Multi-person deep squat real-time counting method based on human key point metric

Abstract

The application discloses a multi-person deep squat real-time counting method based on human key point measurement, which can improve detection accuracy and robustness, support multi-person simultaneous detection, and be suitable for multiple scenes. The method comprises the following steps: based on the multi-target real-time detection and tracking technology of the CSTrack algorithm, multiple persons to be measured can be simultaneously recognized and tracked, and the motion trajectory and position of each person can be recorded and analyzed; the postures of the multiple persons to be measured are estimated based on the BlazePose algorithm, the human posture data of each person to be measured is detected and tracked in real time, the distance and angle change between the joint points are measured to judge the start and end of the action; the posture of the human body is analyzed and optimized in real time by combining the similarity discrimination of KNN and the deep feature learning of CNN, whether the person to be measured is performing a deep squat action and the deep squat category are determined, and then deep squat counting and action standard scoring are performed.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of image data processing, in particular to a multi-person deep squat real-time counting method based on human key point measurement. BACKGROUND

[0002] Deep squat is a classic bodybuilding movement, and real-time counting of deep squat movement is very important for physical exercise. When deep squat detection and counting are performed, manual counting is prone to misjudgment and omission, and it is difficult to meet the demand of multi-person deep squat real-time counting. With the progress of artificial intelligence technology, the intelligent level of schools, sports facilities, fitness centers and various sports facilities is continuously improved. However, the existing technology mostly focuses on single-person deep squat counting, and there are problems such as time-consuming counting, low efficiency, and non-objective scoring when a large number of persons are measured. In addition, in the multi-person deep squat scene, there is a mutual occlusion problem, which may cause the action of part of the measured person to be unable to be accurately identified, affecting the accuracy of counting. In view of the fact that various variants of deep squat (such as wide squat, narrow squat, hand deep squat, kneeling deep squat, wall deep squat, kettlebell deep squat, uneven ground deep squat, split squat, deep squat jump, etc.) may cause the identification system to have difficulty in accurate counting, the present application proposes a multi-person deep squat real-time counting method based on human key point measurement. The method designed in the present patent is helpful to promote the intelligent process of sports facilities, and can effectively save manpower and material resources during physical measurement. SUMMARY

[0003] In order to overcome the above technical defects, the embodiment of the present application provides a multi-person deep squat real-time counting method based on human key point measurement, and the specific scheme is as follows:

[0004] S1, measured person detection based on MTCNN algorithm:

[0005] After the video is obtained, the MTCNN algorithm is used for face recognition of the measured person to determine whether the identity information of the measured person changes. If the identity information changes, the deep squat counting of the measured person is cleared, the deep squat counting of the measured person is ended, and the data is stored in the background. If there is no change, the deep squat real-time counting is continued;

[0006] S2, real-time tracking of the measured person based on CSTrack algorithm:

[0007] After the face recognition verification confirms that the identity information of the measured person does not change, the CSTrack algorithm is used to identify and track the motion trajectory and position of the measured person, and a unique identity tag, i.e. ID, is assigned to the measured person;

[0008] S3, human pose estimation based on BlazePose algorithm:

[0009] Based on the BlazePose algorithm, each human body detection frame in the environment is detected first, and then the posture of each human body is independently detected. The human body posture data is obtained by describing the human body posture through the three-dimensional coordinates of 33 joint points in each frame.

[0010] S4, deep squat start and end time judgment based on human key point measurement:

[0011] The bending angle of the knee of the measured person is calculated from the human body posture data. When the bending angle of the knee is between 35° and 65°, it is determined that the action starts and the deep squat action detection starts. When the bending angle of the knee is between 95° and 105°, it is determined that the action ends and the deep squat action detection ends this time;

[0012] S5, deep squat category discrimination based on KNN algorithm model:

[0013] The KNN algorithm is used to calculate the maximum distance and average distance between the left and right hips (23, 24) and the left and right knees (25, 26), and the left and right hips (23, 24) and the left and right ankles (27, 28) to identify the deep squat category to which the current posture of the measured person belongs.

[0014] S6, deep squat action standardity counting and scoring by CNN algorithm model:

[0015] The CNN algorithm model is used to compare the current average distance value with the reference average distance to evaluate the similarity between the deep squat action and the standard deep squat action, and obtain the action standardity score.

[0016] Further specifically, the 33 joint points are: nose, left eye, right eye, left eye, right eye, left eye, right eye, left ear, right ear, left mouth, right mouth, left shoulder, right shoulder, left elbow, right elbow, left wrist joint, right wrist joint, left little finger, right little finger, left hand index finger, right hand index finger, left thumb, right thumb, left hip, right hip, left knee, right knee, left ankle, right ankle, left heel, right heel, left foot and right foot.

[0017] Further specifically, the video is captured by multiple cameras from different angles at the same time, and the captured video image is preprocessed before face recognition. The preprocessing includes denoising, illumination correction and deblurring.

[0018] The technical solutions adopted by the embodiments of the present application can achieve the following technical effects:

[0019] High precision and robustness: application of KNN and CNN algorithms for screening, combined with changes in human joint point distance, accurate judgment of action category, reduction of misjudgment, improvement of counting stability and precision, and strong anti-interference ability.

[0020] Support multi-person squat counting: solve the problem of time-consuming and low accuracy of manual counting, suitable for group training environment, can monitor multiple subjects and count multiple types of squat actions in real time, and the output squat action standard score is highly feasible.

[0021] Strong adaptability: fusion of multi-view data, high recognition rate in complex environment, suitable for various scenes, convenient for squat counting and action standard scoring. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings illustrate exemplary embodiments and together with the description, serve to explain the principles of the embodiments. The illustrated embodiments are not limited to the illustrated examples. In all the drawings, like reference numerals refer to like but not necessarily identical elements.

[0023] Figure 1 A flowchart of the multi-person squat real-time counting method based on human key point measurement in the present application;

[0024] Figure 2 A schematic diagram of the coordinate positions of 33 key points used in the present application;

[0025] Figure 3 A schematic diagram of the angle between the knee and the vertical direction in the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0027] It should be noted that the technical solutions of each embodiment of the present application can be combined with each other, but it must be based on the fact that a person of ordinary skill in the art can realize it, when the combination of technical solutions appears contradictory or cannot be realized, it should be considered that the combination of technical solutions does not exist, nor is it within the scope of protection claimed by the present application.

[0028] In the description of the present application, it should be understood that the numerical reference before the step does not indicate the order of execution before and after the step, but is only used to facilitate the description of the present application and to distinguish each step, therefore it should not be understood as a limitation of the present application.

[0029] First, the terms related to the present application are explained:

[0030] MTCNN algorithm: English full name is Multi-Task Cascaded Convolutional Networks, which is a deep learning algorithm for face detection and face alignment. It is widely used in face detection, facial expression recognition, facial expression change analysis and other fields. The core idea is to simultaneously perform face detection and key point positioning (eye, nose, mouth and other key points) tasks, and gradually refine the detection results through cascading, so as to realize efficient multi-task learning.

[0031] CSTrack algorithm: a single-stage multi-target tracking algorithm based on ReID (Re-Identification), the main goal of this algorithm is to track multiple targets in real time in video surveillance, autonomous driving and other applications, and through ReID technology to re-identify the same target between different video frames, solve the problem of target occlusion, dispersion, cross-camera in traditional multi-target tracking method.

[0032] BlazePose algorithm: a real-time human pose tracking algorithm proposed by Google, aiming to accurately and quickly identify human key points, especially for real-time human pose estimation on mobile devices, providing an efficient solution to estimate the key point position of human body (such as head, arm, leg, etc.) and provide 2D or 3D position information of each part of the body.

[0033] KNN algorithm: full name is K-Nearest Neighbors, a supervised machine learning algorithm, mainly used for classification and regression tasks, the core idea is to give a to-be-predicted sample, find the K samples in the training data set that are most similar (i.e. closest) to the sample, and then predict the class (classification problem) or value (regression problem) of the to-be-predicted sample through the class (or value) of these K neighbors.

[0034] CNN algorithm: a deep learning model suitable for image processing, which extracts image features through convolution operation and reduces data volume through pooling operation, and finally classifies or recognizes through fully connected layer.

[0035] Secondly, in order to facilitate the understanding of the technical solutions provided by the person skilled in the art, the related technology is described as follows:

[0036] Referring to Figure 1 The multi-person deep squat real-time counting method based on human key point measurement in the embodiment of the present application includes three parts of recognition tracking, pose estimation and counting score, which are as follows:

[0037] I. Recognition tracking

[0038] S1, subject detection based on MTCNN algorithm

[0039] After the video is acquired, the MTCNN algorithm is used for face recognition of the subject to determine whether the identity information of the subject has changed. If the identity information has changed, the subject's deep squat count is cleared, the deep squat count of the subject is ended, and the data is stored in the background. If there is no change, the deep squat real-time count is continued.

[0040] Further, the video is captured from different angles by multiple cameras, and further, the captured video image can be pre-processed before face recognition. The pre-processing can include denoising, illumination correction, deblurring, etc.

[0041] S2, subject real-time tracking based on CSTrack algorithm

[0042] After the face recognition verification confirms that the identity information of the subject has not changed, the CSTrack algorithm is used to identify and track the motion trajectory and position of the subject, and a unique identity tag, i.e., ID, is assigned to the subject. The uniqueness of the identity tag can achieve continuous tracking of the subject during the entire deep squat process. The subsequent deep squat count data is stored in the background based on the ID.

[0043] II. Pose estimation

[0044] S3, human pose estimation based on BlazePose algorithm

[0045] Based on the BlazePose algorithm, each human detection box in the environment is detected from top to bottom first, and then the pose of each human body is independently detected. The human pose is described by the three-dimensional coordinates of 33 joints in each frame, and the human pose data is obtained.

[0046] Specifically, the 33 joints are: nose, left eye, right eye, left eye, right eye, left eye, right eye, left ear, right ear, left mouth, right mouth, left shoulder, right shoulder, left elbow, right elbow, left wrist joint, right wrist joint, left little finger, right little finger, left hand index finger, right hand index finger, left thumb, right thumb, left hip, right hip, left knee, right knee, left ankle, right ankle, left heel, right heel, left foot and right foot. The specific positions are as shown in Figure 2 positions represented by numbers 0-32 in the middle.

[0047] Squat counting requires extremely high real-time performance. Traditional human feature extraction based on RGB features is greatly affected by the environment and computing equipment, and its real-time performance is poor, which poses a challenge to squat recognition and counting systems. In contrast, the human posture feature extraction system based on the BlazePose algorithm can effectively capture the positions of 33 key joints of multiple human bodies in space during squats. It independently detects the posture of each human body and records relevant data by using the three-dimensional coordinates of the 33 key joint positions in each frame.

[0048] S4. Determining the start and end times of squats based on human body key point measurements.

[0049] The knee flexion angle of the subject is calculated from the human posture data. When the knee flexion angle is between 35° and 65°, the movement is determined to start and the squat movement detection begins. When the knee flexion angle is between 95° and 105°, the movement is determined to end and the squat movement detection ends.

[0050] The range of 35° to 65° refers to a value greater than or equal to 35° and less than or equal to 65°. Similarly, the range of 95° to 105° refers to a value greater than or equal to 95° and less than or equal to 105°.

[0051] More specifically, such as Figure 3 As shown, the angle between the knee and the distances between the knee and hip, and between the knee and ankle, is selected as the knee flexion angle. The specific calculation of the angle value is as follows:

[0052] S4.1 Calculate the vectors from the knee to the hip and from the knee to the ankle using formula (1).

[0053] Among them, (x k ,y k ,z k ), (x h ,y h ,z h ( ) are the coordinates of the knee and hip, respectively.

[0054] S4.2 Calculate the vector using formula (2) Vectors in the perpendicular direction The angle cos(θ) between the two sides gives the knee flexion angle:

[0055]

[0056] in, Consider the dot product of vectors and the magnitude of the vectors, respectively. It should be understood that since the vector perpendicular to the z-axis is in the z-direction, the vector perpendicular to the z-axis...

[0057] III. Counting and Scoring

[0058] Before combining KNN and CNN algorithms to analyze and optimize human posture in real time for squat counting and scoring, a multi-view fusion training set of squat videos is pre-collected as sample data for training the KNN and CNN models. The sample data includes human posture features extracted from 19 different squat videos of various perspectives and types using the BlazePose algorithm, converted into objects containing joint coordinates and squat posture samples. These posture features include the position and movement trajectory of each joint, comprehensively capturing the details of the squat movement. Then, this feature data is used as a sample set to train the KNN and CNN models, learning and recognizing standard squat postures, providing a standard for subsequent steps of movement recognition, counting, and similarity evaluation.

[0059] S5. Squat category discrimination based on KNN algorithm model

[0060] The KNN algorithm was used to identify the squat category of the subject's current posture by calculating the maximum distance and average distance between the left and right hips (23, 24) and left and right knees (25, 26), as well as between the left and right hips (23, 24) and left and right ankles (27, 28).

[0061] Specifically, identification involves determining whether the squat sample matches the KNN classification label, identifying the squat sample most similar to the current posture, and performing a similarity assessment to determine the squat category it belongs to.

[0062] The above-mentioned identification of the squat category to which the subject's current posture belongs based on maximum distance filtering and average distance filtering specifically includes:

[0063] First, for each pair of joints (ankle and hip, hip and knee), calculate the Euclidean distance between them, as shown in formula (3):

[0064]

[0065] Where k represents the k-th person, It is the distance between the key points i and j of the k-th person. and These are the coordinates of joint i and joint j of the k-th person, respectively.

[0066] The distances between the joint pairs that need to be calculated in this application include: the distances between the left hip (23) and the left knee (25), the right hip (24) and the right knee (26), the left hip (23) and the left ankle (27), and the right hip (24) and the right ankle (28).

[0067] Secondly, the distance between the joint pairs calculated using formula (4) is normalized to obtain the normalized distance:

[0068]

[0069] in, It is the normalized distance. and These are the minimum and maximum values ​​of the distance between joints measured by the k-th person.

[0070] Next, the set of all normalized distances of the test subjects is filtered by maximum distance, and samples that are highly similar to the target squat sample but have a small difference in feature vector are removed, resulting in the filtered set of maximum distances.

[0071] Finally, for the set of maximum distances after filtering, calculate its average distance value.

[0072] Specifically, let D (k) For the set of all normalized distances for the k-th person, a threshold T is set. max Filter out the set D with the maximum distance max (k) The specific formula is shown in formula (5):

[0073]

[0074] For the set of maximum distances after filtering, calculate its average distance value using formula (6):

[0075]

[0076] Finally, the squat category of the subject's current posture is determined based on the calculated average distance value.

[0077] S6. Perform standard counting and scoring of squat movements using a CNN algorithm model.

[0078] The CNN algorithm model is used to compare the current average distance value with the reference average distance to evaluate the similarity between the squatting movement and the standard squatting movement, and obtain a similarity score (i.e., the standardization score of the movement).

[0079] Specifically, the formula for calculating similarity (7) is as follows:

[0080]

[0081] Among them, S (k) For the standardization score of the k-th person's action, μ ref The average distance of the squatting motion obtained from the training of the sample is .

[0082] The KNN algorithm is used to identify the squat category most similar to the current posture. Then, a CNN algorithm model is used to evaluate the similarity between the squatting motion features trained on samples of that type, thereby obtaining a motion standardization score to help the test subject correct any non-standard movements in a timely manner. In summary, by automating and standardizing the measurement process, the algorithm reduces human error, significantly improves the accuracy and reliability of multi-person squat measurements, and helps test subjects optimize their movements promptly.

[0083] In squat counting, the start and end of the movement are accurately determined by analyzing the distance between key points on the human body, and the standardization of the movement is evaluated. Squat types include various variations (such as wide squats, narrow squats, etc.), but the common characteristic of squat exercises is that the hips and knees must be parallel to or below the knees when squatting down, and the distance between the hip and ankle joints is small at this point. This patented design for acquiring the movement initiation signal first checks whether the coordinates of the human joint points are within the video frame, and then calculates the angle between the human knee and the vertical direction (e.g., ...). Figure 3 The angle between the vector formed by the knee and hip and the z-axis is used to determine the start and end of the movement. If the movement begins and then the calculations are performed at the hip and ankle joints (…), the movement is considered complete. Figure 2 Key points 23, 24, 27, and 28) and the hip and knee joints of the human body ( Figure 2 The similarity assessment and standardized scoring are performed based on the distance between key points 23, 24, 25, and 26. Furthermore, if the angle between the knee and the vertical direction is between 35° and 65°, the signal is "start"; if the angle is between 95° and 105°, the signal is "end". This dual recognition significantly improves the accuracy of squat recognition.

[0084] The squat counting module based on edge devices aims to achieve accurate motion recognition and counting by analyzing users' squatting movements in real time. The squat counting system built using the above steps can be directly deployed to various edge devices. By monitoring the movement process in real time through a camera, as mentioned above, squat counting can be achieved by detecting angle changes at key moments. A squat is counted when the angle between the knee and the vertical direction gradually increases from less than 65° to more than 95°, and the corresponding score is output in real time. After the exercise is completed, the squat count result and motion standardization score for each ID are output. In this way, this application can be applied to scenarios such as gyms, training grounds, and school physical tests to achieve real-time counting of multiple squats.

[0085] The technical solution of this application can bring about the following technical effects:

[0086] First, improve the accuracy and robustness of squat recognition.

[0087] By applying maximum and average distance filtering using KNN and CNN algorithms, and combining the distance changes between human joints during movement, a squat start-end moment recognition, counting, and motion standardization scoring system was constructed, achieving high-precision motion start and end judgment and motion standardization scoring. Specifically, during the squat, the method proposed in this patent accurately determines the start and end of each movement by analyzing the angle change between the knee and the vertical direction, and analyzes the distance between key points such as the hip and ankle, and hip and knee for similarity assessment, thereby ensuring the accuracy of counting and motion standardization scoring. Maximum distance filtering helps to exclude samples that are very similar to the target sample but have minor differences in feature vectors, while average distance filtering ensures that the overall posture is consistent with the standard sample. This method effectively reduces misjudgments caused by prediction jitter and improves the stability of counting. The system combines human posture features with data from different perspectives obtained through multiple cameras to reduce occlusion problems and provide more accurate skeletal key point localization, thereby improving the accuracy of squat motion recognition. It has stronger anti-interference capabilities against interference factors such as changes in lighting and complex backgrounds, thus maintaining high accuracy and stable performance in various environments. Furthermore, CSTrack's multi-target real-time detection and tracking technology provides comprehensive support for training and competition in sports scenarios, ensuring that the movement trajectory of each participant is accurately recorded, while also providing data for skill improvement. Through this efficient monitoring and analysis, sports teams can gain a greater competitive advantage and drive continuous improvement in athletic performance.

[0088] Secondly, it can simultaneously perform real-time detection and tracking of multiple test subjects, thus enabling multi-person support functionality.

[0089] In physical education assessments, counting is typically done manually by instructors, which is prone to errors and omissions. With a large number of participants, this can lead to low efficiency and time-consuming processes. The system proposed in this patent can simultaneously identify, count, and score the squatting movements of multiple participants, improving counting efficiency in group training scenarios. Through advanced deep learning algorithms, it can monitor the posture and number of squats for each participant in real time, ensuring correct training for everyone. The system's real-time counting and scoring feedback allows for timely correction of posture or adjustment of training intensity. Furthermore, the system can acquire overall data to analyze each participant's performance and progress, enabling the design of more efficient training plans. The multi-person support function also enhances the interactivity and competitiveness of training.

[0090] Third, it is applicable to a variety of scenarios and has strong adaptability.

[0091] By integrating data from different perspectives, the system can maintain a high recognition rate in various complex environments and adapt to different training scenarios. This design not only improves the naturalness and convenience of the user experience but also makes the system suitable for more usage scenarios, allowing for easy counting and evaluation of squat movements at home, in the gym, or outdoors.

[0092] Obviously, those skilled in the art should understand that the modules or steps of the embodiments of this application described above can be implemented using general-purpose computer devices. They can be centralized on a single computer device or distributed across a network of multiple computer devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computer device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of this application are not limited to any particular combination of hardware and software.

[0093] It should be noted that the above are merely preferred embodiments of this application and do not limit the scope of patent protection of this application. Any equivalent structural or procedural changes made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A real-time counting method for multiple-person squats based on human body key point measurement, characterized in that, include: S1. Subject detection based on MTCNN algorithm: After acquiring the video, the MTCNN algorithm is used to perform facial recognition on the subject to determine whether the subject's identity information has changed. If it has changed, the subject's squat count is reset to zero, the squat count for that subject ends, and the data is stored in the background. If it has not changed, the squat count continues in real time. S2. Real-time tracking of test subjects based on the CSTrack algorithm: After completing the facial recognition verification to confirm that the subject's identity information has not changed, the CSTrack algorithm is used to identify and track the subject's movement trajectory and location, and a unique identity label, i.e., ID, is assigned to the subject. S3. Human pose estimation based on the BlazePose algorithm: Based on the BlazePose algorithm, a top-down approach is used to first detect each human body bounding box in the environment, and then to independently detect the pose of each human body. The human body pose is described by the three-dimensional coordinates of 33 joints in each frame, and human body pose data is obtained. S4. Determining the start and end times of a squat based on measurements of key human body points: The knee flexion angle of the subject is calculated from the human posture data. When the knee flexion angle is between 35° and 65°, the movement is determined to start and the squat movement detection begins. When the knee flexion angle is between 95° and 105°, the movement is determined to end and the squat movement detection ends. S5. Squat category discrimination based on KNN algorithm model: The KNN algorithm was used to identify the squat category of the subject's current posture by calculating the maximum distance and average distance between the left and right hips (23, 24) and left and right knees (25, 26), as well as between the left and right hips (23, 24) and left and right ankles (27, 28). S6. Perform standard counting and scoring of squat movements using a CNN algorithm model: The CNN algorithm model is used to compare the current average distance value with the reference average distance to evaluate the similarity between the squatting movement and the standard squatting movement, and obtain the movement standardization score.

2. The method for real-time counting of multiple squats according to claim 1, characterized in that, In S4, the angle between the knee and the distances between the knee and hip, and between the knee and ankle, is selected as the knee flexion angle. The specific calculation of the angle value is as follows: S4.1 Calculate the vectors from the knee to the hip and from the knee to the ankle using the following formula (1). S4.2 Calculate the vector using the following formula (2). Vectors in the perpendicular direction The angle cos(θ) between the two sides gives the knee flexion angle:

3. The method for real-time counting of multiple squats according to claim 1, characterized in that, S5 identifies the squat category of the subject's current posture based on maximum distance filtering and average distance filtering, specifically including: First, calculate the Euclidean distance between each pair of joints, as shown in formula (3): Each pair of joints includes: left hip (23) and left knee (25), right hip (24) and right knee (26), left hip (23) and left ankle (27), right hip (24) and right ankle (28); Secondly, the distance between the joint pairs calculated using formula (4) is normalized to obtain the normalized distance: Next, the set of all normalized distances of the test subjects is filtered by maximum distance, and samples that are highly similar to the target squat sample but have a small difference in feature vector are removed, resulting in the filtered set of maximum distances; for the filtered set of maximum distances, the average distance value is calculated. Let D (k) For the set of all normalized distances for the k-th person, a threshold T is set. max Filter out the set D with the maximum distance max (k) The specific formula is shown in formula (5): D max (k) ={d i ' j (k) ∈D (k) |d i ' j (k) ≤T max } (5) For the set of maximum distances after filtering, calculate its average distance value using formula (6): Finally, the squat category of the subject's current posture is determined based on the calculated average distance value.

4. The method for real-time counting of multiple squats according to claim 3, characterized in that, The formula for calculating the standardization score of movement in S6 is shown in formula (7):

5. The method for real-time counting of multiple squats according to claim 1, characterized in that, The 33 joints are: nose, left inner eye, right inner eye, left eye, right eye, left outer eye, right outer eye, left ear, right ear, left mouth, right mouth, left shoulder, right shoulder, left elbow, right elbow, left wrist joint, right wrist joint, left little finger, right little finger, left index finger, right index finger, left thumb, right thumb, left hip, right hip, left knee, right knee, left ankle, right ankle, left heel, right heel, left toes, and right toes.

6. The method for real-time counting of multiple squats according to claim 1, characterized in that, The video is captured simultaneously from different angles using multiple cameras. The captured video images are preprocessed before facial recognition is performed. The preprocessing includes noise reduction, illumination correction, and deblurring.

Images