Crank arm overhanging motion detection method, device and equipment and storage medium

Abstract

The invention provides a method, a device, equipment and a storage medium for detecting the overhanging motion of a bent arm, and the method comprises the steps: obtaining image data through a camera disposed right in front of the overhanging motion of the bent arm, recognizing the joint points of an athlete in the image data through a target recognition model, and when the shoulder joint points are shielded by a horizontal bar, recognizing the joint points of the athlete; performing coordinate completion by adopting a completion algorithm based on a human body structure relationship; acquiring motion characteristic data of an athlete through a multi-modal sensor; performing quantitative analysis based on the joint points and the motion feature data to form a time sequence including a quantitative analysis result, inputting the time sequence into a fusion model including feature extraction and sequence analysis, and outputting an intelligent evaluation result; and matching the quantitative analysis result with the intelligent evaluation result by using a preset scenarized template to obtain a motion detection result, and filing and storing the motion detection result. Multi-source data are integrated through image processing and multi-source sensor equipment, and the problems of subjective evaluation and insufficient precision are solved.

Description

Technical Field

[0001] This invention relates to the field of motion detection technology, and in particular to a method, apparatus, equipment, and storage medium for detecting motion during flexed arm suspension. Background Technology

[0002] The bent-arm hang, a classic exercise for measuring upper limb strength and core stability, is widely used in physical fitness assessments, fitness training, and rehabilitation.

[0003] Among related technologies, traditional detection and evaluation techniques still have significant limitations. Traditional arm-bending hang detection relies entirely on manual observation, judging the standard of movement such as chin over the bar and elbow flexion angle by the naked eye, and recording the duration or number of times completed. Different evaluators have different judgment standards, with an error rate of over 20%. Moreover, one examiner can only monitor a maximum of 3-4 subjects at the same time. Large-scale assessments require a large amount of manpower and are prone to misjudgment due to fatigue. Furthermore, traditional methods can only detect whether basic movements are completed and cannot judge the quality of key movements such as body swing and elbow angle. In recent years, visual recognition technology has begun to be applied to motion detection, using open-source frameworks to identify key points of the human skeleton to judge whether the movement meets the standard. However, the horizontal bar can easily obscure key shoulder nodes, causing the coordinates of the skeletal points to deviate and making it impossible to accurately calculate the elbow flexion angle. In addition, the definition of valid movements varies in different scenarios, and the existing judgment standards use fixed thresholds, which lack flexibility.

[0004] Based on the above analysis of the development status of this technology field, the existing technologies lack a scheme that uses a completion algorithm based on human body structure relationship for coordinate completion during visual processing, integrates visual, mechanical and physiological data to construct a multi-source evaluation method, and sets up a configurable evaluation system. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, device, and storage medium for detecting bent-arm suspension motion, in order to solve the aforementioned problems in the prior art.

[0006] According to a first aspect of the present invention, a method for detecting motion during a bent-arm suspension is provided, comprising: Image data is acquired by a camera deployed directly in front of the bent-arm hanger. The target recognition model is used to identify the joints of the athlete in the image data. When the shoulder joint is obscured by the horizontal bar, a completion algorithm based on human body structure relationship is used to complete the coordinates. Acquire motion characteristic data of athletes using multimodal sensors; Quantitative analysis is performed based on key point and motion feature data to form a time series including the quantitative analysis results. The time series is then input into a fusion model that includes feature extraction and sequence analysis to output intelligent evaluation results. The motion detection results are obtained by matching the quantitative analysis results and intelligent evaluation results with the preset scenario templates, and the motion detection results are archived and stored.

[0007] According to a second aspect of the present invention, a flexed arm suspension motion detection device is provided, comprising: The image processing module is used to acquire image data by a camera deployed directly in front of the bent-arm hang, use a target recognition model to identify the joints of the athlete in the image data, and use a completion algorithm based on human body structure relationship to complete the coordinates when the shoulder joint is obscured by the horizontal bar. The multimodal sensing module is used to acquire motion characteristic data of athletes through multimodal sensors; The initial analysis module is used to perform quantitative analysis based on key point and motion feature data, forming a time series including the quantitative analysis results. The time series is then input into a fusion model that includes feature extraction and sequence analysis, and the intelligent evaluation results are output. The template matching module is used to match the quantitative analysis results and intelligent evaluation results with preset scenario-based templates to obtain motion detection results, and then archive and 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 flexed arm suspension motion 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 flexed arm suspension motion detection method provided in the first aspect of the present disclosure.

[0010] The technical solution provided by the embodiments of the present invention includes the following beneficial effects: during visual processing, a completion algorithm based on human body structure is used for coordinate completion, so that the completed shoulder joint points conform to human ergonomic structure and ensure the authenticity of the output data; through image processing and multi-source sensor devices, visual, mechanical and physiological data are integrated to construct a multi-source evaluation method, which solves the problems of insufficient subjective evaluation and accuracy, and sets up a configurable evaluation system to alleviate the problem of single scene adaptability.

[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 bent-arm suspension motion detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the complete motion of the bent-arm suspension according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the bent-arm suspension motion 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 Example 1 According to an embodiment of the present invention, a method for detecting motion during a bent-arm suspension is provided. Figure 1 This is a flowchart of the bent-arm suspension motion detection method according to an embodiment of the present invention, as follows: Figure 1 As shown, the method for detecting bent-arm suspension motion according to an embodiment of the present invention specifically includes: In step S110, image data is acquired by a camera positioned directly in front of the bent-arm hanger. A target recognition model is used to identify the athlete's joints in the image data. When the shoulder joint is obscured by the horizontal bar, a completion algorithm based on human structural relationships is used for coordinate completion. Specifically, this includes: Figure 2 This is a schematic diagram of the complete motion of the bent-arm suspension according to an embodiment of the present invention, as shown below. Figure 2 As shown, the complete process of the bent-arm hang exercise is demonstrated, including the preparation phase, starting posture, force exertion phase, hang maintenance, and ending movement.

[0016] Image data is captured using a camera configured at 60 frames per second and 4K resolution, with the camera positioned parallel to the horizontal bar used for the bent-arm hang exercise, i.e., directly in front of the movement.

[0017] The OpenPose target recognition model uses the newly added horizontal bar apparatus mask module to identify the athlete's joint points. The horizontal bar is identified by the horizontal bar apparatus mask module to distinguish it from the athlete. The joint points include the shoulder joint point, elbow joint point, wrist joint point, and jaw point. In this embodiment of the invention, the identification error of the joint points is controlled within 3 centimeters; The coordinate completion process is as follows: After the athlete has taken off the bar but has not yet begun the flexed arm hanging motion, the spatial coordinates of the wrist joint are fixed, and the forearm length and upper arm length are determined by the spatial coordinates of the shoulder joint and elbow joint, forming the human body structural relationship, which is equivalent to forming a rigid triangular structure. Preferably, the determination of forearm length and upper arm length can be completed more accurately before the topswing; When the arm-bending hanging exercise begins and the shoulder joint is obscured by the bar, the spatial direction of the upper arm movement is identified by the inertial measurement unit (IMU) sensor worn on the upper arm. Based on the current spatial coordinates of the elbow joint, the spatial direction, and the length of the upper arm, the spatial coordinates of the unique shoulder joint are determined. The obtained shoulder joint is smoothed by a Kalman filter to output a stable and accurate shoulder joint movement trajectory. Theoretically, knowing the elbow joint point, the direction of upper arm movement, and the length of the upper arm can also reveal the coordinates of the shoulder joint point. However, the two-point determination method is unreliable. For example, if the elbow joint point coordinates are misidentified during movement, it will also lead to the shoulder joint point coordinates being misidentified. However, when this happens, it cannot be quickly identified. The wrist joint point can be fixed as an anchor point, and the forearm length and upper arm length can be used as anchor lengths. Therefore, it is necessary to determine whether the current forearm length and upper arm length are within the preset range of the values ​​at the time of calibration. Under normal circumstances, the length will not change. If it is not within the preset range, manual coordinate completion is performed, or existing difference filling technology is used. For example, if the forearm length is significantly reduced, it proves that there is a problem with the recognition, such as a slight offset of the wrist joint.

[0018] In step S120, the athlete's motion characteristic data is acquired through a multimodal sensor, specifically including: The trunk tilt angle was measured by an inertial measurement IMU sensor worn on the athlete's waist, directly behind the navel, and the elbow joint angular velocity was measured by a 6-axis inertial measurement IMU sensor worn on the athlete's upper arm, on the outside of the biceps. The IMU sampling rate was 500 Hz. Preferably, the torso tilt angle can determine whether the chest is hunched or the waist is collapsed, and the elbow joint angular velocity can identify invalid half-bent arm movements; EMG electromyography signals were measured using an electromyography sensor worn on the upper arm of the athlete. The force is measured by a pressure sensor deployed on a horizontal bar.

[0019] In step S130, quantitative analysis is performed based on key point and motion feature data to form a time series including the quantitative analysis results. The time series is then input into a fusion model that includes feature extraction and sequence analysis to output intelligent evaluation results, specifically including: The collected data is transformed into calculable objective indicators to establish a three-dimensional evaluation system that includes movement form, force distribution, and muscle activation. The core quantitative indicators are as follows: Using the elbow joint as the vertex and the shoulder and wrist joints as the edge points, calculate the angle between the vectors as the elbow bending angle, i.e., cos = vector dot product / modulus product, in degrees (°). The vertical distance between the mandibular point and the horizontal bar is calculated as the mandibular clearance. A positive value indicates clearance, and a negative value indicates failure to clear the bar. Calculate the rate of change of the trunk tilt angle as a measure of stability. The peak time difference of the EMG electromyography signals of the biceps brachii and triceps brachii was calculated as the muscle synergistic efficiency. The system generates time series data including elbow flexion angle, chin over bar range, stability, and muscle coordination efficiency. This system transforms traditional subjective judgment into data quantification, increasing the consistency of results from different evaluators to over 95% and achieving a timing accuracy of 0.01 seconds.

[0020] Use an AI-based fusion model to address the lack of motion quality assessment; During the training of the fusion model, 5,000 sets of data on flexed arm hangs in different scenarios, namely physical fitness assessment, fitness, and rehabilitation, were collected. The training labels were marked as effective movements, ineffective movements (such as swing compensation and half-flexed arm), and risky movements (such as excessive shoulder joint movement).

[0021] The time series data is input into the CNN-LSTM fusion model. The CNN convolutional model is used to extract the feature vectors at each time step. The LSTM long short-term memory network is used to learn the temporal features of the feature vectors. The action effectiveness score and risk level are output as intelligent evaluation results. The action effectiveness score is between 0 and 100, and the risk level includes low, medium and high levels.

[0022] In this embodiment of the invention, the CNN-LSTM fusion model is deployed on an edge computing terminal with a processing latency of ≤50ms. It is used for real-time data analysis. Preferably, it can analyze data in real time or detect an elbow angle >90° for 2 seconds before calling the analysis of the fusion model and issuing a warning through the terminal.

[0023] In step S140, the preset scenario-based template is used to match the quantitative analysis results and the intelligent evaluation results to obtain the motion detection results, and the motion detection results are archived and stored, specifically including: Preferably, a comprehensive score is obtained by weighting all quantitative analysis results and using the comprehensive score as a quantitative indicator in the template. The weight of different indicators is adjusted in combination with the athlete's basic information such as age, weight, and training goals to avoid a one-size-fits-all judgment standard. Users can select preset scenario templates based on their actual needs. The preset scenario templates include physical fitness assessment templates, fitness training templates, and rehabilitation training templates, in descending order of threshold strictness. The quantitative analysis results and intelligent evaluation results are compared with the threshold ranges of each indicator in the preset scenario template. An example of the template content is as follows: A physical fitness assessment template with strict thresholds: The requirements are: elbow flexion angle ≤90°, chin clearance over bar ≥2cm, stability ≤5° / s, muscle coordination efficiency ≤0.3 seconds, and movement effectiveness score between 90-100 points. The risk level is low. Fitness training modules with medium thresholds: The requirements are: elbow flexion angle ≤100°, chin clearance over bar ≥2cm, stability ≤3° / s, muscle coordination efficiency ≤0.3 seconds, and movement effectiveness score between 80-100 points. The risk level is low. A rehabilitation training template with a lenient threshold: The requirements are: elbow flexion angle ≤120°, chin clearance over bar ≥2cm, stability ≤1° / s, muscle coordination efficiency ≤0.3 seconds, and movement effectiveness score between 70-100 points. The risk level is low. In this embodiment of the invention, users can manually adjust parameters such as elbow angle threshold, chin over bar amplitude, and stability requirements, and can also customize and add constraints, such as setting rules to prohibit swaying, to meet the differentiated needs of different scenarios.

[0024] The report outputs motion detection results in the form of a report, which includes quantitative index curves, motion effectiveness scores, and risk warning records. In addition to reports, the archive storage includes motion videos from cameras encoded and compressed with H.265, raw data from multiple sensors, and trend charts of indicator changes generated weekly / monthly. Preferably, based on the archive data, targeted suggestions are generated through a rule engine. For example, if "the balance of force exertion in both hands is >20%", it is recommended to "increase unilateral dumbbell curl training to strengthen the weaker arm". If "the trunk stability is not up to standard", it is recommended to "add plank training to improve core stability". The guidance plan is presented on the training terminal in the form of text and dynamic illustrations, and can be downloaded to mobile phones, forming a closed loop of detection-analysis-guidance-review, which solves the problem that traditional equipment only outputs results without improvement suggestions.

[0025] Method Example 2 The implementation of this invention relies on a three-layer architecture of hardware acquisition layer - software processing layer - application output layer. The selection, deployment and parameter configuration of the components of each layer are as follows to meet the high-precision detection requirements of flexed arm suspension motion. 4K high-definition cameras are deployed directly in front of the horizontal bar training area, 3 meters away from the bar and 1.6 meters high, using an integrated fixed design. The lenses focus on the upper body movement area of ​​the trainee. Four pressure sensors are built into the horizontal bar, evenly distributed along the bar at 18cm intervals. The sensor surfaces are covered with anti-slip silicone pads, and wires connect to a data acquisition box, which is placed on the bar base. Two inertial measurement units (IMUs) are fixed to the trainee's waist (4cm directly behind the navel) and upper arm (midpoint of the lateral biceps brachii) with elastic straps, close to the skin. The strap tightness should be comfortable without causing significant discomfort. Two sets of electromyography (EMG) sensors are attached to the biceps brachii (midpoint of the anterior upper arm) and triceps brachii (midpoint of the posterior upper arm), respectively. The skin is cleaned and conductive paste is applied before attachment. The data acquisition box is placed inside the bar base, with a USB 3.0 interface connecting the sensors and IMUs, a gigabit Ethernet connection connecting the cameras, and a built-in lithium battery. The model training terminal is placed 1.5 meters away from the bar, 1 meter high, and supports Wi-Fi 6. Connect the data acquisition box.

[0026] The software is developed based on Python 3.9 and adopts a modular and microservice architecture. The functions related to the bent-arm suspension motion detection method are deployed on Alibaba Cloud ECS g7.xlarge (4 cores, 8GB RAM, 500GB SSD) and a training terminal. The functions are as follows: Data reception and preprocessing functions: The MQTT protocol receives hardware data, performs multi-threaded parallel processing of visual data, Gaussian filtering, kernel size=3×3 noise reduction, and Kalman filtering to correct outliers; missing video frames are filled using interpolation to ensure data integrity; Pose recognition and coordinate extraction function: Using the improved version of the OpenPose framework with the newly added single bar mask module, the human skeletal joints are detected, and the two-dimensional coordinates of the shoulder, elbow, wrist, jaw, and waist joints are extracted. The best ones are converted into three-dimensional coordinates using triangulation, with the origin being the ground directly below the midpoint of the single bar, with an accuracy error of ≤3cm. The coordinates are then completed based on the completion algorithm of human structure relationship. Quantitative indicator calculation function: Real-time calculation of core indicators; elbow joint angle is calculated using the vector angle formula (cosθ=(a b) / (|a|×|b|)); The chin clearance is calculated by the vertical distance between the chin and the bar, in cm; Trunk stability is calculated by the rate of change of the lumbar IMU tilt angle, in ° / s.

[0027] AI risk assessment function: CNN-LSTM fusion model (TensorFlow 2.8), input 10 frames of quantized index sequence, output action effectiveness score (0-100 points) and risk level (low / medium / high); the model is trained on 5000 sets of data (accuracy 92%), and generates warning instructions to the terminal when an anomaly occurs; Scene configuration and solution generation functions: Based on indicator scores and trainee basic information, a weighted algorithm generates a comprehensive score; it has three built-in templates for physical fitness assessment, fitness, and rehabilitation, and supports manual parameter adjustment.

[0028] File management function: MySQL 8.0 stores files, including trainer ID, video path, indicator table, risk record, and Redis caches high-frequency data; supports querying by ID / time, generates weekly / monthly PDF review reports, the reports include indicator curves and weakness analysis, and can be downloaded to the terminal.

[0029] The core interface of the terminal APP displays start / pause / end buttons, and refreshes training duration and effective attempts in real time. Below, it displays live feeds from three cameras, which can be switched between single or triple feeds. The dashboard displays indicators such as elbow angle and bar clearance, marked "normal / abnormal". Abnormalities trigger a red alert. Clicking on an abnormality allows viewing the historical curves of the last 10 movements. When an alert is triggered, it automatically pops up, displaying the warning type, level, and correction suggestions, and supports voice broadcast. Training plans are displayed by "daily / weekly", and clicking on them allows viewing demonstration videos. The archive page displays a review report, which can be shared to social media platforms.

[0030] In this embodiment of the invention, during the multi-source data synchronization process, all devices undergo NTP protocol and cloud calibration at a frequency of once per minute, with a timestamp error ≤10ms. A unified format timestamp is added when the data acquisition box uploads data to ensure data alignment. A 12×9 checkerboard calibration board is placed in the horizontal bar area, and 30 images from different angles are captured. The OpenCVcalibrateCamera function calculates the camera's intrinsic and extrinsic parameters to generate a calibration file, correcting coordinate deviations caused by lens distortion. The pressure sensor is calibrated by placing 5kg, 10kg, and 15kg standard weights on the horizontal bar and recording the voltage values. A linear regression fitting curve of "weight-voltage" is established to generate a calibration formula, ensuring that the pressure measurement error is ≤5N.

[0031] In the dataset construction process, the fusion model recruited 100 subjects, including 20 professional athletes, 30 military and police personnel, and 50 fitness enthusiasts. Each subject received 10 sets of data: 5 standard sets and 5 error sets. Anomaly types and risk levels were labeled, and the datasets were divided into training / validation / test sets in a 7:2:1 ratio. CNN was used to extract spatial features, LSTM to extract temporal features, a Softmax output layer, a cross-entropy loss function, and an Adam optimizer with a learning rate of 0.001 and a decay rate of 0.0001. An early stopping strategy was employed: if the validation set loss did not decrease after 5 rounds, the model was stopped to prevent overfitting. The final test set accuracy was 92%, and the recall rate was 90%. Model deployment required conversion to ONNX format, and TensorRT was used to accelerate inference, reducing inference time from 50ms to 15ms, meeting the real-time warning requirement of latency ≤50ms.

[0032] In this embodiment of the invention, taking a high school physical fitness test scenario as an example, the complete implementation process is as follows: Before the exercise begins, staff turn on the data collection box, camera, and terminal. The APP device performs a self-check to confirm that the connection is normal, that is, the camera video is smooth and the sensor data is refreshed. If the pressure sensor has no data, check the wire connection and unplug and plug it back in. Students enter their name, student ID, age, and weight on the terminal, and the APP queries historical data; staff assist in wearing the IMU and EMG sensors to ensure they fit snugly against the skin; students complete three standard movements as prompted: standing, raising arms, and gripping the bar; the APP automatically calibrates the joint coordinates and pressure sensors; once calibration and preparation are complete, the test can begin. Staff click "Start Testing," the camera records video, and the sensors collect data; the student follows the APP prompts to grip the bar and begin hanging with bent arms, and the terminal displays the elbow angle, bar clearance, and trunk stability in real time. Normal indicators are green, and abnormal indicators are red warnings. At 30 seconds into the test, the student's trunk tilt rate increased to 8° / s, triggering a medium-risk warning from the AI ​​module. The terminal displayed a message: "Tortoise swaying is excessive, medium risk level. Please tighten your core muscles and maintain body stability," along with a voice announcement. After the student adjusted, the stability dropped to 4° / s, and the warning was lifted. The data acquisition box uploaded 10 seconds of video footage and sensor data to the cloud every 10 seconds, and the software layer calculated the indicators in real time and updated them to the terminal. Staff monitored the test status through the terminal, pausing the test and providing guidance when the risk was high.

[0033] Once the test is complete, the cloud-based system processes the data to generate an archive and test report, displaying "Effective suspension time 45 seconds, elbow angle compliance rate 90%, trunk stability good, test result qualified." Students can view the report on their terminals and download PDF reports. Staff export all student test results to create a class test summary table. Staff remove the sensors, turn off the equipment power, and clean the sweat from the sensor surfaces to prepare for the next batch of tests.

[0034] In summary, to address the existing problems, this invention proposes a flexed-arm suspension motion detection method. During visual processing, it employs a completion algorithm based on human anatomy to complete the coordinates, ensuring that the completed shoulder joint points conform to human anatomy and guaranteeing the authenticity of the output data. Through image processing and multi-source sensor equipment, it integrates visual, mechanical, and physiological data to construct a multi-source evaluation method, resolving the issues of subjective evaluation and insufficient accuracy. An intelligent evaluation using a CNN-LSTM fusion model can simultaneously capture instantaneous anomalies and long-term trends, forming perfect functional complementarity. A configurable evaluation system alleviates the problem of limited scene adaptability. Finally, the motion detection results are archived, achieving a leap from single-result recording to full-cycle optimization.

[0035] Device Examples According to an embodiment of the present invention, a device for detecting the motion of a bent-arm suspension is provided. Figure 3 This is a schematic diagram of the bent-arm suspension motion detection device according to an embodiment of the present invention, as shown below. Figure 3 As shown, the flexed arm suspension motion detection device according to an embodiment of the present invention specifically includes: Image processing module 30 is used to acquire image data through a camera deployed directly in front of the bent-arm hanger, identify the athlete's joint points in the image data using a target recognition model, and perform coordinate completion using a completion algorithm based on human structural relationships when the shoulder joint point is obscured by the horizontal bar. Specifically, it is used for: Image data is captured using a camera configured at 60 frames per second and in 4K resolution, with the camera positioned parallel to the horizontal bar used for the bent-arm hang exercise.

[0036] The OpenPose target recognition model uses the newly added horizontal bar apparatus mask module to identify the athlete's joint points. The horizontal bar is identified by the horizontal bar apparatus mask module to distinguish it from the athlete. The joint points include the shoulder joint point, elbow joint point, wrist joint point, and jaw point. After the athlete has taken the swing but has not yet started the flexed arm hanging motion, fix the spatial coordinates of the wrist joint point, and use the spatial coordinates of the shoulder joint point and elbow joint point to mark the forearm length and upper arm length, thus forming the human body structural relationship. When the arm-bending hanging exercise begins and the shoulder joint is obscured by the horizontal bar, the spatial direction of the upper arm movement is identified by the inertial measurement unit (IMU) sensor worn on the upper arm. The spatial coordinates of the shoulder joint are determined based on the current spatial coordinates and spatial direction of the elbow joint and the length of the upper arm. The obtained shoulder joint is smoothed by a Kalman filter, and it is determined whether the current forearm length and upper arm length are within the preset range of the values ​​at the time of calibration. If they are not within the preset range, manual coordinate completion is performed.

[0037] The multimodal sensing module 32 is used to acquire the athlete's motion characteristic data through a multimodal sensor, specifically for: The torso tilt angle was measured by an inertial measurement unit (IMU) sensor worn on the athlete's waist, and the elbow joint angular velocity was measured by an inertial measurement unit (IMU) sensor worn on the athlete's upper arm. EMG electromyography signals were measured using an electromyography sensor worn on the upper arm of the athlete. The force is measured by a pressure sensor deployed on a horizontal bar.

[0038] The initial analysis module 34 is used to perform quantitative analysis based on key point and motion feature data, forming a time series including the quantitative analysis results. The time series is then input into a fusion model that includes feature extraction and sequence analysis, outputting intelligent evaluation results. Specifically, it is used for: Using the elbow joint as the vertex and the shoulder and wrist joints as the edges, calculate the angle between the vectors as the elbow flexion angle. Calculate the vertical distance between the mandibular point and the horizontal bar as the mandibular clearance. Calculate the rate of change of the trunk tilt angle as a measure of stability. The peak time difference of the EMG electromyography signals of the biceps brachii and triceps brachii was calculated as the muscle synergistic efficiency. The time series of data includes elbow flexion angle, jaw clearance over bar, stability, and muscle synergy efficiency.

[0039] The time series data is input into the CNN-LSTM fusion model. The CNN convolutional model is used to extract the feature vectors at each time step. The LSTM long short-term memory network is used to learn the temporal features of the feature vectors. The action effectiveness score and risk level are output as intelligent evaluation results.

[0040] Template matching module 36 is used to match the quantitative analysis results and intelligent evaluation results using preset scenario-based templates to obtain motion detection results, and archive and store the motion detection results. Specifically, it is used for: Select a preset scenario template, which includes, in descending order of threshold strictness, a physical fitness assessment template, a fitness training template, and a rehabilitation training template. The quantitative analysis results and intelligent evaluation results are compared with the threshold ranges of each indicator in the preset scenario template, and the motion detection results are output in the form of a report.

[0041] In summary, to address the existing problems, this invention, a flexed-arm suspension motion detection device, employs a coordinate completion algorithm based on human anatomy during visual processing. This ensures that the completed shoulder joint points conform to human anatomy, guaranteeing the authenticity of the output data. Through image processing and multi-source sensor equipment, it integrates visual, mechanical, and physiological data to construct a multi-source evaluation method, resolving the issues of subjective evaluation and insufficient accuracy. The use of a CNN-LSTM fusion model for intelligent evaluation simultaneously captures instantaneous anomalies and long-term trends, forming perfect functional complementarity. A configurable evaluation system alleviates the problem of limited scene adaptability. Finally, the motion detection results are archived, achieving a leap from single-result recording to full-cycle optimization.

[0042] 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.

[0043] 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.

[0044] 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.

[0045] 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 the bent-arm suspension motion detection method in any of the exemplary embodiments described above.

[0046] 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 above exemplary embodiments of the bent-arm suspension motion detection method.

[0047] 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.

[0048] 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.

[0049] 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 motion during bent-arm suspension, characterized in that, include: Image data is acquired by a camera deployed directly in front of the bent-arm hanger. The joints of the athlete in the image data are identified using a target recognition model. When the shoulder joint is obscured by the horizontal bar, a completion algorithm based on human body structure is used to complete the coordinates. Acquire motion characteristic data of athletes using multimodal sensors; Based on the key points and the motion feature data, a quantitative analysis is performed to form a time series including the quantitative analysis results. The time series is then input into a fusion model that includes feature extraction and sequence analysis to output intelligent evaluation results. The quantitative analysis results and the intelligent evaluation results are matched using a preset scenario template to obtain motion detection results, which are then archived and stored.

2. The method according to claim 1, characterized in that, The acquisition of image data by a camera deployed directly in front of the bent-arm suspension specifically includes: Image data is captured by a camera configured at 60 frames per second and 4K resolution, wherein the camera is deployed parallel to the horizontal bar used for bent-arm hanging exercises.

3. The method according to claim 1, characterized in that, The step of using a target recognition model to identify the athlete's joint points in the image data, and employing a coordinate completion algorithm based on human structural relationships to complete the coordinates when the shoulder joint point is obscured by the horizontal bar, specifically includes: The OpenPose target recognition model uses a newly added horizontal bar equipment mask module to identify the athlete's joint points. The horizontal bar is identified by the horizontal bar equipment mask module to distinguish it from the athlete. The joint points include the shoulder joint point, elbow joint point, wrist joint point, and jaw point. After the athlete has taken off the bar but has not yet started the flexed arm hanging motion, the spatial coordinates of the wrist joint are fixed, and the forearm length and upper arm length are determined by the spatial coordinates of the shoulder joint and elbow joint to form the human body structure relationship. When the arm-bending hanging exercise begins and the shoulder joint is obscured by the horizontal bar, the spatial direction of the upper arm movement is identified by the inertial measurement unit (IMU) sensor worn on the upper arm. The spatial coordinates of the shoulder joint are determined based on the current spatial coordinates of the elbow joint, the spatial direction, and the upper arm length. The obtained shoulder joint is smoothed by a Kalman filter, and it is determined whether the current forearm length and upper arm length are within the preset range of the values ​​at calibration. If they are not within the preset range, manual coordinate completion is performed.

4. The method according to claim 1, characterized in that, The acquisition of athlete motion characteristic data through multimodal sensors specifically includes: The torso tilt angle was measured by an inertial measurement unit (IMU) sensor worn on the athlete's waist, and the elbow joint angular velocity was measured by an inertial measurement unit (IMU) sensor worn on the athlete's upper arm. EMG electromyography signals were measured using an electromyography sensor worn on the upper arm of the athlete. The force is measured by a pressure sensor deployed on a horizontal bar.

5. The method according to any one of claims 3 or 4, characterized in that, The step of performing quantitative analysis based on the joint points and the motion feature data to form a time series including the quantitative analysis results specifically includes: Using the elbow joint as the vertex and the shoulder joint and wrist joint as the edge points, calculate the included vector angle as the elbow joint bending angle. The vertical distance between the mandibular point and the horizontal bar is calculated as the mandibular clearance amplitude. Calculate the rate of change of the trunk tilt angle as the degree of stability; The peak time difference of the EMG electromyography signals of the biceps brachii and triceps brachii is calculated as the muscle synergistic efficiency. A time series is formed, including the elbow flexion angle, the chin over the bar amplitude, the stability, and the muscle coordination efficiency.

6. The method according to claim 1, characterized in that, The step of inputting the time series data into a fusion model that includes feature extraction and sequence analysis, and outputting intelligent evaluation results, specifically includes: The time series is input into the CNN-LSTM fusion model, the CNN convolutional model is used to extract the feature vectors at each time step, the LSTM long short-term memory network is used to learn the temporal features of the feature vectors, and the action effectiveness score and risk level are output as intelligent evaluation results.

7. The method according to claim 1, characterized in that, The step of matching the quantitative analysis results and the intelligent evaluation results using a preset scenario template to obtain the motion detection results specifically includes: Select a preset scenario template, wherein the preset scenario templates include physical fitness assessment template, fitness training template and rehabilitation training template in descending order of threshold strictness; The quantitative analysis results and the intelligent evaluation results are compared with the threshold ranges of each indicator in the preset scenario template, and the motion detection results are output in the form of a report.

8. A device for detecting motion during bent-arm suspension, characterized in that, include: The image processing module is used to acquire image data by a camera deployed directly in front of the bent-arm hang, use a target recognition model to identify the joints of the athlete in the image data, and use a completion algorithm based on human body structure relationship to complete the coordinates when the shoulder joint is obscured by the horizontal bar. The multimodal sensing module is used to acquire motion characteristic data of athletes through multimodal sensors; The initial analysis module is used to perform quantitative analysis based on the joint points and the motion feature data, form a time series including the quantitative analysis results, input the time series into a fusion model including feature extraction and sequence analysis, and output intelligent evaluation results. The template matching module is used to match the quantitative analysis results and the intelligent evaluation results using a preset scenario-based template to obtain motion detection results, and to archive 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 flexed arm suspension motion 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 information transmission implementation program, which, when executed by a processor, implements the steps of the flexed arm suspension motion detection method as described in any one of claims 1 to 7.

Images