Taijiquan action recognition method and device
By identifying Tai Chi moves and extracting the rotation angle change curves of the shoulders and waist, performing differential processing and comparison, the problem that the existing technology cannot accurately identify the Tai Chi force mode, and the accurate identification and auxiliary exercises of the Tai Chi force mode are achieved.
Patent Information
- Application Number
- CN202411971934.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-25
AI Technical Summary
The existing Tai Chi action recognition methods cannot accurately identify the force mode during the Tai Chi posture switching drill, and cannot assist practitioners in mastering the force mode of the silk-wrapped force.
By obtaining the key point data of the target human body in the pending video sequence, input the pre-trained Tai Chi move recognition model, identifying Tai Chi moves and obtaining its key point data. Then, based on the identified moves, the key point data of the force-acting action is selected, the rotation angle change curve of the shoulder and waist is extracted, differential processing is performed and superimposed, and finally compared with the pre-established force-acting feature library to identify the force-acting action mode.
The accurate identification of Tai Chi force action mode can help practitioners better master the force action mode of silk-wrapped force and improve the practice effect of Tai Chi.
Smart Images
Figure CN119888852A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of motion recognition, and in particular to a Tai Chi motion recognition method and device. Background Art
[0002] Chansi Jin is a special Jin method in Tai Chi, which emphasizes that the limbs present a spiral winding state during movement. This Jin method requires not only coherent and smooth movements, but also the coordination and unity of internal strength. Through the practice of Chansi Jin, you can gradually master the essence of Tai Chi, which can not only improve the fighting level of Tai Chi, but also promote physical and mental health, and play a strategic role in practical applications. Therefore, in the practice of Tai Chi, you should pay attention to the practice and mastery of Chansi Jin's force-generating movements.
[0003] In the power-generating action pattern of Chansi Jin, there are two typical trunk movements, one is to drive the whole body movement only through the twisting of the spine, and the other is to drive the whole body movement by driving the waist rotation through the twisting of the spine. However, most practitioners cannot distinguish well between controlling the waist rotation and the spinal twisting. Practitioners are often more likely to control the waist rotation alone during practice, which causes practitioners to often mistakenly control the spinal twisting and waist rotation; for example, practitioners often mistakenly control the waist rotation to drive the spinal rotation to achieve body movements driven only by the spinal twisting, or to achieve body movements driven by the spinal twisting and then the waist rotation.
[0004] At present, the recognition method of Tai Chi movements generally includes the following steps: 1. Obtain Tai Chi practice video; 2. Extract frames from the video to obtain an image sequence, and pre-process the image sequence; 3. Use a human key point detection algorithm to identify the various joints of the human body in the image sequence; 4. Extract the relevant features of the key parts of the human body based on the positions of the identified joints of the human body. The extracted features generally include the position, angle, movement speed and joint movement trajectory of the joints; 5. According to the relevant features of the key parts of the human body, identify Tai Chi movements. The above recognition method can accurately identify the various postures of Tai Chi, but it cannot accurately identify the force mode during the Tai Chi posture switching exercise. Therefore, the existing Tai Chi movement recognition method is only to assist practitioners in learning the external form of Tai Chi, and cannot assist practitioners in mastering the force action mode of Chansi Jin.
[0005] For this reason, an improved Tai Chi movement recognition method and device are urgently needed. Summary of the invention
[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a Tai Chi action recognition method and device, which can accurately recognize the force pattern during Tai Chi practice.
[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:
[0008] In a first aspect, the present invention provides a Tai Chi action recognition method, comprising the following steps:
[0009] S1, obtaining key point data of the target human body in the video sequence to be processed;
[0010] S2, inputting the key point data into a pre-trained Tai Chi moves recognition model, identifying Tai Chi moves, and obtaining the key point data corresponding to the moves;
[0011] S3, according to the identified Tai Chi moves, using a preset key point data selection rule corresponding to the moves, selecting key point data of the force-generating action from the key point data corresponding to the moves;
[0012] S4. Extract shoulder key point data and waist key point data according to the key point data of the force-generating action, and draw a rotation angle change curve of each shoulder key point over time and a rotation angle change curve of each waist key point over time;
[0013] S5, performing differential processing on the curve of the rotation angle change over time of each shoulder key point to obtain the differential curve of each shoulder key point, adding and averaging the differential curves of all shoulder key points to obtain the shoulder differential curve; performing differential processing on the curve of the rotation angle change over time of each waist key point to obtain the differential curve of each waist key point, adding and averaging the differential curves of all waist key points to obtain the waist differential curve;
[0014] S6. Superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than a first preset threshold, compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with a pre-established force action feature library, and identify the force action pattern.
[0015] Optionally, key point data of a target human body in a video sequence to be processed is obtained, including: inputting the video sequence to be processed into a pre-trained target human body detection model and outputting a motion sequence of the target human body; inputting the motion sequence of the target human body into a pre-trained Alphapose network model, extracting key points of the target human body in the motion sequence, and obtaining key point data.
[0016] Optionally, the pre-set key point data selection rule of the moves includes: selecting key point data within a preset time period from the key point data corresponding to the moves as the key point data of the force-generating action.
[0017] Optionally, in S4, the extracted shoulder key points and waist key points correspond one to one in the ground normal direction.
[0018] Optionally, in S5, according to a preset differential distance, two rotation angles separated by a differential distance in each rotation angle variation curve are differentiated to obtain the sensing optical fiber in each differential curve.
[0019] Optionally, in S6, the shoulder differential curve and the waist differential curve are superimposed on the same coordinate system, data points in the shoulder differential curve and the waist differential curve that are less than a first preset threshold are deleted, and the transition data segments in the shoulder differential curve and the waist differential curve are deleted, and the shoulder differential curve and the waist differential curve are divided at the transition data segments to obtain multiple shoulder differential curves and multiple waist differential curves, and each shoulder differential curve and each waist differential curve are fitted into a straight line segment, and the position distribution of the multiple shoulder straight line segments and the multiple waist straight line segments in the time domain are compared with a pre-established force action feature library to identify the force action pattern.
[0020] Optionally, the transition data segments in the shoulder differential curve and the waist differential curve are deleted, including: subtracting adjacent values in the shoulder differential curve, and if the difference is greater than a second preset threshold, the two values are judged to be transition data and deleted; subtracting adjacent values in the waist differential curve, and if the difference is greater than a third preset threshold, the two values are judged to be transition data and deleted.
[0021] Optionally, each shoulder difference curve and each waist difference curve are fitted into a straight line segment by using the least square method.
[0022] Optionally, the pre-established force action feature is the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain.
[0023] In a second aspect, the present invention provides a Tai Chi action recognition device, comprising:
[0024] The first key point extraction module is used to obtain key point data of a target human body in a video sequence to be processed;
[0025] A move recognition module is used to input key point data into a pre-trained Tai Chi move recognition model, recognize Tai Chi moves, and obtain key point data corresponding to the moves;
[0026] The second key point extraction module is used to select the key point data of the force-generating action from the key point data corresponding to the identified Tai Chi moves using a preset key point data selection rule corresponding to the moves;
[0027] The first curve drawing module is used to extract shoulder key point data and waist key point data according to the key point data of the force-generating action, and draw a rotation angle change curve of each shoulder key point over time and a rotation angle change curve of each waist key point over time;
[0028] The second curve drawing module is used to perform differential processing on the curve of the rotation angle change over time of each shoulder key point to obtain the differential curve of each shoulder key point, add and average the differential curves of all shoulder key points to obtain the shoulder differential curve; perform differential processing on the curve of the rotation angle change over time of each waist key point to obtain the differential curve of each waist key point, add and average the differential curves of all waist key points to obtain the waist differential curve;
[0029] The force action recognition module is used to superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than the first preset threshold, and compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with the pre-established force action feature library to identify the force action pattern.
[0030] The beneficial effects of the present invention are:
[0031] The Tai Chi action recognition method and device provided by the present invention can accurately recognize the force-generating action pattern of Tai Chi by extracting shoulder rotation features and waist rotation features that are interconnected in the time domain. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 The figure is a flow chart of a Tai Chi movement recognition method according to a specific embodiment of the present invention. DETAILED DESCRIPTION
[0033] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.
[0034] like Figure 1 As shown, the embodiment of the present invention provides a Tai Chi action recognition method, comprising the following steps:
[0035] Step S1, obtaining key point data of a target human body in a video sequence to be processed.
[0036] Among them, the video sequence to be processed is a video sequence of the target human body practicing Tai Chi movements; the key points of the target human body refer to the points representing the characteristic body parts of the human body, and the characteristic body parts of the human body include nose, left eye, right eye, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle, and right ankle.
[0037] Among them, obtaining the key point data of the target human body in the video sequence to be processed includes: inputting the video sequence to be processed into a pre-trained target human body detection model, and outputting the action sequence of the target human body; inputting the action sequence of the target human body into a pre-trained Alphapose network model, extracting the key points of the target human body in the action sequence, and obtaining the key point data.
[0038] The key point data obtained include the position and confidence of each key point, as well as the connection relationship (i.e., the temporal distribution of the same key points) and confidence between video frames.
[0039] Specifically, the video sequence to be processed is input into a pre-trained target human body detection model, and the target human body action sequence is output, including:
[0040] Step A11, preprocessing the video sequence to be processed;
[0041] Step A12: extract video frames from the video sequence to be processed at set time intervals, and input the extracted video frames into the target human detection model frame by frame, identify and select the target human body, and extract the action sequence of the target human body.
[0042] Step S2: input the key point data into a pre-trained Tai Chi moves recognition model, identify the Tai Chi moves, and obtain the key point data corresponding to the moves.
[0043] Specifically, in step S2, the key point data is input into the Tai Chi moves recognition model, and the time domain distribution characteristics of each key point are matched with the time domain distribution characteristics of the key points corresponding to each predefined Tai Chi moves, and the Tai Chi moves sequence distributed in the time domain corresponding to the key point data is identified, and the key point data corresponding to each Tai Chi move is obtained. The force mode (i.e., the target force mode) driven by the spinal twisting generally exists in several specific Tai Chi moves, so it is helpful to select Tai Chi moves related to the target force mode according to the key point data, and then analyze the force mode of the key point data corresponding to the Tai Chi moves, reduce the amount of calculation, and make the analysis accurate.
[0044] The Tai Chi moves involved in this embodiment are different in different schools and routines, such as each of the 18 moves of Chen-style Tai Chi, each of the 42 moves of Yang-style Tai Chi, etc.
[0045] Step S3: based on the identified Tai Chi moves, using the pre-set key point data selection rules corresponding to the moves, select the key point data of the force-generating action from the key point data corresponding to the moves.
[0046] The pre-set key point data selection rules for the moves include: selecting key point data within a preset time period from the key point data corresponding to the moves as key point data for the force-generating action. Generally speaking, during the practice of a Tai Chi move, the force-generating action often occurs within a specific time interval in the move practice time domain. Therefore, the key point data for the force-generating action is selected by using the pre-set key point data selection rules for the corresponding moves, which is convenient to calculate and accurate to select.
[0047] Step S4, extracting shoulder key point data and waist key point data according to the key point data of the force-generating action, and drawing a curve of the rotation angle change of each shoulder key point over time and a curve of the rotation angle change of each waist key point over time.
[0048] The extracted shoulder key points and waist key points correspond one to one in the direction of the ground normal line, so that it is convenient to form an accurate position distribution of the curve in the time domain in the subsequent process, so as to accurately compare it with the pre-established force action feature library.
[0049] Step S5, performing differential processing on the curve of the rotation angle change over time of each shoulder key point to obtain the differential curve of each shoulder key point, adding and averaging the differential curves of all shoulder key points to obtain the shoulder differential curve; performing differential processing on the curve of the rotation angle change over time of each waist key point to obtain the differential curve of each waist key point, adding and averaging the differential curves of all waist key points to obtain the waist differential curve.
[0050] In this way, by performing differential processing on the rotation angle change curve and adding and averaging it, the accuracy of data analysis is improved, the influence of noise is reduced, and the subsequent comparison with the force action feature library is facilitated.
[0051] Specifically, according to the preset differential distance, the two rotation angles separated by the differential distance in each rotation angle change curve are subtracted to obtain the sensing optical fiber in each differential curve. As an example, the preset differential distance is 3, and the numerical queues of the rotation angle change curve are marked as 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.... in sequence. According to the preset differential distance, the two marked points separated by the differential distance in each rotation angle change curve are subtracted, and the operations performed are 4-1, 5-2, 6-3, 7-4, 8-5, 9-6, 10-7...., and the new queue obtained is the differential curve.
[0052] Step S6, superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than the first preset threshold, compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with the pre-established force action feature library, and identify the force action pattern.
[0053] After superimposing the shoulder differential curve and the waist differential curve in the same coordinate system, there are still some noise data. These noise data are smaller than the data points in the shoulder differential curve and the waist differential curve that are truly valuable for feature comparison. Therefore, the noise data is filtered out by the first preset value to avoid affecting the position distribution of the shoulder differential curve and the waist differential curve in the time domain, thereby causing errors when comparing with the pre-established force action feature library and failing to accurately identify the force action pattern.
[0054] Preferably, in step S6, the shoulder differential curve and the waist differential curve are superimposed on the same coordinate system, data points in the shoulder differential curve and the waist differential curve that are less than the first preset threshold are deleted, and the transition data segments in the shoulder differential curve and the waist differential curve are deleted, and the shoulder differential curve and the waist differential curve are divided at the transition data segments to obtain multiple shoulder differential curves and multiple waist differential curves, and each shoulder differential curve and each waist differential curve are fitted into a straight line segment, and the position distribution of the multiple shoulder straight line segments and the multiple waist straight line segments in the time domain is compared with the pre-established force action feature library to identify the force action mode. Each shoulder differential curve and each waist differential curve are a wavy line that fluctuates with a straight line parallel to the horizontal axis of the coordinate system as a reference, so each shoulder differential curve and each waist differential curve can be fitted into a straight line segment parallel to the horizontal axis of the coordinate system. In this way, through the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain, the practitioner's force movement pattern can be described in more detail, and then compared with the pre-established force movement feature library, the force movement pattern can be identified in more detail and accurately; and the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain makes the representation of the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain simpler, which is conducive to further computational analysis and has a small amount of calculation.
[0055] Furthermore, the transition data segments in the shoulder differential curve and the waist differential curve are deleted, including: subtracting adjacent values in the shoulder differential curve, and if the difference is greater than a second preset threshold, the two values are judged to be transition data and deleted; subtracting adjacent values in the waist differential curve, and if the difference is greater than a third preset threshold, the two values are judged to be transition data and deleted.
[0056] Furthermore, the least square method is used to fit each shoulder difference curve and each waist difference curve into a straight line segment.
[0057] The pre-established force action feature is the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain.
[0058] In summary, the Tai Chi action recognition method provided by the embodiment of the present invention can accurately identify the force action pattern of Tai Chi by extracting the shoulder rotation features and waist rotation features that are interconnected in the time domain.
[0059] The embodiment of the present invention further provides a Tai Chi action recognition device, comprising:
[0060] The first key point extraction module is used to obtain key point data of a target human body in a video sequence to be processed.
[0061] The move recognition module is used to input key point data into a pre-trained Tai Chi move recognition model, identify Tai Chi moves, and obtain the key point data corresponding to the moves.
[0062] The second key point extraction module is used to select the key point data of the force-generating action from the key point data corresponding to the identified Tai Chi moves using a preset key point data selection rule corresponding to the moves.
[0063] The first curve drawing module is used to extract shoulder key point data and waist key point data according to the key point data of the force-generating action, and draw a curve of the rotation angle change of each shoulder key point over time and a curve of the rotation angle change of each waist key point over time.
[0064] The second curve drawing module is used to perform differential processing on the curve of the rotation angle change of each shoulder key point over time to obtain the differential curve of each shoulder key point, add and average the differential curves of all shoulder key points to obtain the shoulder differential curve; perform differential processing on the curve of the rotation angle change of each waist key point over time to obtain the differential curve of each waist key point, add and average the differential curves of all waist key points to obtain the waist differential curve.
[0065] The force action recognition module is used to superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than the first preset threshold, and compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with the pre-established force action feature library to identify the force action pattern.
[0066] Other functions implemented by the first key point extraction module, the move recognition module, the second key point extraction module, the first curve drawing module, the second curve drawing module and the force action recognition module are similar to those in the above-mentioned Tai Chi action recognition method and will not be repeated here.
[0067] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0068] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.
[0069] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.
[0070] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0071] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.
[0072] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.
Claims
1. A Tai Chi action recognition method, characterized in that: The following steps are involved: S1, obtaining key point data of the target human body in the video sequence to be processed; S2, inputting the key point data into a pre-trained Tai Chi moves recognition model, identifying Tai Chi moves, and obtaining the key point data corresponding to the moves; S3, according to the identified Tai Chi moves, using a preset key point data selection rule corresponding to the moves, selecting key point data of the force-generating action from the key point data corresponding to the moves; S4. Extract shoulder key point data and waist key point data according to the key point data of the force-generating action, and draw a rotation angle change curve of each shoulder key point over time and a rotation angle change curve of each waist key point over time; S5, performing differential processing on the curve of the rotation angle change over time of each shoulder key point to obtain the differential curve of each shoulder key point, adding and averaging the differential curves of all shoulder key points to obtain the shoulder differential curve; performing differential processing on the curve of the rotation angle change over time of each waist key point to obtain the differential curve of each waist key point, adding and averaging the differential curves of all waist key points to obtain the waist differential curve; S6. Superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than a first preset threshold, compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with a pre-established force action feature library, and identify the force action pattern.
2. The Tai Chi action recognition method according to claim 1, characterized in that: Obtain key point data of the target human body in the video sequence to be processed, including: The video sequence to be processed is input into a pre-trained target human body detection model to output the target human body action sequence; the target human body action sequence is input into a pre-trained Alphapose network model, the key points of the target human body in the action sequence are extracted to obtain the key point data.
3. The Tai Chi action recognition method according to claim 1, characterized in that: The pre-set key point data selection rules for moves include: The key point data within a preset time period is selected from the key point data corresponding to the moves as the key point data of the force-generating action.
4. The Tai Chi action recognition method according to claim 1, characterized in that: In S4, the extracted shoulder key points and waist key points correspond one to one in the ground normal direction.
5. The Tai Chi action recognition method according to claim 1, characterized in that: In S5, according to the preset differential distance, two rotation angles separated by the differential distance in each rotation angle variation curve are differentiated to obtain the sensing optical fiber in each differential curve.
6. The Tai Chi action recognition method according to claim 1, characterized in that: In S6, The shoulder differential curve and the waist differential curve are superimposed on the same coordinate system, the data points in the shoulder differential curve and the waist differential curve that are less than a first preset threshold are deleted, and the transition data segments in the shoulder differential curve and the waist differential curve are deleted, and the shoulder differential curve and the waist differential curve are divided at the transition data segments to obtain multiple shoulder differential curves and multiple waist differential curves, and each shoulder differential curve and each waist differential curve are fitted into a straight line segment, and the position distribution of the multiple shoulder straight line segments and the multiple waist straight line segments in the time domain are compared with a pre-established force action feature library to identify the force action pattern.
7. The Tai Chi action recognition method according to claim 6, characterized in that: Delete the transition data segments in the shoulder differential curve and waist differential curve, including: Subtract adjacent values in the shoulder difference curve, and if the difference is greater than a second preset threshold, the two values are judged to be transition data and deleted; The adjacent values in the waist difference curve are subtracted. If the difference is greater than a third preset threshold, the two values are determined to be transition data and are deleted.
8. The Tai Chi action recognition method according to claim 6, characterized in that: The least square method is used to fit each shoulder difference curve and each waist difference curve into a straight line segment.
9. The Tai Chi action recognition method according to claim 6, characterized in that: The pre-established force action characteristics are the position distribution of multiple shoulder straight line segments and multiple waist straight line segments in the time domain.
10. A Tai Chi action recognition device, characterized in that: include: The first key point extraction module is used to obtain key point data of a target human body in a video sequence to be processed; A move recognition module is used to input key point data into a pre-trained Tai Chi move recognition model, recognize Tai Chi moves, and obtain key point data corresponding to the moves; The second key point extraction module is used to select the key point data of the force-generating action from the key point data corresponding to the identified Tai Chi moves using a preset key point data selection rule corresponding to the moves; The first curve drawing module is used to extract shoulder key point data and waist key point data according to the key point data of the force-generating action, and draw a rotation angle change curve of each shoulder key point over time and a rotation angle change curve of each waist key point over time; The second curve drawing module is used to perform differential processing on the curve of the rotation angle change over time of each shoulder key point to obtain the differential curve of each shoulder key point, add and average the differential curves of all shoulder key points to obtain the shoulder differential curve; perform differential processing on the curve of the rotation angle change over time of each waist key point to obtain the differential curve of each waist key point, add and average the differential curves of all waist key points to obtain the waist differential curve; The force action recognition module is used to superimpose the shoulder differential curve and the waist differential curve in the same coordinate system, delete the data points in the shoulder differential curve and the waist differential curve that are less than the first preset threshold, and compare the position distribution of the shoulder differential curve and the waist differential curve in the time domain with the pre-established force action feature library to identify the force action pattern.