Supine crunch movement detection method, system and equipment and medium

Abstract

The embodiment of the invention provides a supine crunch movement detection method, system and device and a medium, and the method comprises the steps: synchronously collecting multi-view video data of the supine crunch movement of a trainee through two cameras deployed in different directions of the trainee; preprocessing the collected multi-view video data, identifying the preprocessed video data based on a deep learning posture identification model, and extracting a plurality of joint point coordinates of a trainer; based on the joint point coordinates, performing quantitative judgment on the initial supine state, the crunch process and the lowering process of the supine crunch action to judge whether the action reaches the standard or not, and performing effective counting or non-standard action marking according to an action judgment result; on the basis of the joint point coordinates and the time sequence data, at least one quantitative index of the supine crunch movement is calculated, the quantitative index comprises the movement speed, the force generation balance and the waist arching height, and according to the quantitative index and a movement judgment result, real-time risk early warning and training guidance are output.

Description

Technical Field

[0001] This document relates to the field of computer vision recognition technology for sports, and in particular to a method, system, device and medium for detecting supine abdominal crunches. Background Technology

[0002] The supine crunch is a classic training and assessment exercise for evaluating core muscle strength and trunk control. It is widely used in fitness and body shaping, rehabilitation training, and military / police physical fitness testing. The standard procedure is as follows: The trainee lies supine on a flat surface with knees bent at 90 degrees and feet flat on the ground. Hands lightly support the sides of the ears. Using abdominal muscle contraction, the trainee lifts the upper back off the ground, then slowly lowers back to the starting supine position, completing one full repetition. While this exercise seems simple, scientific training and accurate testing are crucial to avoiding injury and maximizing results. Currently, the training and testing of supine crunches still rely on traditional methods, which have several technical shortcomings, and existing optimization technologies have not yet overcome these core limitations.

[0003] Traditional supine crunch tests rely entirely on manual observation and counting. Examiners must simultaneously assess multiple dimensions, but manual judgment of factors such as whether the lower back is fully on the ground and the height of the upper back lift lacks objective evidence, leading to misjudgments and omissions. Especially in scenarios with multiple examiners simultaneously, examiners' attention is scattered, and judgment standards fluctuate with fatigue levels, resulting in an error rate of 15%-20%, severely impacting fairness. Furthermore, traditional tests only record the single result of valid repetitions, failing to capture crucial details such as movement speed, force balance, and postural deviation. Trainees struggle to understand the root causes of their movement problems, and coaches cannot develop targeted improvement plans based on data, leading to blind repetition. This not only results in ineffective core muscle training but may also lead to compensatory habits due to incorrect force application, hindering breakthroughs in performance. Manual testing requires examiners to observe, time, and count each individual throughout the entire process, with each test taking at least 2 minutes. When facing large-scale scenarios such as military training or group classes at fitness institutions, the assessment process is lengthy and requires the simultaneous participation of a large number of examiners, resulting in high labor costs. At the same time, different examiners may have different understandings of the judgment criteria, which can easily lead to disputes in the assessment and further reduce the efficiency and credibility of the test.

[0004] In summary, current supine crunch testing technology still faces core problems such as subjective judgment, fragmented data, and inefficient operation. It lacks an integrated solution that can achieve real-time posture monitoring, accurate standard judgment, quantitative data analysis, and large-scale efficient testing. It cannot meet the needs of fitness, rehabilitation, military and police fields for scientific training and fair assessment. It is urgent to break through the existing bottlenecks through technological innovation. Summary of the Invention

[0005] This specification provides one or more embodiments of a method for detecting supine abdominal crunches, including: Two cameras deployed in different directions around the trainee simultaneously collect multi-view video data of the trainee performing supine crunches. The collected multi-view video data is preprocessed, and the preprocessed video data is identified based on a deep learning pose recognition model to extract the coordinates of multiple joints of the trainee. Based on the coordinates of the joints, the initial supine state, the crunching process, and the lowering process of the supine crunch are quantitatively judged to determine whether the movement meets the standard. Based on the movement judgment results, effective counts or non-standard movements are marked. Based on joint coordinates and time series data, calculate at least one quantitative indicator for supine abdominal crunches, including movement speed, force balance, and lumbar arch height. Based on the quantitative indicators and movement judgment results, output real-time risk warnings and training guidance.

[0006] Furthermore, of the two cameras, the first camera has a shooting angle perpendicular to the trainee's torso and is used to collect data on the fit between the waist and the training mat; the second camera has a shooting angle parallel to the trainee's torso and is used to collect data on the joints of the scapula, head, and hands. During the data acquisition process, metadata from both cameras, including frame rate, timestamp, and focal length, is synchronized to ensure data time consistency.

[0007] Furthermore, the collected multi-view video data is preprocessed, and the preprocessed video data is identified based on a deep learning pose recognition model to extract the coordinates of multiple key points of the trainee. Specifically, this includes: By using the camera's wide dynamic range function, supplementary lighting, and Gaussian filtering noise reduction algorithm, the impact of environmental interference on image accuracy is eliminated, thereby eliminating the interference of light and shadow on key point recognition. A two-dimensional coordinate system is established with the training mat plane as the reference. The X-axis of the two-dimensional coordinate system is parallel to the direction of the trainee's torso, and the Y-axis is perpendicular to the training mat plane. The real-time coordinates of each joint point in the coordinate system are output, including the coordinates of the head, neck, scapula, waist, sacrum, hand, knee joint and ankle joint. The Kalman filter algorithm is used to dynamically correct the key point coordinates, eliminating coordinate fluctuations caused by slight body swaying of the trainee and slight camera shake. When the trainee makes slight adjustments to the position of their hand or turns their head slightly, the tracking of joints is continuously maintained without loss by predicting the trajectory of joints in adjacent frames.

[0008] Furthermore, the quantitative determination of the initial supine state includes: Determine whether the waist is touching the ground based on the vertical distance between the waist joint point and the plane of the training mat; The determination of whether the legs are bent to a preset angle is based on the vector angle between the knee joint and the ankle joint; The hand position is determined based on the distance between the hand joints and the head joints to determine whether the hand position is compliant.

[0009] Furthermore, the quantitative determination of the abdominal crunch process includes: The vertical distance between the scapular joint and the training pad plane determines whether the scapula is off the ground at a preset height; The distance between the lumbar joint and the training mat plane determines whether the lumbar region remains in contact with the ground. The presence of excessive forward head tilt is determined by the angle between the head joints and the torso.

[0010] Furthermore, the quantitative indicators and calculation standards for exercise include: Action speed is calculated based on the time required to complete one full action cycle; Equal force exertion is calculated based on the difference in contraction speed between the left and right abdominal muscle groups. The height of the lower back arch is calculated based on the maximum vertical distance between the lower back joint and the plane of the training mat during the movement cycle.

[0011] Furthermore, the method further includes constructing and storing a training profile, which includes trainee basic information, quantitative data, key images, and risk warning records; the trainee basic information includes name and training time; the quantitative data includes the number of effective movements, the percentage of non-standard movements, and three types of effective curves; the key images include typical frames of qualified and non-compliant movements.

[0012] This specification provides one or more embodiments of a supine crunch exercise detection system, including: Data acquisition module: used to simultaneously acquire multi-view video data of the trainee performing supine crunch exercises using two cameras deployed in different directions of the trainee; Data processing module: used to preprocess the acquired multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee; Action judgment module: Based on the coordinates of the joint points, it is used to quantify and judge the initial supine state, the crunch process and the lowering process of the supine crunch, so as to determine whether the action meets the standard, and to perform effective counting or mark non-standard actions according to the action judgment results. Risk warning module: Based on joint coordinates and time series data, it calculates at least one quantitative indicator for supine abdominal crunch exercise, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, it outputs real-time risk warnings and training guidance.

[0013] This specification provides one or more embodiments of an electronic device, including: Processor; and, A memory is configured to store computer-executable instructions, which, when executed, cause the processor to perform the steps of the supine crunch exercise detection method described above.

[0014] This specification provides one or more embodiments of a storage medium for storing computer-executable instructions that, when executed, implement the steps of the supine abdominal crunch exercise detection method described above.

[0015] This invention, through a combination of multi-view visual acquisition and deep learning analysis, achieves objective, precise, and intelligent detection of supine abdominal crunches. It transforms subjective experience-based motion judgment into quantitative analysis based on spatial coordinates and motion timing, solving the problems of inconsistent standards and low efficiency inherent in manual detection. By using dual cameras in synergy, it simultaneously captures waist alignment and upper body posture, and uses an optimized posture estimation model to stably extract key joint data, transforming motion standards into precisely measurable distance and angle thresholds, ensuring the objectivity and consistency of the judgment results. By defining multi-dimensional quantitative indicators, it achieves analysis of motion quality, automatically identifying and prompting non-standard movements, and assessing force application patterns and stability risks in real time. A three-level early warning mechanism is established, providing immediate feedback based on risk levels to effectively prevent sports injuries, comprehensively improving the scientific nature, safety, and management efficiency of training.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0017] 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.

[0018] Figure 1 A flowchart illustrating a method for detecting supine abdominal crunches provided in one or more embodiments of this specification; Figure 2 A schematic diagram of a supine abdominal crunch exercise detection system provided for one or more embodiments of this specification; Figure 3 This is a schematic diagram of the structure of an electronic device provided for one or more embodiments of this specification. Detailed Implementation

[0019] 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.

[0020] Method Implementation Examples According to an embodiment of the present invention, a method for detecting supine abdominal crunches is provided. Figure 1 A flowchart illustrating a method for detecting supine abdominal crunches provided in one or more embodiments of this specification, such as... Figure 1 As shown, the supine abdominal crunch exercise detection method according to an embodiment of the present invention specifically includes: S1. Multi-view video data of the trainee performing supine crunches is collected simultaneously by two cameras deployed in different directions from the trainee.

[0021] The two cameras are configured as follows: the first camera is positioned directly above the trainee's torso, with its optical axis perpendicular to the plane of the training mat and its shooting angle perpendicular to the trainee's torso. It is used to collect data on the fit between the waist and the training mat, and to identify whether the waist has an abnormal arch or gap due to compensatory force. The sacrum or lower back joints are usually used as references. The second camera is positioned to the side of the trainee's torso, with its optical axis parallel to the long axis of the torso and at a certain height from the training mat. Its shooting angle is parallel to the trainee's torso. It is used to collect joint data of the scapula, head, and hands, track the movement details of the upper body and limbs, and accurately capture the height of both scapulae off the ground, the spatial posture of the head and neck, including the angles of forward tilt and backward tilt, and whether the hands are placed beside the ears or improperly clasped behind the head.

[0022] To ensure that the data from the two cameras are strictly aligned in the time dimension and form a unified analysis frame sequence, metadata synchronization is performed during the acquisition process. This includes calibrating the acquisition frame rate of the two cameras and adding a unified timestamp to each frame. In addition, intrinsic parameters such as focal length that affect spatial coordinate calculation are collaboratively calibrated to ensure data temporal consistency.

[0023] S2. Preprocess the collected multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee.

[0024] After acquiring the original multi-view video stream, considering potential strong light reflections or low light conditions indoors, the camera's wide dynamic range (WDR) and controllable supplementary lighting unit are comprehensively utilized to eliminate the impact of environmental interference on image accuracy, thereby eliminating the interference of light and shadow on keypoint recognition. The camera's WDR function can compress extreme brightness contrasts in the scene, preserving details in shadow and highlight areas. The controllable supplementary lighting unit, including supplementary lights, can increase the light intensity to the minimum threshold required for analysis. Subsequently, digital image processing algorithms such as Gaussian filtering are applied to smooth and denoise the frame images, removing random noise and irrelevant textures.

[0025] The preprocessed video frames are fed in real-time into a deep learning posture recognition model. This model, trained on massive amounts of human posture data and supplemented with a library of supine crunch-specific exercises, can locate and identify key anatomical joints closely related to motion analysis within a single pixel. These include the lumbar and sacral regions for assessing core stability, the scapula for determining range of motion, the head and neck for monitoring compensation, and the hands, knees, and ankles for confirming posture compliance. To convert the identified pixel locations into motion data, a two-dimensional coordinate system is established with the training mat plane as the reference. The X-axis of this coordinate system is parallel to the trainee's torso, describing the forward and backward displacement of the body; the Y-axis is perpendicular to the training mat plane, used to measure the vertical height of each joint. Each identified joint is mapped into this coordinate system, and the real-time coordinates of each joint in the coordinate system are output.

[0026] The original coordinate sequence may contain high-frequency noise introduced by slight human body tremors, breathing fluctuations, or the camera's own slight vibrations. To address this noise, a Kalman filter algorithm is introduced. Based on the kinematic model of the joints, the coordinates of the current frame are predicted and fused with the actual values ​​identified by the model. This distinguishes between real limb movements and meaningless random jitters, and dynamically corrects the joint coordinates.

[0027] To address the possibility of temporary occlusion during actual movement, such as when a trainee makes a slight adjustment to their hand position or a slight turn of their head, causing a joint to be temporarily lost or have low confidence in a particular frame, the current position of the joint is estimated using interpolation or a short-term prediction model based on the joint's trajectory prediction in adjacent frames and its speed and direction of movement in the preceding frames. This ensures that joint tracking is continuous and not lost.

[0028] S3. Based on the joint coordinates, the initial supine state, the crunching process, and the lowering process of the supine crunch are quantitatively judged to determine whether the movement meets the standard. Based on the movement judgment results, effective counts or non-standard movement marks are made.

[0029] Quantitative determination of the initial supine position includes: First, the vertical distance between the lumbar joint and the training mat plane is used to determine if the lower back is touching the ground. The lumbar joint is typically located in the lower back or above the hips. The Y-axis coordinate of the lumbar joint in the established two-dimensional coordinate system is calculated and compared with the training mat plane to determine if the vertical distance is less than a preset minimum threshold. The training mat plane has Y=0, and the minimum threshold is set to 0.5 cm. Simultaneously, by connecting the hip, knee, and ankle joints to form vectors for the knee and ankle joints, the angle between these vectors is calculated to quantify the knee flexion angle and determine if both legs are bent to a preset angle. Furthermore, by calculating the Euclidean distance between the hand and head joints (specifically, the Euclidean distance between the wrist and chin), the compliance of hand position is determined, identifying and eliminating the common improper use of hands behind the head.

[0030] After quantitatively assessing whether the initial supine position meets the standard, the abdominal crunch process is quantitatively assessed, including: The system continuously tracks the vertical displacement of the scapular joint relative to the mat surface. Based on the vertical distance between the scapular joint and the training mat plane, it determines whether the scapula has lifted off the ground to a preset height. When the lift height consistently exceeds the preset value, it is considered a valid scapular lift. Simultaneously, the Y-axis coordinate of the lumbar joint is monitored, requiring it to remain within a very small fluctuation range. The distance between the lumbar joint and the training mat plane determines whether the lumbar region remains on the ground. At the same time, by analyzing the angle of the vector formed by the directions of the head and torso relative to the vertical direction, the forward head tilt angle is calculated in real time. If this angle exceeds a preset safety threshold, it is considered a risky movement involving excessive forward head tilt, i.e., compensating with neck muscles. The direction of the torso is the line connecting the neck and hips.

[0031] In this embodiment, the quantitative judgment logic for the three stages of initialization, crunching, and lowering is shown in the table below:

[0032] S4. Based on joint coordinates and time series data, calculate at least one quantitative indicator for supine abdominal crunch exercise, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, output real-time risk warnings and training guidance.

[0033] The quantitative indicators for supine abdominal crunches include movement speed, force balance, and the height of the lower back arch, and their calculation standards are as follows: Action speed is calculated based on the duration of completing a full action cycle. By capturing a full action cycle, that is, the start and end frames from the initial supine position through the peak contraction and back to the initial state, combined with the video frame rate, the time taken to complete a single action is accurately calculated, which can then be converted into the number of times per second.

[0034] The balance of force exertion is calculated based on the difference in contraction speed between the abdominal muscle groups on the left and right sides of the body. By analyzing the relative kinematic changes of the corresponding joints on the left and right sides of the trunk during the crunch, the difference in contraction speed between the internal and external oblique muscles and the rectus abdominis muscles on both sides is indirectly calculated. By calculating the percentage of this difference, the degree of imbalance in force exertion on both sides is quantified.

[0035] The height of the lumbar arch is calculated based on the maximum vertical distance between the lumbar joint and the training pad plane during the movement cycle; the vertical distance between the lumbar joint and the training pad plane is continuously recorded, and the maximum value is extracted. This maximum value is the severity of the failure of the abdominal core muscles, which leads to excessive forward curvature of the lumbar spine.

[0036] The method further includes constructing and storing a training profile, which includes trainee basic information, quantitative data, key images, and risk warning records; the trainee basic information includes name and training time; the quantitative data includes the number of effective movements, the percentage of non-standard movements, and three types of effective curves; the key images include typical frames of qualified and non-compliant movements.

[0037] The beneficial effects of this invention are as follows: This invention, through a combination of multi-view visual acquisition and deep learning analysis, achieves objective, precise, and intelligent detection of supine abdominal crunches. It transforms subjective experience-based motion judgment into quantitative analysis based on spatial coordinates and motion timing, solving the problems of inconsistent standards and low efficiency inherent in manual detection. By using dual cameras in synergy, it simultaneously captures waist alignment and upper body posture, and uses an optimized posture estimation model to stably extract key joint data, transforming motion standards into precisely measurable distance and angle thresholds, ensuring the objectivity and consistency of the judgment results. By defining multi-dimensional quantitative indicators, it achieves analysis of motion quality, automatically identifying and prompting non-standard movements, and assessing force application patterns and stability risks in real time. A three-level early warning mechanism is established, providing immediate feedback based on risk levels to effectively prevent sports injuries, comprehensively improving the scientific nature, safety, and management efficiency of training.

[0038] System Implementation Examples According to an embodiment of the present invention, a supine abdominal crunch exercise detection system is provided. Figure 2 This is a schematic diagram illustrating the composition of a supine abdominal crunch exercise detection system provided in one or more embodiments of this specification, as shown below. Figure 3 As shown, the supine abdominal crunch exercise detection system according to an embodiment of the present invention specifically includes: Data acquisition module 20: used to simultaneously acquire multi-view video data of the trainee performing supine crunch exercises using two cameras deployed in different directions of the trainee; Data processing module 22: used to preprocess the acquired multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee; Action judgment module 24: Based on the coordinates of the joint points, it is used to quantify and judge the initial supine state, the crunching process and the lowering process of the supine crunch, so as to judge whether the action meets the standard, and to perform effective counting or mark non-standard actions according to the action judgment results. Risk warning module 26: It is used to calculate at least one quantitative indicator of supine abdominal crunch based on joint coordinates and time series data, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, it outputs real-time risk warning and training guidance.

[0039] The embodiments of the present invention are system embodiments corresponding to the above method embodiments. The specific operation of each module can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0040] Device Example 1 This invention provides an electronic device, such as... Figure 3 As shown, it includes: a memory 30, a processor 32, and a computer program stored in the memory 30 and executable on the processor 32. When the computer program is executed by the processor 32, it performs the following method steps: S1. Multi-view video data of the trainee performing supine crunches is collected simultaneously by two cameras deployed in different directions from the trainee. S2. Preprocess the collected multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee; S3. Based on the joint coordinates, the initial supine state, the crunching process, and the lowering process of the supine crunch are quantitatively judged to determine whether the movement meets the standard. Based on the movement judgment results, effective counts or non-standard movement marks are made. S4. Based on joint coordinates and time series data, calculate at least one quantitative indicator for supine abdominal crunch exercise, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, output real-time risk warnings and training guidance.

[0041] Device Example 2 This invention provides a computer-readable storage medium storing an information transmission implementation program. When executed by a processor 32, the program performs the following method steps: S1. Multi-view video data of the trainee performing supine crunches is collected simultaneously by two cameras deployed in different directions from the trainee. S2. Preprocess the collected multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee; S3. Based on the joint coordinates, the initial supine state, the crunching process, and the lowering process of the supine crunch are quantitatively judged to determine whether the movement meets the standard. Based on the movement judgment results, effective counts or non-standard movement marks are made. S4. Based on joint coordinates and time series data, calculate at least one quantitative indicator for supine abdominal crunch exercise, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, output real-time risk warnings and training guidance.

[0042] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0043] 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 supine abdominal crunches, characterized in that, include: Two cameras deployed in different directions around the trainee simultaneously collect multi-view video data of the trainee performing supine crunches. The collected multi-view video data is preprocessed, and the preprocessed video data is identified based on a deep learning pose recognition model to extract the coordinates of multiple joints of the trainee. Based on the coordinates of the joints, the initial supine state, the crunching process, and the lowering process of the supine crunch are quantitatively judged to determine whether the movement meets the standard. Based on the movement judgment results, effective counts or non-standard movements are marked. Based on joint coordinates and time series data, calculate at least one quantitative indicator for supine abdominal crunches, including movement speed, force balance, and lumbar arch height. Based on the quantitative indicators and movement judgment results, output real-time risk warnings and training guidance.

2. The method according to claim 1, characterized in that, The two cameras are configured such that the first camera's shooting angle is perpendicular to the trainee's torso and is used to collect data on the fit between the waist and the training mat; the second camera's shooting angle is parallel to the trainee's torso and is used to collect data on the joints of the scapula, head, and hands. During the data acquisition process, metadata from both cameras, including frame rate, timestamp, and focal length, is synchronized to ensure data time consistency.

3. The method according to claim 1, characterized in that, The collected multi-view video data is preprocessed, and the preprocessed video data is identified based on a deep learning pose recognition model. The extraction of multiple keypoint coordinates of the trainee specifically includes: By using the camera's wide dynamic range function, supplementary lighting, and Gaussian filtering noise reduction algorithm, the impact of environmental interference on image accuracy is eliminated, thereby eliminating the interference of light and shadow on key point recognition. A two-dimensional coordinate system is established with the training mat plane as the reference. The X-axis of the two-dimensional coordinate system is parallel to the direction of the trainee's torso, and the Y-axis is perpendicular to the training mat plane. The real-time coordinates of each joint point in the coordinate system are output, including the coordinates of the head, neck, scapula, waist, sacrum, hand, knee joint and ankle joint. The Kalman filter algorithm is used to dynamically correct the key point coordinates, eliminating coordinate fluctuations caused by slight body swaying of the trainee and slight camera shake. When the trainee makes slight adjustments to the position of their hand or turns their head slightly, the tracking of joints is continuously maintained without loss by predicting the trajectory of joints in adjacent frames.

4. The method according to claim 3, characterized in that, The quantitative determination of the initial supine state includes: Determine whether the waist is touching the ground based on the vertical distance between the waist joint point and the plane of the training mat; The determination of whether the legs are bent to a preset angle is based on the vector angle between the knee joint and the ankle joint; The hand position is determined based on the distance between the hand joints and the head joints to determine whether the hand position is compliant.

5. The method according to claim 3, characterized in that, The quantitative assessment of the abdominal crunch process includes: The vertical distance between the scapular joint and the training pad plane determines whether the scapula is off the ground at a preset height; The distance between the lumbar joint and the training mat plane determines whether the lumbar region remains in contact with the ground. The presence of excessive forward head tilt is determined by the angle between the head joints and the torso.

6. The method according to claim 1, characterized in that, The quantitative indicators and calculation standards for exercise include: Action speed is calculated based on the time required to complete one full action cycle; Equal force exertion is calculated based on the difference in contraction speed between the left and right abdominal muscle groups. The height of the lower back arch is calculated based on the maximum vertical distance between the lower back joint and the plane of the training mat during the movement cycle.

7. The method according to claim 1, characterized in that, The method further includes constructing and storing a training profile, which includes trainee basic information, quantitative data, key images, and risk warning records; the trainee basic information includes name and training time; the quantitative data includes the number of effective movements, the percentage of non-standard movements, and three types of effective curves; the key images include typical frames of qualified and non-compliant movements.

8. A supine abdominal crunch exercise detection system, characterized in that, include: Data acquisition module: used to simultaneously acquire multi-view video data of the trainee performing supine crunch exercises using two cameras deployed in different directions of the trainee; Data processing module: used to preprocess the acquired multi-view video data, identify the preprocessed video data based on the deep learning pose recognition model, and extract the coordinates of multiple joint points of the trainee; Action judgment module: Based on the coordinates of the joint points, it is used to quantify and judge the initial supine state, the crunch process and the lowering process of the supine crunch, so as to determine whether the action meets the standard, and to perform effective counting or mark non-standard actions according to the action judgment results. Risk warning module: Based on joint coordinates and time series data, it calculates at least one quantitative indicator for supine abdominal crunch exercise, including movement speed, force balance and lumbar arch height. Based on the quantitative indicators and movement judgment results, it outputs real-time risk warnings and training guidance.

9. An electronic device, characterized in that, include: processor; as well as, A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the steps of the supine crunch exercise detection method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, Used to store computer-executable instructions, which, when executed, implement the steps of the supine crunch exercise detection method as described in any one of claims 1 to 7.

Images