Method for measuring motion based on ai recognition and motion measuring device

Abstract

The application discloses a kind of based on AI recognition's motion measurement method, the method includes: based on AI recognition model identification human skeleton data, according to human skeleton data to the user posture is identified;According to user posture determines the training project type;According to training project type determines corresponding preset specified action, according to preset specified action division standard posture determination area and call preset standard posture determination strategy, according to standard posture determination strategy real-time determination current user posture is eligible;If yes, then to standard posture determination area mark corresponding decomposition action stage and real-time according to the marking result between each standard posture determination area obtains measurement result;If no, then generate interactive information including the difference calculation result between current user posture and standard posture.The application also discloses a kind of motion measurement device.The application can improve detection efficiency and accuracy, reduce equipment investment cost, and can provide correct action guidance for user.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a motion measurement method and device based on AI recognition. Background Technology

[0002] Traditional sports training and assessment equipment often requires manual supervision during use. Humans need to judge the standard of the test subjects' movements and read the results (such as push-ups, sit-ups, etc.). The manual judgment method is prone to mistaking non-standard movements as correct and including them in the score. At the same time, it is easy for test subjects to cheat. Ultimately, these devices affect the fairness, impartiality and accuracy of the test results because they rely too much on human subjective judgment.

[0003] In recent years, electronic sports training and assessment equipment has emerged to replace manual labor. However, data collection, processing, and feedback are still handled by separate devices. Taking long-distance running tests as an example, multiple data collection devices are deployed along the route to collect data, which is then transmitted to a backend server for processing and analysis. Finally, the data is transmitted to a display device for feedback. This approach undoubtedly increases equipment costs and makes equipment deployment relatively cumbersome.

[0004] In addition, traditional sports training and assessment equipment are limited in function and lack universality, often requiring the relevant stakeholders to purchase a set of equipment for a single project. If the demand for a project is large, this will inevitably increase the purchase cost. Summary of the Invention

[0005] This invention provides a motion measurement method and device based on AI recognition, which can improve the efficiency and accuracy of motion detection, reduce equipment investment costs, and provide users with correct motion guidance.

[0006] To address the aforementioned technical problems, one technical solution adopted by the present invention is to provide a motion measurement method based on AI recognition, the motion measurement method comprising:

[0007] Based on the AI ​​recognition model, the user's human skeleton data is identified in real time from the continuous frame images acquired by the camera unit, and the human skeleton data of each frame image is obtained.

[0008] The user's movements are identified based on the human skeletal data to obtain the user's posture in each frame image;

[0009] The type of training program is determined based on the user's body shape.

[0010] According to the type of the test and training project, the corresponding preset prescribed actions are determined, the preset prescribed actions are decomposed to obtain decomposed actions, and the continuous frame images are statically divided according to the decomposed actions, and at least two standard posture judgment areas are obtained in each frame image.

[0011] The system monitors the user's posture in real time. When the user's posture overlaps with the corresponding standard posture judgment area, it invokes a preset standard posture judgment strategy that matches the type of the test or training project.

[0012] The current user's posture is determined to be qualified based on the decomposed actions corresponding to the standard posture determination area in the standard posture determination strategy.

[0013] If so, the standard body posture determination area is marked with the corresponding stage of the decomposed action, and the measurement result is obtained in real time based on the marking results between each standard body posture determination area;

[0014] If not, then interactive information is generated that includes the calculation results of the difference between the current user posture and the corresponding decomposed action.

[0015] According to an embodiment of the present invention, the step of performing real-time human skeleton data recognition on continuous frame images acquired by the camera unit based on an AI recognition model to obtain human skeleton data for each frame image includes:

[0016] The camera unit acquires a video stream of the user undergoing training in the motion area, and extracts continuous frame images from the video stream;

[0017] The frame images are input into the AI ​​recognition model, and human skeleton data is recognized in the frame images to obtain human skeleton data for each frame image.

[0018] According to an embodiment of the present invention, the step of inputting the frame image into the AI ​​recognition model, performing human skeleton data recognition on the frame image, and obtaining human skeleton data for each frame image includes:

[0019] The frame image is input into the AI ​​recognition model, and skeletal joint features are extracted from the frame image based on a convolutional neural network;

[0020] The skeletal joint features are classified using a decision tree algorithm to obtain the skeletal joint types;

[0021] The mean shift method based on weighted Gaussian kernels is used to predict the position of each type of bone joint, thereby obtaining the position and confidence level of each bone joint. The human skeletal data includes the position of the bone joint and the corresponding confidence level.

[0022] According to an embodiment of the present invention, the step of recognizing the user's movements based on the human skeletal data to obtain the user's body posture in each of the frame images includes:

[0023] Calculate the distance between bone joints based on the positions of each bone joint and the confidence level;

[0024] The angle between the lines connecting the bone joints is obtained based on the distance between the bone joints.

[0025] The user's body posture is determined based on the angle.

[0026] According to one embodiment of the present invention, the step of extracting consecutive frame images from the video stream further includes:

[0027] In the video stream, a preset number of consecutive raw frame images acquired by the camera unit are cached.

[0028] Perform statistical processing on the user data in the cached original frame images;

[0029] The original frame images are filtered based on the statistical processing results to extract consecutive frame images.

[0030] According to one embodiment of the present invention, immediately following the step of determining the type of training program based on the user's body posture, the method further includes:

[0031] Based on the type of the training project, the target key limb part of the current user's body posture is determined. Based on the target key limb part, the continuous frame images are dynamically divided into regions, and a corresponding foul posture judgment region is obtained in each frame image. The foul posture judgment region is used to track the user's target key limb part in the continuous frame images.

[0032] Before the step of determining whether the current user's posture is qualified based on the decomposed actions corresponding to the standard posture judgment area in the standard posture judgment strategy, the method further includes:

[0033] The standard posture judgment strategy determines whether the target key limb part is compliant based on the preset prohibited decomposition action corresponding to the foul posture judgment area.

[0034] If so, the current user posture is determined to be valid, and further determination is made as to whether the current user posture is qualified;

[0035] If not, the current user posture is determined to be invalid, and the determination continues to assess whether the target key limb part is compliant in the next standard posture determination area.

[0036] According to an embodiment of the present invention, the step of obtaining measurement results in real time based on the marking results between each of the standard body posture determination regions includes:

[0037] Until the end of the measurement, starting from the first marking of the standard body shape determination area, the marking results of each subsequent standard determination area are monitored sequentially;

[0038] When it is continuously monitored that each of the standard judgment areas is marked with the stage of the decomposed action, it is further determined whether the marking order of each of the standard judgment areas is consistent with the preset prescribed action order corresponding to the test and training project type;

[0039] If so, it is determined that the user has completed a preset action and accumulated one measurement and / or accumulated measurement time, and the marking results of each of the monitored standard judgment areas are eliminated.

[0040] If not, it is determined that the user has not completed a preset action and starting from the standard posture judgment area of ​​the corresponding marker that starts the decomposition action, the marking results of each subsequent standard judgment area are monitored sequentially, and the marking results of each standard posture judgment area before the start of the decomposition action are eliminated.

[0041] According to one embodiment of the present invention, the types of test and training items include: running, ball games, and movement postures; the standard posture judgment strategy is set in a one-to-one correspondence with the types of test and training items.

[0042] To address the aforementioned technical problems, another technical solution adopted by the present invention is: providing an AI-based motion measurement device, which is used to measure training and assessment items. The motion measurement device integrates an AI chip and a camera unit connected to the AI ​​chip. The AI ​​chip stores an AI recognition model for recognizing human skeletal data. The motion measurement device includes:

[0043] The first recognition module is used to recognize the user's human skeleton data in real time based on the AI ​​recognition model of the continuous frame images acquired by the camera unit, and obtain the human skeleton data of each frame image.

[0044] The second recognition module is used to recognize the user's actions based on the human skeleton data and obtain the user's body posture in each frame image;

[0045] The determination module is used to determine the type of training program based on the user's body posture.

[0046] The first division module is used to determine the corresponding preset prescribed actions according to the type of the test and training project, decompose the preset prescribed actions to obtain decomposed actions, and divide the continuous frame images into static regions according to the decomposed actions, so that at least two standard posture judgment regions are obtained in each frame image.

[0047] The monitoring module is used to monitor the user's posture in real time. When the user's posture is detected to overlap with the corresponding standard posture judgment area, a preset standard posture judgment strategy matching the test and training project type is invoked.

[0048] The first determination module is used to determine whether the current user's posture is qualified based on the decomposed actions corresponding to the standard posture determination area in the standard posture determination strategy.

[0049] The first execution module is used to mark the stage of the decomposed action corresponding to the standard body posture judgment area and obtain the measurement result in real time based on the marking result between each standard body posture judgment area if the first determination module determines that it is true.

[0050] The second execution module is used to generate interactive information including the difference calculation result between the current user posture and the corresponding decomposed action if the first determination module determines that it is not true.

[0051] According to one embodiment of the present invention, the motion measuring device further includes:

[0052] The second segmentation module is used immediately after the determination module determines the training item type based on the user's body posture, determines the target key limb parts of the current user's body posture based on the training item type, and dynamically segments the continuous frame images based on the target key limb parts, obtaining a corresponding foul posture judgment region in each frame image; the foul posture judgment region is used to track the user's target key limb parts in the continuous frame images.

[0053] The second determination module is used to determine whether the target key limb part is compliant based on the preset prohibited actions corresponding to the violation posture determination area in the standard posture determination strategy before the first determination module determines whether the current user posture is qualified based on the decomposed actions corresponding to the standard posture determination area in the standard posture determination strategy.

[0054] If so, the current user posture is determined to be valid and the first determination module is instructed to further determine whether the current user posture is qualified;

[0055] If not, the current user posture is determined to be invalid, and the determination continues to assess whether the target key limb part is compliant in the next standard posture determination area.

[0056] The beneficial effects of this invention are as follows: By using an AI recognition model to identify human skeletal data, and then identifying the user's posture based on this data, the method first determines the type of training exercise based on the user's posture. Then, it divides the standard posture judgment area according to the preset prescribed movements of the training exercise type and calls the corresponding standard posture judgment strategy. Finally, it judges whether the user's posture in each frame image is qualified according to the standard posture judgment strategy. This invention improves the efficiency and accuracy of motion detection. When the user's posture is judged as unqualified, it can provide correct movement guidance to the user based on the difference between the user's posture and the decomposed movements representing the corresponding standard posture. The motion measurement device adapted to this invention can be applied to different types of motion. With an integrated structure containing an AI chip and camera unit for recognizing human skeletal data, it can replace the cumbersome combination of monitoring camera sampling + network transmission + server-side computation and analysis in traditional motion measurement environments, thereby reducing equipment investment costs. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the structure of a motion measuring device according to an embodiment of the present invention;

[0058] Figure 2 This is a flowchart illustrating an AI-based motion measurement method according to an embodiment of the present invention.

[0059] Figure 3 This is a flowchart illustrating another embodiment of the motion measurement method based on AI recognition according to the present invention;

[0060] Figure 4 This is a schematic diagram of the structure of a motion measuring device according to another embodiment of the present invention;

[0061] Figure 5 This is a schematic diagram of the structure of a motion measuring device according to another embodiment of the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] The terms "first," "second," and "third" used in this invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this invention are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0064] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0065] This invention relates to the field of artificial intelligence technology, and provides a motion measurement method based on AI recognition. This motion measurement method is applied to the measurement of training and assessment items (sports training and assessment items) using motion measurement devices. Please refer to [link to relevant documentation]. Figure 1 The motion measurement device integrates an AI chip 10 and a camera unit 11 connected to the AI ​​chip 10. The AI ​​chip 10 stores an AI recognition model for identifying human skeletal data. In this embodiment, the camera unit 11 is integrated with the AI ​​chip 10 through a MIPI interface and an I2C interface. The camera unit 11 acquires image data in real time and transmits it to the AI ​​chip 10. The AI ​​chip 10 uses the AI ​​recognition model to identify human skeletal data in the image, realizing the data acquisition and processing capabilities that traditional sports training and assessment require multiple devices and network transmission. That is, the acquired data can be transmitted, processed, and stored independently within the same device.

[0066] In one feasible embodiment, please refer to Figure 1The motion measurement device also includes an illumination unit 12 connected to the AI ​​chip 10. The illumination unit 12 can be a high-brightness or adjustable high-power LED. The AI ​​chip 10 controls the illumination unit 12 to provide real-time supplementary lighting for the camera environment, which can be used to improve the quality of the images acquired by the camera unit 11.

[0067] In one feasible embodiment, please refer to Figure 1 The motion measurement device also includes a storage unit 13 connected to the AI ​​chip 10, such as a RAM unit and a Flash unit. The RAM unit is used for memory and has temporary data processing functions, while the Flash unit has application storage functions.

[0068] In one feasible embodiment, please refer to Figure 1 The motion measurement device also includes a display touch unit 14 and an audio unit 15 connected to the AI ​​chip 10. These two units can provide audio and video guidance for sports training and assessment projects, as well as timely feedback on results, based on human-computer interaction, while also facilitating device operation. Furthermore, combined with the display touch unit 14 and audio unit 15, it can assist users in completing functions such as automatic login based on facial recognition and identity verification (anti-cheating).

[0069] In one feasible embodiment, please refer to Figure 1 The motion measurement device is also equipped with a peripheral interface unit 16 connected to the AI ​​chip 10. The peripheral interface unit 16 provides functions including USB, serial port, TF card, and wireless data SIM card interface, supports other application expansion, and can connect to other peripherals.

[0070] In other embodiments, the motion measurement device further includes a network transmission unit 17 supporting wired and wireless transmission, and a power supply unit 18 for mobile use of the motion measurement device. Both the network transmission unit 17 and the power supply unit 18 are connected to the AI ​​chip 10. The power supply unit 18 can be a high-power rechargeable battery, facilitating the use of the motion measurement device in outdoor settings.

[0071] Based on the above structure, in actual testing and use, the testing and training project's measurement requirements can be met by deploying a single set of the motion measurement device of this invention, thus replacing the cumbersome combination of monitoring camera sampling + network transmission + server-side computation and analysis required in traditional motion measurement environments. Simultaneously, this motion measurement device uses human skeletal data recognition to achieve testing functions for various assessments and training exercises, such as running (e.g., short, middle, and long distance running, shuttle runs, serpentine runs, etc.), ball sports (e.g., volleyball, basketball, etc.), and movement postures (e.g., parallel bars, pull-ups, push-ups, sit-ups, etc.), which can improve the efficiency and accuracy of motion detection compared to traditional electronic testing and training equipment. Overall, the motion measurement device of this invention can reduce the equipment investment costs for the testing and training project's measurement requirements.

[0072] Figure 2 This is a flowchart illustrating a motion measurement method based on AI recognition according to an embodiment of the present invention. It should be noted that if substantially the same result is obtained, the method of the present invention is not necessarily identical. Figure 2 The illustrated process sequence is limited. For example... Figure 2 As shown, the method includes the following steps:

[0073] Step S101: Based on the AI ​​recognition model, the user's human skeleton data is identified in real time from the continuous frame images acquired by the camera unit to obtain the human skeleton data of each frame image.

[0074] In step S101, the AI ​​recognition model can be an image recognition model based on a convolutional neural network. This AI recognition model is stored in the AI ​​chip. The camera unit transmits continuously acquired NV12 format images in real time to the AI ​​chip via the MIPI interface. The AI ​​recognition model in the AI ​​chip is used to identify the human skeleton data of each frame image. The human skeleton data can include the position and confidence level of each bone joint. Since the human body has numerous bone joints, this embodiment only needs to identify the major bone joints that have a significant impact on posture. The bone joints in this embodiment include, but are not limited to, the nose, right eye, left eye, right ear, left ear, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, right hip, right knee, right ankle, left hip, left knee, and left ankle.

[0075] In one feasible embodiment, a video stream of a user undergoing training in a moving area is acquired by a camera unit, and continuous frame images are extracted from the video stream. The frame images are preprocessed, such as scaling and cropping, to convert them into a preset input image size. The preprocessed frame images are then input into an AI recognition model to perform human skeleton data recognition on the frame images, obtaining human skeleton data for each frame. Furthermore, accurate recognition of human images ensures the subsequent extraction of human skeleton data. In existing technologies, it is difficult to guarantee the accuracy of human skeleton data recognition by relying on corresponding recognition algorithms when performing further human skeleton data recognition processing on the frame images acquired by the camera unit. For example, if the background color is similar to the human clothing, or if objects resembling human body parts are mistakenly identified in the background, recognition jumps can easily occur when extracting the target human body part. For instance, the position of the target human body part obtained from a certain frame image may be abnormal and not match the actual position of the human body, or the area where the target human body part is located may be adhered to a similar background. Obviously, the aforementioned abnormally altered frame images are detrimental to the extraction of human skeleton data. In one feasible embodiment, during the extraction of continuous frame images from the video stream, a preset number of continuous raw frame images acquired by the camera unit can be cached. Statistical processing of user data in the raw frame images in the cache can be performed to remove frame images with abnormal user data, thereby ensuring the accuracy of the subsequent human skeleton data extraction process.

[0076] In one feasible embodiment, the specific process of using an AI recognition model to identify human skeletal data can be as follows: A frame image is input into the AI ​​recognition model; skeletal joint features are extracted from the frame image based on a convolutional neural network; the skeletal joint features are classified using a decision tree algorithm to obtain skeletal joint types; and the position of each skeletal joint type is predicted using a weighted Gaussian kernel mean shift method to obtain the position and confidence level of each skeletal joint. Specifically, human skeletal data includes skeletal joint positions and corresponding confidence levels. The decision tree algorithm in this embodiment is, for example, the random decision forest algorithm. In this algorithm, the forest is a collection of trees, each consisting of intermediate nodes and leaf nodes. Intermediate nodes are determined based on skeletal joint features. At the leaf nodes of the trees, the learned distribution of body part labels is stored. The average distribution of all trees in the forest is calculated to obtain the skeletal joint classification. The decision tree algorithm enables fast and effective classification of skeletal joints, improving recognition efficiency and accuracy.

[0077] Step S102: Recognize the user's movements based on human skeletal data to obtain the user's posture in each frame of the image.

[0078] In step S102, in one feasible embodiment, the distance between skeletal joints is calculated based on the position and confidence level of each skeletal joint; the angle between the lines connecting the skeletal joints is obtained based on the distance between the skeletal joints; and the user's posture is determined based on the angle. In this embodiment, users have different postures when performing different training tests. For example, the user's posture for pull-ups includes the upward movement, the bent-arm hanging movement, and the downward movement, while the user's posture for sit-ups includes: knees bent, hands on shoulders, elbows touching thighs, etc.

[0079] Step S103: Determine the type of training program based on the user's body shape.

[0080] In step S103, the training item types include: running, ball games, and movement postures. Running items include short, middle, and long distance running, shuttle runs, and serpentine runs; ball games include volleyball and basketball; and movement postures include parallel bars, pull-ups, push-ups, and sit-ups. Because different training items test different movement postures, each training item has a set of preset prescribed movements. This embodiment uses real-time comparison of the user's body posture with the preset prescribed movements of each training item. If the current user's body posture matches a preset prescribed movement for a certain item, and a training item can be uniquely distinguished from several similar preset prescribed movements by the sequence of the user's continuous body postures, then the training item can be determined as the type of training item the user is currently performing. By determining the training item type, the complete preset prescribed movements of the determined training item type can be invoked. The method provided in this embodiment can automatically determine the user's training items in real time. It does not require the user to complete a full set of prescribed actions before the training items can be automatically determined. Therefore, it reduces the amount of data processing. At the same time, it can promptly start the process of determining whether the user's physical posture is qualified after the training items are determined, resulting in higher work efficiency.

[0081] Step S104: Determine the corresponding preset prescribed actions according to the type of test and training project, decompose the preset prescribed actions to obtain the decomposed actions, divide the continuous frame images into static regions according to the decomposed actions, and obtain at least two standard posture judgment regions in each frame image.

[0082] In step S104, the preset prescribed movements are decomposed to obtain decomposed movements. For a training exercise type, the corresponding preset prescribed movements are a continuous process, while decomposed movements refer to representative postures that can be used to determine whether the movement is in place within this continuous process. These representative postures represent the main postures at each stage of the movement process, and their spatial positions are often fixed. For example, the decomposed movements of a pull-up include the preparatory posture and the bent-arm hanging posture, while the decomposed movements of a sit-up include the elbow-to-thigh posture, the downward posture, the lying posture, and the upward posture. Among these, the preparatory posture is the starting decomposed movement for a pull-up, and the elbow-to-thigh posture is the starting decomposed movement for a sit-up.

[0083] The system divides static regions into consecutive frames based on the decomposed actions, obtaining a standard posture judgment region in each frame. Each decomposed action corresponds to one standard posture judgment region. Each standard posture judgment region refers to the preset action position area for the user in the frame image based on the determined training item type, and is used to determine whether the user's posture is qualified. This invention can determine whether the user's posture is qualified simply by performing the judgment within the standard posture judgment region, reducing the judgment range and improving measurement efficiency. Each standard posture judgment region is set to start sequentially according to a preset action order, starting from the initial decomposed action corresponding to the training item type, to judge the user's posture respectively. Furthermore, each standard posture judgment region can automatically start according to the identified current decomposed action stage the user is actually in. The identified current decomposed action stage can be predicted based on factors such as the user's current spatial position and direction of movement; this invention does not impose any limitations on this. In one feasible embodiment, the standard posture judgment area corresponding to the user's initial decomposition action is the first standard posture judgment area to be activated. After the standard posture judgment area completes the user's posture judgment, the judgment function is closed, and then the next standard posture judgment area is activated to judge whether the next user's posture is qualified, and so on.

[0084] Step S105: Monitor the user's posture in real time. When the user's posture overlaps with the corresponding standard posture judgment area, call the preset standard posture judgment strategy that matches the test and training item type.

[0085] In step S105, the standard posture judgment strategy is set in a one-to-one correspondence with the test / training item type. For example, if the test / training item type is pull-ups, the corresponding standard posture judgment strategy is the pull-up standard posture judgment strategy. When the user's posture overlaps with the corresponding activated standard posture judgment area, only the standard posture judgment strategy corresponding to that standard posture judgment area needs to be called. For other standard posture judgment areas that are not involved, there is no need to call the corresponding judgment strategy, thereby improving strategy efficiency and avoiding misjudgment, thus improving measurement accuracy.

[0086] Step S106: Determine whether the current user's posture is qualified based on the decomposed actions corresponding to the standard posture judgment area in the standard posture judgment strategy.

[0087] In step S106, this embodiment determines whether the user's posture in each frame of the image is qualified according to the standard posture determination strategy, which improves the efficiency and accuracy of motion detection, can be applied to different types of motion, and reduces equipment investment costs.

[0088] The standard posture judgment strategy of this embodiment, for example, the standard posture judgment strategy for pull-ups includes: (1) Arm straightening judgment in the preparatory posture: when the angle formed between the wrist joint, elbow joint and shoulder joint of the arm is about 180°, the arm is considered to be straight; (2) Chin higher than the bar surface in the bent-arm hanging posture: compare the vertical coordinate of the real-time position of the chin with the vertical position of the bar surface to determine whether the chin is higher than the bar surface; (3) Body swing judgment in the bent-arm hanging posture: the approximate rectangle ABCD formed by the joints of the two shoulders and the two hips is projected onto the vertical plane as AB'C'D, and the size of the included angle ∠BAB' reflects the swing amplitude of the body relative to the vertical plane, thereby judging the body swing. For example, the standard posture judgment strategy for sit-ups includes: (1) Posture judgment when the elbow touches the leg: calculate the distance between the thigh (i.e., the line connecting the hip joint and the knee joint) and the elbow joint, and judge whether the elbow touches the leg according to the preset threshold; (2) Determination of the downward posture: the angle between the line connecting the shoulder joint and the hip joint and the horizontal plane is less than 5°; (3) Determination of the lying posture: the scapula touches the ground, i.e. the line connecting the shoulder joint and the hip joint is parallel to the horizontal plane; (4) Determination of the upward posture: the angle between the line connecting the shoulder joint and the hip joint and the horizontal plane is more than 15°.

[0089] Step S107: If so, mark the corresponding decomposed action stage for the standard posture judgment area and obtain the measurement result in real time based on the marking results between each standard posture judgment area.

[0090] In one feasible embodiment, before the measurement ends, starting from the first marked standard posture judgment area, the marking results of subsequent standard judgment areas are sequentially monitored; when it is detected that each standard judgment area is marked with the corresponding decomposition action stage, it is further determined whether the marking order of each standard judgment area is consistent with the preset prescribed action order corresponding to the test and training item type; if so, it is determined that the user has completed one preset prescribed action and accumulated one measurement count and / or accumulated measurement time, and the marking results of each monitored standard judgment area are eliminated; if not, it is determined that the user has not completed one preset prescribed action and starting from the standard posture judgment area corresponding to the start of the decomposition action, the marking results of subsequent standard judgment areas are sequentially monitored, and the marking results of each standard posture judgment area before the start of the decomposition action are eliminated.

[0091] For example, in push-up training, the process from the user's body starting in the extended position to the user lowering their body until the elbows are level with the shoulders is defined into four states: posture 1 represents the extended position (body and legs straight, arms straight, body and horizontal ground angle close to 20°, the same below), posture 2 represents the initial lowering position (e.g. close to 15°), posture 3 represents the deep lowering position (e.g. close to 5°), and posture 4 represents the prone position (e.g. close to 0°). The complete push-up breakdown includes 1-2-3-4-3-2-1. If a user's posture is continuously monitored and meets the pattern 1-2-3-4-3-2-1, it means that the corresponding decomposed actions in each standard posture judgment area have been marked continuously, and the marking order is consistent with the preset prescribed action order. It is considered that the user has completed the preset prescribed action once, and the number of times is accumulated. At the same time, the marking result is cleared to facilitate subsequent marking based on the user's posture. If a user's posture is continuously monitored and meets the pattern 1-2-3-2-1 (representing an incorrect action of not reaching the bottom) or 4-3-2-3-4 (representing an incorrect action of not being prepared), it means that the corresponding decomposed actions in each standard posture judgment area have not been marked continuously. In this case, monitoring needs to continue until the corresponding decomposed actions in each standard posture judgment area are continuously marked.

[0092] If we consider adjacent repetitive actions consecutively, there may be special cases, such as 2-3-4-3-2-1-2-3-4-3-2-1 (meaning that a certain posture at the beginning was not qualified, but the postures after subsequent adjustments were qualified). The preceding segment 2-3-4-3-2-1 also satisfies the condition that all standard posture judgment areas have been marked according to the corresponding decomposed actions, but the marking order is inconsistent with the preset prescribed action order. In this case, starting from the standard posture judgment area corresponding to posture 1 (the initial decomposed action) that is monitored, the marking results of each standard posture judgment area should be monitored again, and the marking results of each standard posture judgment area before posture 1 (the initial decomposed action) should be eliminated.

[0093] Step S108: If not, generate interactive information including the calculation results of the difference between the current user's posture and the corresponding decomposed action.

[0094] In step S108, the user's posture is analyzed, and the difference between it and the corresponding decomposed movements is calculated. This provides the user with correct movement guidance, thereby improving the user experience. In this embodiment, the first interactive information can facilitate human-computer interaction through voice or video.

[0095] Based on the above embodiments, in one feasible embodiment, please refer to [link to relevant documentation]. Figure 3 Following step S103, the following is also included:

[0096] Step S109: Determine the target key limb parts of the current user's body posture according to the type of test and training project, and dynamically divide the continuous frame images according to the target key limb parts, so as to obtain the corresponding foul posture judgment area in each frame image.

[0097] The foul posture determination area in this embodiment is used to track the user's target key limb parts in consecutive frame images to directly determine whether the current user's posture is a foul, that is, to determine whether the corresponding user posture is valid. For example, in pull-ups, the target key limb parts include the feet, so in each frame image, it tracks in real time whether the feet touch the bar; in sit-ups, the target key limb parts are the knees, so in each frame image, it tracks in real time whether the knee bending angle is approximately 90°; or if the key limb parts are the hands, it tracks in real time whether the hands are placed behind the head. Such postures are preset prohibited actions for each training item, which are commonly referred to as foul actions. Prohibited actions do not directly reflect the stage of the user's posture, but are used to continuously monitor whether the user has committed a foul throughout the entire training item.

[0098] Based on the above embodiments, in one feasible embodiment, please refer to [link to relevant documentation]. Figure 3 Before step S106, the method further includes:

[0099] Step S110: Determine whether the target key limb part is compliant based on the preset prohibited actions corresponding to the foul posture judgment area in the standard posture judgment strategy.

[0100] In step S110, if yes, the current user posture is determined to be valid and step S106 is executed. If no, the current user posture is determined to be invalid and the determination continues to determine whether the target key limb parts are compliant in the next standard posture determination area.

[0101] This embodiment obtains the foul posture judgment area through dynamic region division. In the foul posture judgment area, it can judge whether a foul has occurred for a key part of the human body, which further improves the accuracy and reliability of the measurement and can prevent cheating.

[0102] Figure 4 This is a schematic diagram of the motion measurement device based on AI recognition according to an embodiment of the present invention. Figure 4 As shown, the device 30 includes a first identification module 31, a second identification module 32, a determination module 33, a first division module 34, a monitoring module 35, a first judgment module 36, a first execution module 37, and a second execution module 38.

[0103] The first recognition module 31 is used to recognize the user's human skeleton data in real time based on the AI ​​recognition model of the continuous frame images acquired by the camera unit, and obtain the human skeleton data of each frame image.

[0104] The second recognition module 32 is used to recognize the user's movements based on human skeletal data and obtain the user's posture in each frame of the image;

[0105] Module 33 is used to determine the type of training program based on the user's body shape;

[0106] The first division module 34 is used to determine the corresponding preset prescribed actions according to the type of test and training project, decompose the preset prescribed actions to obtain the decomposed actions, and divide the continuous frame images into static regions according to the decomposed actions, so that at least two standard posture judgment regions are obtained in each frame image.

[0107] The monitoring module 35 is used to monitor the user's posture in real time. When it detects that the user's posture overlaps with the corresponding standard posture judgment area, it calls the preset standard posture judgment strategy that matches the test and training project type.

[0108] The first judgment module 36 is used to determine whether the current user's posture is qualified based on the decomposed actions corresponding to the standard posture judgment area in the standard posture judgment strategy.

[0109] The first execution module 37 is used to mark the corresponding decomposed action stage in the standard body posture judgment area and obtain the measurement result in real time based on the marking results between each standard body posture judgment area if the first determination module determines that it is true.

[0110] The second execution module 38 is used to generate interactive information including the calculation results of the difference between the current user's body posture and the corresponding decomposed action if the first determination module determines that it is not true.

[0111] In one feasible embodiment, please refer to Figure 5 The device 30 also includes a second division module 39 and a second determination module 310.

[0112] The second division module 39 is used to determine the target key limb part of the current user's body shape according to the target key limb part after the determination module 33 determines the test and training item type based on the user's body shape, and to dynamically divide the continuous frame images according to the target key limb part, so that the corresponding foul body shape judgment area is obtained in each frame image.

[0113] The second judgment module 310 is used to determine whether the target key limb parts are compliant based on the preset prohibited actions corresponding to the violation posture judgment area in the standard posture judgment strategy before the first judgment module 36 determines whether the current user's posture is qualified according to the decomposed actions corresponding to the standard posture judgment area in the standard posture judgment strategy.

[0114] If so, the current user's posture is determined to be valid and the first determination module 36 is instructed to further determine whether the current user's posture is qualified;

[0115] If not, the current user's posture is determined to be invalid, and the system will continue to determine whether the target key limb parts are compliant in the next standard posture determination area.

[0116] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0117] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0118] The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A motion measurement method based on AI recognition, characterized in that, The motion measurement method comprises: real-time user human skeleton data recognition of continuous frame images obtained by a camera unit based on an AI recognition model, to obtain human skeleton data of each frame image; user body posture of each frame image is obtained by recognizing the user's action according to the human skeleton data; determine the type of the test project according to the user body posture; according to the test project type, determine the corresponding preset specified action, decompose the action, obtain the decomposed action, and divide the continuous frame images according to the decomposed action, to obtain at least two standard body posture judgment areas in each frame image; real-time monitoring of the user body posture, when the user body posture and the corresponding standard body posture judgment area exist, the standard body posture judgment strategy matched with the test project type is called; whether the current user body posture is qualified is determined according to the decomposed action corresponding to the standard body posture judgment area in the standard body posture judgment strategy; if yes, the standard body posture judgment area is marked with the corresponding decomposed action, and the measurement result is obtained according to the marking result between each standard body posture judgment area; if not, the interactive information including the difference calculation result between the current user body posture and the corresponding decomposed action is generated. 2.The AI recognition-based motion measurement method of claim 1, wherein, The step of real-time user human skeleton data recognition of continuous frame images obtained by a camera unit based on an AI recognition model, to obtain human skeleton data of each frame image comprises: extracting continuous frame images from a video stream obtained by the camera unit when the user is testing in a motion area; input the frame image into the AI recognition model, and perform human skeleton data recognition on the frame image to obtain human skeleton data of each frame image. 3.The AI recognition-based motion measurement method of claim 2, wherein, The step of inputting the frame image into the AI recognition model, performing human skeleton data recognition on the frame image, and obtaining human skeleton data of each frame image comprises: input the frame image into the AI recognition model, and extract skeletal joint features from the frame image based on a convolutional neural network; use a decision tree algorithm to classify and process the skeletal joint features to obtain skeletal joint types; predict the positions of each skeletal joint type based on a weighted Gaussian kernel mean shift method to obtain skeletal joint positions and confidence levels; the human skeleton data comprises the skeletal joint positions and corresponding confidence levels. 4.The AI recognition-based motion measurement method of claim 3, wherein, The step of recognizing the user's action according to the human skeleton data to obtain the user body posture of each frame image comprises: calculate the distance between skeletal joint nodes according to the skeletal joint positions and the confidence levels; obtain the angle between the skeletal joint node connecting lines according to the distance between the skeletal joint nodes; determine the user body posture according to the angle. 5.The AI recognition-based motion measurement method of claim 2, wherein, The step of extracting continuous frame images from the video stream further comprises: in the video stream, buffer the continuous original frame images obtained by the camera unit according to a preset number; statistically process the user data in the buffered original frame images; According to the statistical processing result, the original frame images are screened, and continuous frame images are extracted. 6.The AI recognition-based motion measurement method of claim 1, wherein, After the step of determining the test item type according to the user posture, the method further comprises: According to the test item type, a target key body part of the current user posture is determined, and the continuous frame images are dynamically regionally divided according to the target key body part, so that a rule-breaking posture judgment region is obtained in each frame image; the rule-breaking posture judgment region is used to track the target key body part of the user in the continuous frame images; Before the step of determining whether the current user posture is qualified according to the decomposition action corresponding to the standard posture judgment region in the standard posture judgment strategy, the method further comprises: According to the preset prohibited action corresponding to the rule-breaking posture judgment region in the standard posture judgment strategy, it is determined whether the target key body part is compliant; If yes, it is determined that the current user posture is valid, and it is further determined whether the current user posture is qualified; If no, it is determined that the current user posture is invalid, and it is further determined whether the target key body part is compliant at the next standard posture judgment region. 7.The AI recognition-based motion measurement method of claim 1, wherein, The step of obtaining a measurement result according to the marking result between each standard posture judgment region in real time comprises: Before the measurement is completed, the marking results of the subsequent standard posture judgment regions are sequentially monitored since the standard posture judgment region is first marked; When it is continuously monitored that each standard posture judgment region is marked with the corresponding decomposition action, it is further determined whether the marking sequence of each standard posture judgment region is consistent with the preset specified action sequence corresponding to the test item type; If yes, it is determined that the user completes a preset specified action once, and the measurement number and / or the measurement time are accumulated once, and the marking results of the monitored standard posture judgment regions are eliminated; If no, it is determined that the user does not complete a preset specified action once, and the marking results of the standard posture judgment regions before the starting decomposition action are eliminated since the corresponding marking starting standard posture judgment region is monitored. 8.The AI recognition-based motion measurement method of claim 1, wherein, The test item type comprises running, ball games, and action postures; the standard posture judgment strategy is set in one-to-one correspondence with the test item type. 9.A motion measurement device based on AI recognition, characterized by The sports measurement device is used to measure test items, and an AI chip and a camera unit connected with the AI chip are integrated in the sports measurement device; an AI recognition model for recognizing human body skeleton data is stored in the AI chip; the sports measurement device comprises: A first recognition module is configured to recognize human body skeleton data of a user in continuous frame images obtained by the camera unit in real time based on an AI recognition model, so as to obtain human body skeleton data of each frame image; A second recognition module is configured to recognize an action of the user according to the human body skeleton data, so as to obtain a user posture of each frame image; A determination module is configured to determine a test item type according to the user posture. The first dividing module is configured to determine a corresponding preset specified action according to the test item type, perform action decomposition on the preset specified action, obtain decomposed actions, and perform static region division on continuous frame images according to the decomposed actions. The monitoring module is configured to monitor the user body state in real time, and call a preset standard body state judgment strategy matched with the test item type when the user body state overlaps with the corresponding standard body state judgment region. The first judgment module is configured to determine whether the current user body state is qualified according to the decomposed action corresponding to the standard body state judgment region in the standard body state judgment strategy. The first execution module is configured to mark the standard body state judgment region with a stage corresponding to the decomposed action and obtain a measurement result according to the marking result between the standard body state judgment regions in real time if the first judgment module determines that the current user body state is qualified. The second execution module is configured to generate interaction information including a difference calculation result between the current user body state and the corresponding decomposed action if the first judgment module determines that the current user body state is not qualified.

10. The motion measuring apparatus according to claim 9, wherein The motion measurement device further comprises: The second dividing module is configured to determine a target key limb part of the current user body state according to the test item type after the determining module determines the test item type according to the user body state, perform dynamic region division on continuous frame images according to the target key limb part, and obtain a foul body state judgment region in each frame image. The second judgment module is configured to determine whether the target key limb part is compliant according to a preset prohibited action corresponding to the foul body state judgment region in the standard body state judgment strategy before the first judgment module determines whether the current user body state is qualified according to the decomposed action corresponding to the standard body state judgment region in the standard body state judgment strategy. If yes, it is determined that the current user body state is valid, and the first judgment module is instructed to further determine whether the current user body state is qualified. If no, it is determined that the current user body state is invalid, and whether the target key limb part is compliant when the next standard body state judgment region is determined is continued to be determined.

Images