Physical fitness testing equipment and working method

Abstract

The invention belongs to the field of physical fitness testing equipment, and discloses physical fitness testing equipment and a working method, which realize full-automatic identification and standardized counting of physical fitness testing actions, support various physical fitness testing items, do not need manual intervention, greatly reduce the possibility of manual participation and errors, and improve the working efficiency. Compared with a traditional manual counting mode which depends on personal experience of supervisors and subjective judgment, the problems that results are poor in repeatability, standards are not uniform, and missing recording and mistaken recording are prone to occurring are effectively solved. Through deep learning key point detection and motion classification based on the KNN algorithm, the motion type can be identified more accurately, motion details can be analyzed, whether each motion meets the standard or not can be accurately judged, the motion standardization judgment precision is improved, and the problem that a traditional visual method is difficult to accurately identify the motion details and normalization is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of physical fitness testing equipment, and particularly relates to a physical fitness testing equipment and a working method. BACKGROUND

[0002] Physical fitness testing (physical testing) is an important way to evaluate individual physical fitness and sports ability, and is widely used in school sports, professional selection, police fitness examination and mass fitness fields. Common physical testing projects such as pull-ups, push-ups and sit-ups have strict requirements for the standardization of movements and the number of times completed. With the development of digital and intelligent sports, how to realize scientific management and efficient execution of the whole process of physical testing has become the focus of the industry.

[0003] At present, the technical solutions for physical testing motion detection and counting mainly include the following categories: 1. Manual counting and subjective judgment: This is the most traditional method, which is observed and judged by the examiner or supervisor through naked eye observation to determine whether each movement is standardized and counted manually. Although this method is simple and low in cost, it is easily affected by human fatigue, subjective experience, etc., resulting in counting errors and non-uniform evaluation standards. In large-scale physical testing or high-frequency testing scenarios, manual methods are low in efficiency and high in management cost, and are difficult to meet the needs of digital and intelligent sports management.

[0004] 2. Sensor-assisted detection: Some solutions use wearable sensors (such as accelerometers, gyroscopes, etc.) fixed on key parts of the athlete's body to complete motion counting and standardization judgment through analysis of sensor data. This method improves the accuracy of motion detection to some extent, but wearing sensors is cumbersome and affects the experience, and is affected by factors such as wearing position, body size difference, etc., making it difficult to fully cover all standards of complex physical testing movements, and the scope of application is limited.

[0005] 3. Motion detection based on computer vision: With the progress of computer vision technology, some solutions use cameras to collect video streams, and realize motion detection and counting through image processing and motion trajectory analysis. For example, traditional machine learning methods or image segmentation techniques are used to detect human contours or motion trajectories to roughly count the number of movements. This method realizes non-contact detection, but is usually only applicable to specific scenarios or single movements, has limited motion standardization judgment capability, and has low recognition accuracy.

[0006] 4. Human pose estimation and motion analysis based on deep learning: In recent years, some research and commercial products have introduced deep learning human pose estimation algorithms (such as OpenPose, AlphaPose, HRNet, MoveNet, etc.), which can automatically detect human key points (such as head, shoulder, elbow, hip, knee, ankle, etc.) in videos, and complete automatic recognition, standardization judgment and counting of actions by combining time series analysis methods. For example, some colleges or technology companies have tried to use open source pose estimation algorithms for automatic counting and auxiliary judgment of push-ups, squats and other actions. Such solutions have improved the accuracy and adaptability of action recognition, but they lack support for standardized judgment of diversified physical measurement actions, and have limited data management and real-time feedback functions, making it difficult to meet the needs of large-scale, full-process applications. SUMMARY

[0007] The purpose of the present application is to overcome the problems of limited accuracy and adaptability, lack of unified standards and full-process management of existing solutions, and to provide a physical fitness testing device and working method.

[0008] To achieve the above purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a physical fitness testing device, comprising: a video acquisition and preprocessing module for acquiring video data, standardizing and normalizing the format of the video data, and preprocessing the video frames; a key point detection module for acquiring the normalized video frames and locating the key points in the video frames; an action recognition and standardization judgment module for acquiring key point information, extracting sample features of the key point information, normalizing the sample features, comparing the normalized sample features with features in a preset sample library, and identifying the type of physical measurement action.

[0009] The present application further improves by including a visualization module for displaying the video data and the type of physical measurement action.

[0010] The present application further improves by implementing the functions of the video acquisition and preprocessing module through the following methods: real-time acquisition or import of local physical measurement videos by a camera as video data; format standardization and size normalization of the video data, and adjustment of the video resolution to 640x480; frame rate control, wing image denoising and image enhancement of the video frames.

[0011] The present application further improves by implementing the functions of the key point detection module through the following methods: using a deep model to process the video frames, locate the preset key points of the human body, and output the three-dimensional coordinates of the key points.

[0012] The further improvement of the present application is that the function of the action recognition and standardization determination module is realized by the following method: Obtaining key point information, analyzing the key point information to obtain time sequence coordinate data of each key point; Extracting relevant sample features in the time sequence coordinate data of each key point and performing normalization processing; Using K-nearest neighbor algorithm to compare the normalized relevant sample features with the features in the preset sample library to obtain the corresponding body measurement action type.

[0013] In the second aspect, the present application provides a working method of physical fitness testing equipment, including the following steps: Obtaining video data, performing standardization and normalization processing on the format of the video data, and performing normalization processing on the video frames; Obtaining the normalized video frames, and positioning the key points in the video frames; Obtaining key point information, extracting sample features of the key point information, performing normalization processing on the sample features, comparing the normalized sample features with the features in the preset sample library, and identifying the body measurement action type.

[0014] The further improvement of the present application is that the specific method of obtaining video data, performing standardization and normalization processing on the format of the video data, and performing preprocessing on the video frames is as follows: Real-time collecting or importing local body measurement video through a camera as the video data; Performing format standardization and size normalization on the video data, and adjusting the video resolution to 640x480; Performing frame rate control, wing image denoising and image enhancement on the video frames.

[0015] The further improvement of the present application is that the specific method of obtaining the normalized video frames and positioning the key points in the video frames is as follows: Using a depth model to process the video frames, positioning the preset key points of the human body, and outputting the three-dimensional coordinates of the key points.

[0016] The further improvement of the present application is that the specific method of obtaining key point information, extracting sample features of the key point information, performing normalization processing on the sample features, comparing the normalized sample features with the features in the preset sample library, and identifying the body measurement action type is as follows: Obtaining key point information, analyzing the key point information to obtain time sequence coordinate data of each key point; Extracting relevant sample features in the time sequence coordinate data of each key point and performing normalization processing; The K nearest neighbor algorithm is used to compare the normalized sample features with the features in the preset sample library, so as to obtain the corresponding body measurement action type.

[0017] The further improvement of the present application is that the corresponding body measurement action type is compared with the preset standard action, and whether the body measurement action is standard is judged according to the comparison result, if it is standard, the total number is counted, otherwise, it is not recorded.

[0018] Compared with the prior art, the present application has the following beneficial effects: The present application realizes full-automatic recognition and standardized counting of body measurement actions, supports multiple body measurement items, does not need manual intervention, greatly reduces the possibility of manual participation and error, effectively solves the problems of poor repeatability of results, non-uniform standards, easy to miss and misrecord, and the like compared with the traditional manual counting which depends on the personal experience and subjective judgment of supervisors; the present application can more accurately recognize action types and analyze action details through deep learning key point detection and action classification based on the KNN algorithm, accurately determines whether each action meets the standard, improves the accuracy of action standardized determination, solves the problems that the traditional visual method is difficult to accurately recognize action details and standardization; the present application fully considers the changes of different body types, backgrounds and action amplitudes, has better adaptability and stability, can maintain high-precision recognition in complex actual environments, compared with the existing sensor scheme, does not need to wear additional equipment, avoids the problems of wearing cumbersome affecting user experience and being easily affected by wearing position, body difference and the like; the present application realizes automatic acquisition, storage and analysis of body measurement data, all action detection and determination results can be automatically archived, supports historical data comparison and multi-dimensional statistics, solves the problem of lack of systematic data management and real-time feedback function in the existing scheme; the flexible adaptation of multiple body measurement action types and standard customization of the present application facilitate the expansion of new body measurement items and judgment standards according to different needs, solve the problem of limited support for action standardized determination and multi-action type compatibility in the existing deep learning visual scheme. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a system diagram of the present application; Figure 2 is a flowchart of the present application; Figure 3 is a flowchart of the video acquisition and preprocessing module in the present application; Figure 4 is a flowchart of the key point detection module in the present application; Figure 5 is a flowchart of the action recognition and standardized determination module in the present application. DETAILED DESCRIPTION

[0020] For further understanding of the present application, the following will make a detailed description of the present application in combination with the accompanying drawings and specific embodiments. It should be understood that the embodiments are only to explain the present application and not to limit it.

[0021] Referring to Figure 1 A physical fitness testing device comprises: A video acquisition and preprocessing module is configured to acquire video data, standardize and normalize the format of the video data, and preprocess video frames. A key point detection module is configured to acquire the normalized video frames and locate key points in the video frames. An action recognition and standardization determination module is configured to acquire key point information, extract sample features of the key point information, normalize the sample features, compare the normalized sample features with features in a preset sample library, and identify a physical testing action type.

[0022] Referring to Figure 2 A working method of a physical fitness testing device comprises the following steps: S1. Acquire video data, standardize and normalize the format of the video data, and normalize the video frames.

[0023] S2. Acquire the normalized video frames and locate key points in the video frames.

[0024] S3. Acquire key point information, extract sample features of the key point information, normalize the sample features, compare the normalized sample features with features in a preset sample library, and identify a physical testing action type.

[0025] Embodiment 1 In this embodiment, a physical fitness testing device is provided, which comprises a video acquisition and preprocessing module, a key point detection module, an action recognition and standardization determination module, and a visualization module.

[0026] The video acquisition and preprocessing module of the physical fitness testing device is configured to acquire video data, standardize and normalize the format of the video data, and preprocess video frames. The module acquires video data by real-time acquisition or import of local physical testing video through a camera. After acquiring the video data, the module standardizes the format and normalizes the size of the video data, and adjusts the video resolution to 640x480. In addition, the module also performs frame rate control, image denoising, and image enhancement processing on the video frames to ensure the video quality for subsequent processing.

[0027] The key point detection module is configured to obtain the normalized video frame and locate the key points in the video frame. The module uses a deep model to process the video frame, locates the preset key points of the human body, and outputs the three-dimensional coordinates of the key points. The coordinate data of these key points will be used as the basic data for subsequent action recognition.

[0028] The action recognition and standardization determination module is configured to obtain the key point information, extract the sample features of the key point information, normalize the sample features, compare the normalized sample features with the features in the preset sample library, and recognize the body measurement action type. Specifically, after obtaining the key point information, the module analyzes the key point information to obtain the time sequence coordinate data of each key point, extracts the related sample features in the time sequence coordinate data of each key point, and normalizes the sample features. Finally, the K-nearest neighbor algorithm is used to compare the normalized sample features with the features in the preset sample library to obtain the corresponding body measurement action type.

[0029] The visualization module is configured to obtain the video data and the body measurement action type for display. The module synchronously displays the video data and the recognition result, so that the user can intuitively see the execution process of the body measurement action and the recognition result.

[0030] The visualization module runs through the whole process of body measurement action collection, recognition, determination, and counting. After the user selects a body measurement item in the system interface, the user can collect the movement process in real time through the camera or upload a local video. The system first detects the human body key points of the video frame and superimposes the detection result in the form of skeleton lines and key points on the original picture in real time to realize intuitive visualization of the action process. Subsequently, the system normalizes the obtained key point coordinate information to eliminate the influence of external factors such as body shape and shooting angle, and ensures the accuracy of subsequent determination. Based on the determination standard of the standard action, the system automatically calculates the feature parameters such as the joint angle to determine whether the current action conforms to the standard. For the action determined to be standard, the system automatically counts the number of valid actions and displays the number of valid actions in real time on the interface. If the action is not standard, the system skips this counting, and the determination result is also synchronously fed back on the interface, which facilitates the user to adjust the action in a timely manner. All determination results and action data are automatically stored in the system database, which supports the user to query the historical data at any time and output a detailed action detection report.

[0031] In the working process of the physical fitness testing device, the video collection and preprocessing module first obtains the body measurement video and performs standardization processing. Then, the key point detection module locates the human body key points of the processed video frame. Next, the action recognition and standardization determination module performs action recognition and determination based on the key point information. Finally, the visualization module displays the video data and the recognition result. The whole process forms a complete physical fitness testing process, which can accurately recognize and evaluate the body measurement action of the user.

[0032] Through the cooperation of the above modules, the physical fitness testing equipment can automatically complete the recognition and evaluation of physical testing actions, improve the efficiency and accuracy of physical fitness testing, reduce the subjectivity and inconsistency of manual evaluation, and provide objective technical support for physical education and training.

[0033] Embodiment 2: The embodiment provides a physical fitness testing equipment working method, which includes three main steps of video data acquisition and processing, key point positioning, and action recognition and comparison, and the specific implementation process is as follows: Step 1: Video data acquisition and processing; Referring to Figure 3 , first, the video of the user performing the physical testing action is collected in real time through the camera, or the existing physical testing video is imported from the local storage device as the video data. After obtaining the video data, the video data is standardized in format, and videos of different formats are uniformly converted to a standard format for subsequent processing. At the same time, size normalization processing is performed, and the video resolution is uniformly adjusted to 640x480 pixels to ensure consistency in processing. In addition, frame rate control is performed on the video frames, usually controlled at 25-30 frames per second, to ensure the continuity and accuracy of action capture. During processing, image denoising is also performed on the video frames to remove noise caused by insufficient light, camera equipment quality, etc., to improve image quality. Finally, image enhancement processing is performed to adjust brightness, contrast and other parameters to make the human body outline clearer for subsequent key point positioning.

[0034] Step 2: Key point positioning; Referring to Figure 4 , the video frames after normalization processing are obtained, and a deep learning model is used to process the human body in the video frames. The deep model is based on a convolutional neural network architecture, which can accurately identify and locate the preset key points of the human body, such as the head, shoulders, elbows, wrists, hips, knees and ankles, through multiple layers of feature extraction and position regression. The model outputs three-dimensional coordinate data for each key point after processing, including X, Y and Z three-dimensional spatial position information, where X and Y represent the plane coordinates, and Z represents the depth information. These three-dimensional coordinate data provide basic data support for subsequent action recognition.

[0035] Step 3: Action recognition and comparison; Referring to Figure 5First, the key point information output in step two is obtained, and the information is analyzed to obtain time sequence coordinate data of each key point over time. These time sequence coordinate data record the movement trajectory of each key point in the action process. Next, relevant sample features are extracted from these time sequence coordinate data, including relative position relationship between key points, angle change, speed change, acceleration change and other dynamic features. The extracted sample features are normalized to eliminate differences caused by different body types, action amplitudes and other factors, so that the feature data is comparable. Then, the K-nearest neighbor algorithm is used to compare the normalized relevant sample features with the features in the preset sample library. The preset sample library stores feature templates of various standard physical measurement actions, such as sit-ups, pull-ups, standing long jump, etc. By calculating the distance between feature vectors, the K samples most similar to the current action features are found, and according to the majority voting principle, the most likely physical measurement action type of the current action is determined.

[0036] In a preferred embodiment, the acquisition of video data can be simultaneously collected by multiple angle cameras to obtain more comprehensive human action information and improve the accuracy of key point positioning.

[0037] In another preferred embodiment, the depth model can use an improved pose estimation network that adds a temporal consistency constraint to the standard human pose estimation model, which can better handle the key point tracking problem in continuous actions.

[0038] In yet another preferred embodiment, the K value in the K-nearest neighbor algorithm can be dynamically adjusted according to the size of the sample library, usually selected between 3-7 to balance the accuracy and robustness of the algorithm. At the same time, in the feature comparison process, a weight mechanism can be introduced to assign different importance weights to the features of different key points to improve the accuracy of recognition.

[0039] Through the above steps, the physical test equipment working method can accurately identify the type of physical measurement action performed by the user, providing objective and accurate evaluation basis for physical test, reducing the subjectivity and inconsistency of manual evaluation, and improving the scientificity and fairness of physical test.

[0040] Embodiment 3: This embodiment provides an action recognition and comparison method and an action standardization determination method, and the specific implementation process is as follows: The action recognition and comparison method includes the following steps: S1, Obtain key point information. In this step, key point information in the process of human action is obtained through acquisition equipment. These key points usually include head, shoulder, elbow, wrist, hip, knee and ankle, etc. The acquisition equipment can be a camera, a depth sensor or a wearable device, and the acquisition frequency is 30-60 frames per second to ensure the continuity and integrity of the action data.

[0041] S2, Analyze the key point information to obtain time sequence coordinate data of each key point. In this step, the system converts the originally acquired key point information into coordinate data (x, y, z) in a three-dimensional coordinate system and arranges them in time sequence to form time sequence coordinate data. Each key point has corresponding spatial position information at each time point, and these data constitute the basic representation form of the action. Abnormal data points caused by occlusion or sensor errors are filtered out in the analysis process to ensure data quality.

[0042] S3, Extract relevant sample features in the time sequence coordinate data of each key point and perform normalization processing. In this step, sample features that can represent the characteristics of the action are extracted from the time sequence coordinate data, including but not limited to: relative distance between key points, angle change, speed, acceleration, joint angle change rate, etc. These features can effectively describe the core characteristics of human action. After extraction, these features are normalized to map feature values of different scales to a unified numerical range (usually between 0-1 or -1 to 1), eliminating the influence of individual differences such as body size and action amplitude, so that the features have better comparability. Normalization processing is realized by using methods such as maximum and minimum value normalization or Z-score standardization.

[0043] S4, Compare the normalized relevant sample features with the features in the preset sample library using K-nearest neighbor algorithm to obtain the corresponding body measurement action type. In this step, the K-nearest neighbor (KNN) algorithm is used to compare the normalized sample features with the pre-established action sample library. First, calculate the Euclidean distance or cosine similarity between the to-be-identified action features and the sample features in the sample library, then select the K samples (K value is usually set to 3, 5 or 7) with the closest distance, and determine the type of the current action according to the action type with the highest frequency in the K samples. The preset sample library contains feature data of various standard body measurement actions, such as deep squat, push-up, sit-up, etc. Each action type contains at least 100 standard action samples performed by different people to ensure the accuracy and generalization ability of the recognition.

[0044] The action specification determination method includes the following steps: S1, Obtain the corresponding body measurement action type. After completing the action recognition, the system obtains the type of the current action, such as deep squat, push-up, etc.

[0045] S2, compare with the preset standard action. The system compares and analyzes the recognized action with the preset standard action template. The comparison includes: the degree of coincidence of the key point trajectory, whether the action amplitude meets the standard, whether the action speed is appropriate, whether the key posture point is in place, etc. For different types of physical measurement actions, the system sets different standard judgment standards. For example, for the squat action, the system detects whether the knee is bent to the level of the thigh parallel to the ground, whether the torso is straight, whether the action is completed, etc. In the comparison process, the system calculates the difference between the current action and the standard action template, and generates a standard score of 0-100.

[0046] S3, determine whether the physical measurement action is standard according to the comparison result. The system determines whether the current action is standard according to the standard score calculated in the previous step, combined with the preset standard threshold (usually set to 75 points). When the standard score is greater than or equal to the threshold, the action is determined to be standard; otherwise, it is determined to be non-standard.

[0047] S4, if standard, count the total number, otherwise, do not record. For the action determined to be standard, the system counts it into the total number of valid actions; for the action determined to be non-standard, the system does not count it, and can prompt the user to adjust the action posture through vision or sound. The system also records the standard score of each action, so that the user can view the detailed action evaluation results and understand the aspects that need to be improved.

[0048] Through the combination of the above action recognition and comparison method and the action standard judgment method, accurate recognition and standard evaluation of physical measurement actions can be realized, the scientificity and accuracy of the physical measurement process can be effectively improved, and accurate motion feedback can be provided for the user.

[0049] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that: the specific embodiments of the present application can be modified or replaced by the same, without departing from the spirit and scope of the present application. Any modification or equivalent replacement, which should be covered within the protection scope of the claims of the present application.

Claims

1. A physical fitness testing device, characterized in that, include: The video acquisition and preprocessing module is used to acquire video data, standardize and normalize the format of the video data, and preprocess the video frames. The key point detection module is used to acquire normalized video frames and locate key points in the video frames. The action recognition and standardization judgment module is used to acquire key point information, extract sample features of key point information, normalize the sample features, compare the normalized sample features with features in the preset sample library, and identify the type of body measurement action.

2. The physical fitness testing device according to claim 1, characterized in that, It includes a visualization module for acquiring video data and displaying the types of physical test movements.

3. The physical fitness testing device according to claim 1, characterized in that, The video acquisition and preprocessing module functions through the following methods: Real-time capture of physical test videos via camera or import of local video data; The video data was standardized in format and normalized in size, and the video resolution was adjusted to 640×480; Perform frame rate control, denoising, and image enhancement on the wing image of the video frame.

4. The physical fitness testing device according to claim 1, characterized in that, The key point detection module's functionality is implemented using the following methods: The video frames are processed using a depth model to locate the preset key points of the human body and output the three-dimensional coordinates of the key points.

5. A physical fitness testing device according to claim 1, characterized in that, The functionality of the action recognition and standardization judgment module is implemented through the following methods: Obtain key point information, parse the key point information, and obtain the time-series coordinate data of each key point; Extract relevant sample features from the time-series coordinate data of each key point and perform normalization processing; The K-nearest neighbor algorithm is used to compare the normalized features of relevant samples with the features in the preset sample library to obtain the corresponding physical test action type.

6. A method for operating a physical fitness testing device, characterized in that, Includes the following steps: Acquire video data, standardize and normalize the video data format, and normalize the video frames; Obtain the normalized video frames and locate the key points in the video frames; The system acquires key point information, extracts sample features from the key point information, normalizes the sample features, compares the normalized sample features with features in a preset sample library, and identifies the body measurement action type.

7. The working method of a physical fitness testing device according to claim 6, characterized in that, The specific methods for preprocessing video frames are as follows: Acquire video data, standardize and normalize the video data format, and perform preprocessing on the video frames: Real-time capture of physical test videos via camera or import of local video data; The video data was standardized in format and normalized in size, and the video resolution was adjusted to 640×480; Perform frame rate control, denoising, and image enhancement on the wing image of the video frame.

8. The working method of a physical fitness testing device according to claim 6, characterized in that, The specific method for locating key points in normalized video frames is as follows: The video frames are processed using a depth model to locate the preset key points of the human body and output the three-dimensional coordinates of the key points.

9. The working method of a physical fitness testing device according to claim 6, characterized in that, The specific method for obtaining key point information, extracting sample features from the key point information, normalizing the sample features, and comparing the normalized sample features with features in a preset sample library to identify the body measurement action type is as follows: Obtain key point information, parse the key point information, and obtain the time-series coordinate data of each key point; Extract relevant sample features from the time-series coordinate data of each key point and perform normalization processing; The K-nearest neighbor algorithm is used to compare the normalized features of relevant samples with the features in the preset sample library to obtain the corresponding physical test action type.

10. The method of operating a physical fitness testing device according to claim 9, characterized in that, Obtain the corresponding physical test movement type, compare it with the preset standard movement, and determine whether the physical test movement is standard based on the comparison result. If it is standard, it is included in the total number of times; otherwise, it is not recorded.

Images