to break through ability bottlenecks. In addition, some existing counting methods use visual processing to assist motion detection, which often relies on a single fixed coordinate system in the noise filtering process, making it difficult for the filtering model to distinguish between noise and the trainee's own improper movements.

[0004] Based on the above analysis of the development status of this technology field, the existing technology lacks a solution that uses visual processing technology, employs a dual coordinate system to filter jitter noise, counts movements based on the periodic characteristics of pull-ups, and performs quantitative index analysis. Summary of the Invention

[0005] The purpose of this invention is to provide a pull-up motion detection method, apparatus, device, and storage medium, aiming to solve the above-mentioned problems in the prior art.

[0006] According to a first aspect of the present invention, a pull-up exercise detection method is provided, comprising: The video stream of pull-up movements is captured using a front-facing camera; The motion video stream is input into the posture recognition model to extract the initial joint coordinates. The presence of jitter noise is determined by integrating the fixed coordinate system and the human dynamic coordinate system. Based on the determination result, the Kalman gain is adjusted to filter the jitter noise and obtain the corrected joint coordinates. Based on the corrected joint coordinates, motion counting is performed according to the periodic characteristics of pull-ups. When judging the head over the bar feature, the side camera is used to assist in recognition. The periodic movements included in the counting are quantitatively analyzed, and the motion detection results are output. Risk warnings are issued based on the exercise detection results, and the exercise detection results are stored.

[0007] According to a second aspect of the present invention, a pull-up detection device is provided, comprising: The acquisition module is used to acquire video streams of pull-up movements using a front-facing camera; The smoothing module is used to input the motion video stream into the posture recognition model to extract the initial joint coordinates. It determines whether there is jitter noise by integrating the fixed coordinate system and the human dynamic coordinate system. Based on the judgment result, it adjusts the Kalman gain to filter the jitter noise and obtain the corrected joint coordinates. The results analysis module is used to count motions based on the periodic characteristics of pull-ups, based on the corrected joint coordinates. When judging the head over the bar feature, the side camera is used to assist in recognition. The module also performs quantitative index analysis on the periodic movements included in the count and outputs the motion detection results. The storage module is used to provide risk warnings based on motion detection results and to store the motion detection results.

[0008] According to a third aspect of the present invention, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the pull-up movement detection method provided in the first aspect of the present disclosure.

[0009] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which an information transmission implementation program is stored, which, when executed by a processor, implements the steps of the pull-up movement detection method provided in the first aspect of the present disclosure.

[0010] The technical solution provided by this invention has the following beneficial effects: by integrating a fixed coordinate system with a human dynamic coordinate system and using a Kalman filter to determine whether jitter noise exists, it can effectively distinguish between jitter noise and the non-standard movements of the human body, thus improving the effect of video stream redundancy processing; by counting movements based on the periodic characteristics of pull-ups, the subjective judgment of pull-ups is converted into a preset threshold judgment of objective data, accurately identifying features and improving the accuracy of movement counting; further, quantitative index analysis is performed to support scientific training guidance and reduce the risk of sports injuries.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of the pull-up motion detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the conceptual framework of an embodiment of the present invention; Figure 3 This is a schematic diagram of the pull-up movement detection device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0015] Method Implementation Examples According to an embodiment of the present invention, a method for detecting pull-up movements is provided. Figure 1 This is a flowchart of the pull-up movement detection method according to an embodiment of the present invention, as follows: Figure 1As shown, the pull-up exercise detection method according to an embodiment of the present invention specifically includes: In step S110, the motion video stream of the pull-up is acquired using a front-facing camera, specifically including: Deploy a front-facing camera in a direction parallel to the single bar, where parallel means parallel in a planar view; The front camera is positioned 1-1.2 meters from the horizontal bar and 1.5 meters above it, focusing on the trainee's upper limbs. In enhanced environments, the wide dynamic range function of the front camera is enabled; in low-light environments, the fill light of the front camera is enabled, with the fill light intensity being greater than or equal to 400 lux. Preferably, a Gaussian filtering denoising algorithm is used to eliminate the impact of environmental interference on image accuracy, ensuring that recognition is not affected by light and shadow, and obtaining a motion video stream.

[0016] In step S120, the motion video stream is input into the pose recognition model to extract initial joint coordinates. The presence of jitter noise is determined by integrating a fixed coordinate system with the human dynamic coordinate system. Based on the determination result, the Kalman gain is adjusted to filter the jitter noise, resulting in corrected joint coordinates. Specifically, this includes: Input the motion video stream into the OpenPose pose recognition model; Preferably, the OpenPose posture recognition model has at least 5,000 training samples, including videos of standard and non-standard pull-up movements, to accurately extract the coordinates of the core joints of the trainee; preferably, the OpenPose posture recognition model is equipped with a new horizontal bar mask module to distinguish the horizontal bar from the athlete. The OpenPose pose recognition model outputs initial joint points for the wrist, elbow, shoulder, head, and hip.

[0017] The coordinate fluctuations caused by slight body swaying of the trainee, slight adjustment of the wrist position on the bar, or slight camera shake can be identified as jitter noise. That is, the deviation is not caused by the trainee's non-standard or non-standard movements, but by movement within the allowable error range or by external factors unrelated to the movement. Traditional visual processing methods use a static, fixed coordinate system, which makes it difficult to effectively distinguish between noise and the irregularity of the trainee's movements. Therefore, a dual coordinate system is used in this embodiment of the invention. A two-dimensional coordinate system is established based on the plane of the single bar as a fixed coordinate system, wherein the x-axis of the fixed coordinate system is parallel to the ground and the y-axis is perpendicular to the ground; A dynamic two-dimensional coordinate system is established based on human movement as the human dynamic coordinate system. The x-axis of the human dynamic coordinate system is the direction in which the initial joints of both shoulders are formed, the y-axis is perpendicular to the x-axis, and the midpoint of the initial joints of both shoulders is taken as the origin. In other words, when the trainee moves or rotates as a whole, the human body's dynamic coordinate system also rotates. Observing limb movements in this coordinate system can greatly eliminate coordinate changes caused by the overall movement or rotation of the body.

[0018] In this embodiment of the invention, the coordinate accuracy is ±1 pixel, and the actual mapped distance is ±0.5 cm; The initial joint coordinates are smoothed using a Kalman filter to obtain the corrected joint coordinates: If the initial joint coordinates of the motion video stream wobble beyond a preset threshold in the fixed coordinate system, but do not wobble beyond the preset threshold in the human dynamic coordinate system, then jitter noise is identified and the Kalman gain in the Kalman filter is reduced; otherwise, the Kalman gain is not adjusted. The Kalman filter generates observations and predictions during processing. Reducing the Kalman gain means that the predictions are trusted more because there is a lot of noise in the observations. The embodiments of the present invention can effectively process noise while preserving the essential motion characteristics of the trainee and will not smooth out poor-quality movements. By using Kalman filter to predict the trajectory of adjacent frames, we can ensure that key point tracking is continuous and without losing frames.

[0019] The relative positions of the wrist joint and the hand joint are generally fixed, which can determine the hand's fine-tuning grip position. Preferably, the position of the hand joint can be additionally identified.

[0020] In step S130, based on the corrected joint coordinates, motion counting is performed according to the periodic characteristics of the pull-up. When determining the head-over-bar feature, a side camera is used for assisted recognition. The periodic movements included in the counting are then quantitatively analyzed, and the motion detection results are output, specifically including: Corrected joint coordinates were used in all subsequent results analysis. Once the pull-up exercise has passed the initial hanging state judgment, the pull-up process qualification judgment, and the lowering process qualification judgment in sequence, the exercise is considered successful and the exercise count is incremented by one; Based on the corrected joint coordinates, the elbow angle is calculated according to the elbow joint and the shoulder joint. When the elbow angle is higher than the straightening threshold, the line connecting the hip joint and the shoulder joint is perpendicular to the ground, and the head joint is lower than the bar surface, it is determined to meet the initial hanging state. In this embodiment of the invention, the straightening threshold is 170°, and if this is met, the arm is determined to be straight; the deviation of the y-axis direction from the ground is less than 3 cm, and if this is met, the body is determined to be without tilt; whether the head joint point is lower than the bar surface through the y-axis coordinate is determined, and if this is met, it is determined to be the initial hanging state. If the following conditions are met in sequence: arms straight, body without tilting, and initial hanging, then the initial hanging state is met.

[0021] Based on the corrected joint coordinates, when the head passes the bar above the bar threshold, the elbow angle is less than the force threshold, and the hip joint is less than the deviation threshold, it is determined that the pull-up process meets the standard. In this embodiment of the invention, the difference between the y-axis coordinate of the head joint point and the y-axis coordinate of the bar surface is calculated. The threshold for passing the bar is 1 cm. If the threshold is met, it is determined that the bar has been passed. The force threshold is 90°. If the threshold is met, it is determined that the upper limbs have exerted force in place. The deviation threshold is 10 cm and the x-axis deviation. If the threshold is met, it is determined that there is no excessive swinging of the waist. If the following conditions are met in sequence—clearing the bar, engaging the upper limbs, and avoiding excessive waist swaying—then the pull-up process is considered to have met the standards. Enabling side-view camera-assisted recognition when determining head-over-bar features specifically includes: When the front camera determines that the head is not above the bar threshold, the side camera is activated; the side refers to the direction perpendicular to the bar, that is, the trainee's side is seen in the picture. The frame rate, timestamp, and focal length of the side camera and the front camera are synchronized. If the side camera determines that the head is above the threshold, the result is that the head is above the threshold. Otherwise, the original determination is maintained to avoid recognition errors caused by the interference of the pole.

[0022] Based on the corrected joint coordinates, when the elbow angle recovers from less than the force threshold to greater than the extension threshold, the head joint point falls back below the bar, and the hip joint point is less than the deviation threshold, it is determined that the lowering process meets the standard. In this embodiment of the invention, the elbow angle is tracked to recover from ≤90° to ≥170°. If this condition is met, the arm is determined to be straight again. At the same time, the Y-axis coordinate of the head joint point falls back to below the bar surface, indicating a successful fall. The hip offset is still ≤10cm, indicating that the lowering standard is met. If the arm is straightened again, successfully lowered, and lowered to the target position in sequence, then the lowering process is considered to have met the target.

[0023] In this embodiment of the invention, actions that do not meet the above threshold are considered non-standard and are not counted as valid attempts. If the following situations occur, the actions are not counted as valid attempts, and the problem type is marked: If the elbow angle returns to 160°-169° and is not fully extended: mark "arm not fully extended"; If the head height above the bar is less than 1cm, the standard is not met: mark "Chin not above the bar"; Hip X-axis offset > 10cm, excessive lumbar sway: marked "lumbar compensation".

[0024] Preferably, the stage can be marked, such as "chin not over the bar during the pull-up process" or "lumbar compensation during the lowering process".

[0025] Based on the corrected joint coordinates and time series data, a quantitative index analysis is performed on the periodic movements of pull-ups in a single count. That is, a further quality analysis is performed on the movements that are identified as successful, providing data support for training guidance. The analysis is more detailed, while the problems of movements that cannot be counted successfully are more superficial and more serious, and do not require quantitative index analysis. (1) To measure the speed of movement (reflecting the rhythm of force exertion) Calculate the duration of the motion cycle using Formula 1. This refers to the duration of a single "suspend-pull-lower" motion cycle, measured in seconds. The motion speed is calculated using Formula 2. : Formula 1; Formula 2; in, Indicates the end frame of the action cycle. Indicates the start frame of the action cycle. Indicates frame rate; if A frequency greater than 0.3 beats per second indicates "the movement is too fast and the muscles are not fully engaged." A frequency of less than 0.15 beats per second indicates "insufficient upper limb strength".

[0026] (2) Balance of force exertion Use Formula 3 to calculate the elbow flexion speed. It is necessary to extract data from the left and right elbow joints based on the camera, and then use Formula 4 to calculate the force difference rate. : Formula 3; Formula 4; in, This indicates the elbow flexion angle, which is the angle formed by the wrist, elbow, and shoulder. Indicates time difference, Indicates the speed at which the left elbow bends. Indicates the speed at which the right elbow bends; if A value greater than 20% indicates "uneven force exertion between the left and right sides".

[0027] (3) Waist offset Based on the hip joint points extracted by the camera, the maximum horizontal displacement of the hip relative to the initial hanging position along the x-axis during the movement cycle is calculated, and the lumbar displacement is calculated using Formula 5. : Formula 5; in, This indicates that the hip joint point deviates from its maximum value. This indicates that the hip joint point deviates from the minimum value. This indicates the state of the hip joint point in the initial suspension state, i.e., the corresponding x-axis coordinate value; if A reading greater than 15 cm indicates "lumbar compensation." For the best results, it may further indicate "excessive lumbar compensation, which can easily lead to lumbar and back strain, and requires strengthening core training such as plank exercises."

[0028] In step S140, a risk warning is issued based on the motion detection results, and the motion detection results are stored. Specifically, this includes: Based on the common injury sites in pull-ups—the shoulders, elbows, and lower back—a three-tiered risk mechanism is established: Low risk: Occasional lumbar deviation 10cm < S ≤ 15cm, slight force imbalance 15% < D ≤ 20%, a green text prompt such as "Pay attention to body stability" will pop up on the interactive screen; Medium risk: Multiple instances of improper movements, such as "arms not fully extended" 3 times, or uneven force exertion (D>20%), will trigger a voice broadcast such as "significant difference in force exertion between left and right sides, adjust grip strength" and push basic reinforcement training suggestions, such as resistance band assisted pull-ups; High risk: Waist deviation S>15cm and accompanied by head not passing the bar properly and elbow overextension, i.e. angle>175°, which is prone to elbow joint strain. A red warning window will pop up in real time, with a voice prompt "pause adjustment" and a standard movement animation will be displayed, i.e., the elbow angle and hip position are marked.

[0029] The motion detection results are stored in an archive, which includes the following: Basic information: Trainee's name, ID number, and testing time; Key data: Exercise detection results; effective number of times, types of non-standard movements and their proportions, such as "arms not straightened" accounting for 20%, and quantitative indicator curves, namely, the changes in movement speed and waist deviation over time; Auxiliary information: key action frames, such as "chin over the bar" and "waist shift" (typical images), and risk warning records such as high-risk trigger time points.

[0030] The archive data is uploaded to the central cloud server via an encryption protocol, using the AES-256 encryption algorithm. Trainees or coaches can query historical data through an interactive screen, such as changes in effective repetitions over the past month and trends in improvement of waist offset. Coaches can analyze common problems among trainees in batches based on the archives, such as a high percentage of students in a class whose arms are not fully extended, and optimize group training programs accordingly.

[0031] Based on the above description, the embodiments of the present invention follow the full-process design of "data acquisition - preprocessing - feature extraction - quantitative analysis - judgment and early warning - file integration", and optimize key links for the action cycle of pull-up "hanging-pulling up-lowering".

[0032] Figure 2 This is a schematic diagram illustrating the conceptual framework of an embodiment of the present invention, such as... Figure 2 As shown, the process of forming a pull-up movement detection scheme is illustrated. In response to the existing problems of pull-ups, the scheme analyzes the movement characteristics of the preparation, movement, and finishing phases, considers advanced analysis of inconspicuous movement judgment, and generates subsequent training plans, thereby obtaining the pull-up movement detection scheme of this invention.

[0033] By applying the embodiments of this invention, the following improvements in motion detection data are achieved: Through multi-view visual recognition and quantitative index evaluation, subjective judgment standards such as pull-ups are transformed into objective data. The error in the effective count is ≤1%, far lower than the 5%-10% of manual judgment and the 8%-12% of simple sensor devices. Simultaneously, by correcting joint coordinates using the Kalman filter algorithm, the detection accuracy of key indicators such as waist offset and elbow angle reaches ±0.5cm and ±2°, respectively. This accurately identifies non-standard movements that are easily overlooked by humans, such as "slight waist compensation" and "incomplete elbow extension." The analysis results not only record the effective counts but also output multi-dimensional quantitative data such as movement speed, force balance, and waist offset, helping trainees identify the root cause of problems. Coaches can develop targeted reinforcement training plans based on the commonalities of the class's overall data, changing the blind training mode of existing technologies that only look at results without improvement direction. This shifts pull-up training from experience-driven to data-driven, significantly improving training effectiveness. Experimental verification shows that using this invention... After one week, trainees' effective repetitions increased by an average of 15%-20%, and the proportion of non-standard movements decreased by 40%. A three-level early warning mechanism was established: low-risk trainees received real-time posture adjustment prompts, medium-risk trainees received basic training suggestions, and high-risk trainees received a pause warning. The system immediately provided voice prompts and displayed the standard posture to prevent further injuries caused by incorrect movements. Compared to the injury rate of approximately 12% without real-time early warning technology, this invention can reduce the incidence of pull-up injuries to below 5%, ensuring training safety and continuity.

[0034] The following are the cost improvement figures for the application: For single-person inspections, no manual supervision is required; the system automatically completes posture judgment, counting, and early warning, reducing the average inspection time per person from 3 minutes to 1 minute, an efficiency increase of 200%. For large-scale scenarios such as a military company of 50 people or a school class of 40 people, only one staff member is needed to manage the equipment and complete the inspection of all personnel. Compared to the traditional model requiring 3-4 examiners, labor costs are reduced by 60%-75%, while also avoiding inconsistent judgment standards due to human fatigue, reducing the cost of assessment disputes. The hardware architecture, including cameras, processors, and cloud servers, effectively achieves data interoperability, reducing investment costs by 50%-60%, and adding 2... By adapting the camera to pull-ups and optimizing the posture recognition model, the pull-up detection function can be expanded, making it highly economical and scalable. The pull-up training profiles built can track trainees' data over a long period of time. Coaches no longer need to repeatedly assess trainees' basic abilities and can directly develop advanced plans based on historical data, reducing the time spent on repeated testing and assessment. At the same time, the database can accumulate solutions to common problems, so new trainees do not need to start from scratch, shortening the training cycle by 20%-30% and indirectly reducing the cost of repeated consumption of training consumables.

[0035] The technical effects are as follows: In summary, to address the existing problems, this invention proposes a pull-up motion detection method. It integrates a fixed coordinate system and a human dynamic coordinate system with a Kalman filter to determine the presence of jitter noise. If there is jitter in the fixed coordinate system but stability in the human dynamic coordinate system, it is identified as noise. This effectively distinguishes jitter noise from improper human movement, improving the efficiency of video stream redundancy processing. Motion counting is performed based on the periodic characteristics of pull-ups, converting subjective judgments into objective data based on preset thresholds, thus accurately identifying features and improving the accuracy of motion counting. Furthermore, when the front camera determines that the head is below the threshold for crossing the bar, a side camera is activated, forming a double safety net with the human posture model capable of recognizing the bar, avoiding misjudgments caused by errors in human-bar relationship identification. Further quantitative analysis is conducted to support scientific training guidance and reduce the risk of sports injuries.

[0036] Device Examples According to an embodiment of the present invention, a pull-up exercise detection device is provided. Figure 3 This is a schematic diagram of the pull-up exercise detection device according to an embodiment of the present invention, as shown below. Figure 3 As shown, the pull-up detection device according to an embodiment of the present invention specifically includes: Acquisition module 30 is used to acquire the video stream of pull-up movements via a front-facing camera, specifically for: Deploy a front-facing camera parallel to the direction of the horizontal bar; In enhanced environments, the wide dynamic range function of the front camera is enabled; in low-light environments, the fill light of the front camera is enabled, resulting in a motion video stream.

[0037] The smoothing module 32 is used to input the motion video stream into the pose recognition model to extract the initial joint coordinates. It determines the presence of jitter noise by integrating a fixed coordinate system with the human dynamic coordinate system. Based on the determination result, it adjusts the Kalman gain to filter the jitter noise and obtains the corrected joint coordinates. Specifically, it is used for: Input the motion video stream into the OpenPose pose recognition model; The OpenPose pose recognition model outputs initial joint points for the wrist, elbow, shoulder, head, and hip.

[0038] A two-dimensional coordinate system is established based on the plane of the single bar as a fixed coordinate system, wherein the x-axis of the fixed coordinate system is parallel to the ground and the y-axis is perpendicular to the ground; A dynamic two-dimensional coordinate system is established based on human movement as the human dynamic coordinate system. The x-axis of the human dynamic coordinate system is the direction in which the initial joints of both shoulders are formed, the y-axis is perpendicular to the x-axis, and the midpoint of the initial joints of both shoulders is taken as the origin. The initial joint coordinates are smoothed using a Kalman filter to obtain the corrected joint coordinates. If the initial joint coordinates of the motion video stream fluctuate beyond a preset threshold in the fixed coordinate system, but do not fluctuate beyond the preset threshold in the human dynamic coordinate system, then jitter noise is identified and the Kalman gain in the Kalman filter is reduced.

[0039] The results analysis module 34 is used to count motions based on the corrected joint coordinates and the periodic characteristics of pull-ups. When determining the head-over-bar feature, it uses a side camera for assisted recognition and performs quantitative index analysis on the periodic movements included in the count, outputting the motion detection results. Specifically, it is used for: Once the pull-up exercise has passed the initial hanging state judgment, the pull-up process qualification judgment, and the lowering process qualification judgment in sequence, the exercise is considered successful and the exercise count is incremented by one; Based on the corrected joint coordinates, the elbow angle is calculated according to the elbow joint and the shoulder joint. When the elbow angle is higher than the straightening threshold, the line connecting the hip joint and the shoulder joint is perpendicular to the ground, and the head joint is lower than the bar surface, it is determined to meet the initial hanging state. Based on the corrected joint coordinates, when the head passes the bar above the bar threshold, the elbow angle is less than the force threshold, and the hip joint is less than the deviation threshold, it is determined that the pull-up process meets the standard. Based on the corrected joint coordinates, when the elbow angle recovers from less than the force threshold to greater than the extension threshold, the head joint falls back below the bar, and the hip joint is less than the deviation threshold, it is determined that the lowering process meets the standard.

[0040] Enabling side-view camera-assisted recognition when determining head-over-bar features specifically includes: When the front camera determines that the head is not above the threshold for crossing the bar, the side camera is activated; The frame rate, timestamp, and focal length of the side camera and the front camera are synchronized. If the side camera determines that the head exceeds the threshold, the result is that the head exceeds the threshold.

[0041] Based on the corrected joint coordinates, a quantitative index analysis of the periodic movement of pull-ups in a single count is performed. Calculate the duration of the motion cycle using Formula 1. Calculate the motion speed using Formula 2. : Formula 1; Formula 2; in, Indicates the end frame of the action cycle. Indicates the start frame of the action cycle. Indicates frame rate; Use Formula 3 to calculate the elbow flexion speed. Formula 4 is used to calculate the force difference rate. : Formula 3; Formula 4; in, Indicates the angle of elbow flexion. Indicates time difference, Indicates the speed at which the left elbow bends. Indicates the speed at which the right elbow bends; Calculate the waist offset using Formula 5. : Formula 5; in, This indicates that the hip joint point deviates from its maximum value. This indicates that the hip joint point deviates from the minimum value. This indicates the state of the hip joint points during the initial suspension phase.

[0042] The storage module 36 is used to provide risk warnings based on motion detection results and to store the motion detection results.

[0043] In this embodiment of the invention, the acquisition module 30, equipped with a fill light and wide dynamic range function, acquires images of the trainee and simultaneously outputs video streams and metadata. The smoothing module 32 and the result analysis module 34 both belong to the data processing module. They use an embedded processor (NVIDIA Jetson Xavier NX), 128GB SSD local storage, and have built-in attitude recognition model, quantization algorithm and risk judgment logic. The processing latency is ≤100ms, and temporary data is stored locally to avoid data loss due to network interruption. Storage module 36 uses a central cloud server; Preferably, the embodiments of the present invention also include an interactive display module 38 and a power supply module 310. The interactive display module 38 is a 21.5-inch touch screen (1920×1080 resolution) and a 15W speaker, similar to an edge interactive device, which displays the effective number of times, quantitative indicators, and risk warnings in real time, and supports trainees to query historical data; the speaker outputs voice warnings; the power supply module 310 is an AC 220V power supply and a 12V backup lithium battery with a battery life of ≥4h, which supplies power to each module and avoids detection interruption due to sudden power failure.

[0044] Preferably, in addition to the device, it can also be deployed in a software system. The software system is mainly divided into functional modules, including image processing function, posture recognition function, quantitative analysis function, interactive control function, data encryption and transmission function, and file management function. Image processing functions: realize video frame noise reduction, ambient light adaptation, background segmentation and filtering of interference around the bar, and output clear joint point recognition images; Posture recognition function: Based on the optimized OpenPose model, it extracts the coordinates of the core joints in pull-ups, with a recognition accuracy of ≥96%; Quantitative analysis function: Embedded pull-up calculation algorithms such as elbow angle and waist offset, output quantitative indicators and risk levels; Interactive control functions: Control the touch screen display and speaker alerts, and support trainees to trigger operations such as starting the test and querying files; Data encryption and transmission functions: Enables encrypted local data storage and cloud synchronization; File management function: Supports file creation, query, comparative analysis, and generation of training evaluation reports.

[0045] In summary, addressing the existing problems, this invention presents a pull-up motion detection device. It integrates a fixed coordinate system and a human dynamic coordinate system using a Kalman filter to determine the presence of jitter noise. If there is jitter in the fixed coordinate system but stability in the human dynamic coordinate system, it is identified as noise. This effectively distinguishes jitter noise from improper human movement, improving the efficiency of video stream redundancy processing. Based on the periodic characteristics of pull-ups, motion counting is performed, converting subjective judgments of pull-ups into preset threshold judgments based on objective data, accurately identifying features and improving the accuracy of motion counting. Furthermore, when the front camera determines that the head is below the threshold for crossing the bar, a side camera is activated, forming a double safety net with the human posture model capable of recognizing the bar, avoiding misjudgments caused by errors in human-bar relationship identification. Further quantitative analysis is conducted to support scientific training guidance and reduce the risk of sports injuries.

[0046] Electronic device examples Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device 400 may include at least one processor 410 and a memory 420. The processor 410 can execute instructions stored in the memory 420. The processor 410 is communicatively connected to the memory 420 via a data bus. In addition to the memory 420, the processor 410 can also be communicatively connected to an input device 430, an output device 440, and a communication device 450 via the data bus.

[0047] Processor 410 can be any conventional processor, such as a commercially available CPU. Processors may also include graphics processing units (GPUs), field-programmable gate arrays (FPGAs), systems-on-chips (SoCs), application-specific integrated circuits (ASICs), or combinations thereof.

[0048] The memory 420 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0049] In this embodiment of the present disclosure, the memory 420 stores executable instructions, and the processor 410 can read the executable instructions from the memory 420 and execute the instructions to implement all or part of the steps of any of the pull-up movement detection methods in the exemplary embodiments described above.

[0050] Computer-readable storage medium embodiments In addition to the methods and apparatus described above, exemplary embodiments of this disclosure may also be a computer program product or a computer-readable storage medium storing the computer program product, the computer product including computer program instructions that can be executed by a processor to implement all or part of the steps described in any of the pull-up movement detection methods in the exemplary embodiments described above.

[0051] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. Programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages, and scripting languages ​​(e.g., Python). The program code can be executed entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0052] Computer-readable storage media may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: static random access memory (SRAM) having one or more electrically connected wires; electrically erasable programmable read-only memory (EEPROM); erasable programmable read-only memory (EPROM); programmable read-only memory (PROM); read-only memory (ROM); magnetic storage; flash memory; magnetic disk or optical disk; or any suitable combination thereof.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting pull-up movement, characterized in that, include: The video stream of pull-up movements is captured using a front-facing camera; The motion video stream is input into the posture recognition model to extract the initial joint coordinates. The presence of jitter noise is determined by integrating the fixed coordinate system and the human dynamic coordinate system. Based on the determination result, the Kalman gain is adjusted to filter the jitter noise and obtain the corrected joint coordinates. Based on the corrected joint coordinates, motion counting is performed according to the periodic characteristics of pull-ups. When judging the head-over-bar feature, the side camera is used to assist in recognition. The periodic movements included in the counting are quantitatively analyzed, and the motion detection results are output. Risk warnings are issued based on the motion detection results, and the motion detection results are stored.

2. The method according to claim 1, characterized in that, The specific steps of capturing the pull-up motion video stream using a front-facing camera include: The front camera is deployed parallel to the direction of the single bar; The motion video stream is obtained by activating the wide dynamic range function of the front camera in an enhanced environment and activating the fill light of the front camera in a low-light environment.

3. The method according to claim 1, characterized in that, The step of inputting the motion video stream into the pose recognition model to extract the initial joint coordinates specifically includes: The motion video stream is input into the OpenPose pose recognition model; The OpenPose pose recognition model outputs initial joint points for the wrist, elbow, shoulder, head, and hip.

4. The method according to claim 1, characterized in that, The method of determining the presence of jitter noise by integrating a fixed coordinate system and a human dynamic coordinate system, and adjusting the Kalman gain to filter the jitter noise based on the determination result to obtain the corrected joint point coordinates specifically includes: A two-dimensional coordinate system is established based on the plane of the single bar as the fixed coordinate system, wherein the x-axis of the fixed coordinate system is parallel to the ground and the y-axis is perpendicular to the ground; A dynamic two-dimensional coordinate system is established based on human movement as the human dynamic coordinate system. The x-axis of the human dynamic coordinate system is the direction in which the initial joints of both shoulders are formed, the y-axis is perpendicular to the x-axis, and the midpoint of the initial joints of both shoulders is taken as the origin. The initial joint coordinates are smoothed using a Kalman filter to obtain the corrected joint coordinates: If the initial keypoint coordinates of the motion video stream fluctuate more than a preset threshold in the fixed coordinate system, but do not fluctuate more than the preset threshold in the human dynamic coordinate system, then jitter noise is determined to exist and the Kalman gain in the Kalman filter is reduced.

5. The method according to claim 1, characterized in that, The method of counting movements based on the periodic characteristics of pull-ups specifically includes: Once the pull-up exercise has passed the initial hanging state judgment, the pull-up process qualification judgment, and the lowering process qualification judgment in sequence, the exercise is considered successful and the exercise count is incremented by one; Based on the corrected joint coordinates, the elbow angle is calculated according to the elbow joint and the shoulder joint. When the elbow angle is higher than the straightening threshold, the line connecting the hip joint and the shoulder joint is perpendicular to the ground, and the head joint is lower than the bar surface, it is determined to meet the initial hanging state. Based on the corrected joint coordinates, when the head passes the bar higher than the bar threshold, the elbow angle is less than the force threshold, and the hip joint is less than the deviation threshold, it is determined that the pull-up process meets the standard. Based on the corrected joint coordinates, when the elbow angle recovers from less than the force threshold to greater than the straightening threshold, the head joint falls back below the bar, and the hip joint is less than the deviation threshold, it is determined that the lowering process meets the standard.

6. The method according to claim 5, characterized in that, The specific steps of enabling the side camera to assist in recognition when determining head crossing features include: When the front camera determines that the head crossing the bar is not higher than the bar threshold, the side camera is activated; The frame rate, timestamp, and focal length of the side camera and the front camera are synchronized. If the side camera determines that the head exceeds the threshold, the result is that the head exceeds the threshold.

7. The method according to claim 1, characterized in that, The quantitative analysis of the periodic movements included in the counting, and the output of motion detection results, specifically include: Based on the corrected joint coordinates, a quantitative index analysis is performed on the periodic movements of pull-ups in a single count. Calculate the duration of the motion cycle using Formula 1. Calculate the motion speed using Formula 2. : Official 1; Official 2; in, Indicates the end frame of the action cycle. Indicates the start frame of the action cycle. Indicates frame rate; Use Formula 3 to calculate the elbow flexion speed. Formula 4 is used to calculate the force difference rate. : Official 3; Official 4; in, Indicates the angle of elbow flexion. Indicates time difference, Indicates the speed at which the left elbow bends. Indicates the speed at which the right elbow bends; Calculate the waist offset using Formula 5. : Official 5; in, This indicates that the hip joint point deviates from its maximum value. This indicates that the hip joint point deviates from the minimum value. This indicates the state of the hip joint points during the initial suspension phase.

8. A pull-up exercise detection device, characterized in that, include: The acquisition module is used to acquire video streams of pull-up movements using a front-facing camera; The smoothing module is used to input the motion video stream into the posture recognition model to extract the initial joint coordinates, determine whether there is jitter noise by integrating the fixed coordinate system and the human dynamic coordinate system, and adjust the Kalman gain to filter the jitter noise according to the judgment result to obtain the corrected joint coordinates. The results analysis module is used to count motions based on the corrected joint coordinates and the periodic characteristics of pull-ups. When judging the head-over-bar feature, the side camera is used to assist in recognition. The module also performs quantitative index analysis on the periodic movements included in the count and outputs the motion detection results. The storage module is used to issue risk warnings based on the motion detection results and store the motion detection results.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the pull-up movement detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for information transmission, which, when executed by a processor, implements the steps of the pull-up movement detection method as described in any one of claims 1 to 7.

Images