Attitude data acquisition method and system based on virtual reality technology, and storage medium

Abstract

The invention relates to the technical field of robots, and discloses a posture data collection method and system based on the virtual reality technology and a computer readable storage medium, and the method comprises the steps: determining the current real posture data when a user executes a preset task; mapping the current real posture data into current virtual posture data of a virtual human body model in the virtual reality equipment; driving the virtual human body model to move along with the user according to the current virtual posture data; if the current virtual attitude data does not meet the action quality requirement, determining the current virtual attitude data as rerecording attitude data, and generating action rerecording guidance information corresponding to the rerecording attitude data; and displaying action re-recording guidance information to the user based on the virtual reality equipment to guide the user to re-execute a re-recording task segment corresponding to the re-recording posture data in the preset task. In this way, the quality of the collected attitude data is improved.

Description

Technical Field

[0001] This application relates to the field of robotics, specifically to a posture data acquisition method, system, and computer-readable storage medium based on virtual reality technology. Background Technology

[0002] With the rapid development of computer technology, sensor technology, artificial intelligence, and robotics, humanoid robot technology has continuously progressed and improved. Breakthroughs in technologies such as deep learning and computer vision, in particular, have provided humanoid robots with more powerful perception, learning, and decision-making capabilities. Humanoid robots have broad application prospects in many fields, such as home services, medical care, and public safety. Due to rising living standards and increasing labor costs, the demand for robots capable of replacing humans in repetitive, tedious, or dangerous tasks is increasing daily.

[0003] When training a robot to complete a specific task, it is necessary to collect posture data corresponding to that task for teaching the robot. However, existing technologies typically collect posture data according to preset procedures or human experience, ignoring issues such as joint abnormalities, sudden posture changes, or discontinuities during the collection process. Often, data quality defects can only be discovered after data collection is completed through offline playback or post-processing analysis. This results in a large amount of invalid or low-value data that is beyond the robot's capabilities being generated during the collection process, meaning that only a small portion of the large amount of posture data collected is usable.

[0004] Therefore, improving the quality of the acquired attitude data has become a pressing technical problem. Summary of the Invention

[0005] In view of the above problems, embodiments of this application provide a posture data acquisition method, system and computer-readable storage medium based on virtual reality technology to solve the problem of low quality of acquired posture data in the prior art.

[0006] According to one aspect of the embodiments of this application, a posture data acquisition method based on virtual reality technology is provided. The method includes: determining the current real posture data of a user performing a preset task based on a joint tracking device worn by the user; mapping the current real posture data to the current virtual posture data of a virtual human body model in a virtual reality device, wherein the virtual reality device is worn on the user's head and is used to display the virtual human body model to the user; driving the virtual human body model in the virtual reality device to follow the user's movement according to the current virtual posture data; determining whether the current virtual posture data meets preset motion quality requirements; if the current virtual posture data does not meet the motion quality requirements, determining the current virtual posture data as re-recorded posture data, and generating motion re-recording guidance information corresponding to the re-recorded posture data; displaying the motion re-recording guidance information to the user based on the virtual reality device to guide the user to re-execute the re-recorded task segment in the preset task corresponding to the re-recorded posture data; and storing the real posture data corresponding to the virtual posture data that meets the motion quality requirements during the user's execution of the preset task to complete the posture data acquisition.

[0007] Preferably, the re-recording task segment corresponds to multiple virtual posture data. After displaying action re-recording guidance information to the user based on the virtual reality device to guide the user to re-execute the re-recording task segment corresponding to the re-recording posture data in the preset task, the method further includes: generating prompt information corresponding to the re-recording posture data, wherein the prompt information is used to indicate that the current posture of the virtual human body model is in an unqualified posture; driving the virtual human body model in the virtual reality device to move again according to the multiple virtual posture data corresponding to the re-recording task segment, and presenting prompt information in the virtual reality device when the virtual human body model in the virtual reality device moves to the posture corresponding to the re-recording posture data.

[0008] Preferably, storing the real posture data corresponding to the virtual posture data that meets the action quality requirements during the user's execution of the preset task to complete posture data acquisition includes: identifying repeatedly executed task segments during the user's execution of the preset task to obtain at least one group of repeated task segments, wherein the group of repeated task segments includes multiple repeatedly executed task segments, and each executed task segment corresponds to multiple virtual posture data; determining the number of re-recorded posture data in the virtual posture data corresponding to each executed task segment in the group of repeated task segments as the re-recording amount of the executed task segment; storing the real posture data corresponding to the task segment with the fewest re-recording amounts in the group of repeated task segments, and the real posture data corresponding to the task segments that were not repeatedly executed during the user's execution of the preset task to complete posture data acquisition.

[0009] Preferably, determining whether the current virtual posture data meets the preset motion quality requirements includes: making a preliminary judgment on the current virtual posture data according to preset motion rationality rules to obtain a preliminary judgment result; if the preliminary judgment result is that the current virtual posture is reasonable, then the current virtual posture data is determined to meet the motion quality requirements; if the preliminary judgment result is that the current virtual posture is unreasonable, then the current virtual posture data is input into a pre-trained visual language model to obtain a model judgment result; if the model judgment result is that the current virtual posture is reasonable, then the current virtual posture data is determined to meet the motion quality requirements; if the model judgment result is that the current virtual posture is unreasonable, then the current virtual posture data is determined to not meet the motion quality requirements.

[0010] Preferably, the motion rationality rules include joint movement rules and joint distance rules; the current virtual posture data is initially judged according to the preset motion rationality rules to obtain a preliminary judgment result, including: determining the degree of joint movement of the current virtual posture based on the key joint data in the previous virtual posture data and the key joint data in the current virtual posture data; determining the joint distance between multiple key joints in the current virtual posture based on the key joint data in the current virtual posture data; if the degree of joint movement satisfies the joint movement rules and the joint distance between multiple key joints satisfies the joint distance rules, then the current virtual posture is judged to be rational.

[0011] Preferably, the joint movement rule includes that the degree of joint movement does not exceed a preset joint movement degree; the joint distance rule includes that the joint distances between multiple key joints are all within a preset joint distance range; or, the joint distance rule includes that the current distance ratio is within a preset reasonable ratio range corresponding to the distance ratio item; wherein, the determination process of the current distance ratio being within the preset reasonable ratio range corresponding to the distance ratio item includes: for the preset distance ratio item, based on the current virtual posture data, determining the first distance of the first joint set required to constitute the distance ratio item, and the second distance of the second joint set, wherein the first distance is a geometric metric calculated based on the spatial coordinates of the multiple joints included in the first joint set, and the second distance is... The geometric metric is calculated based on the spatial coordinates of multiple joints contained in the second joint set; the ratio of the first distance to the second distance is determined as the current distance ratio of the distance ratio term; it is determined whether the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio term; if the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio term, it is determined that the joint distance between multiple key joints satisfies the joint distance rule; wherein, the distance ratio term defines the first joint set and the second joint set on which it is based, and the relationship between the first joint set and the second joint set includes at least one of the following: the adjacent relationship of ipsilateral limbs, the symmetrical relationship about the body midline, or the geometric relationship that together form a specific functional triangle.

[0012] Preferably, the joint tracking device includes multiple joint trackers for wearing on key body positions of the user. The key body positions include at least one of the head, torso, arms, hands, pelvis, and lower limbs, and the key joint data is data corresponding to the key body positions.

[0013] According to another aspect of the embodiments of this application, a posture data acquisition system based on virtual reality technology is provided. The system includes a server, a joint tracking device, and a virtual reality device, wherein the server is communicatively connected to the joint tracking device and the virtual reality device respectively; the joint tracking device is worn by a user and collects the user's current real posture data when performing a preset task, and sends the current real posture data to the server; the server is used to execute the method described in any of the preceding claims, wherein the server generates driving data for driving a virtual human body model based on the current real posture data, and sends the driving data and the motion re-recording guidance information corresponding to the re-recorded posture data to the virtual reality device; the virtual reality device is used to display the motion state of the virtual human body model and the motion re-recording guidance information, and drives the virtual human body model to follow the user's movement according to the driving data.

[0014] According to another aspect of the embodiments of this application, a posture data acquisition system based on virtual reality technology is provided. The system includes a joint tracking device and a virtual reality device with a communication connection; the joint tracking device is used to be worn by a user and to collect the current real posture data of the user when performing a preset task, and to send the current real posture data to the virtual reality device; the virtual reality device is used to perform the method described in any of the preceding claims.

[0015] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the posture data acquisition method based on virtual reality technology as described in any of the preceding claims.

[0016] This application embodiment determines the user's current real posture data when performing a task and maps this data in real time to the current virtual posture data of the virtual human model in the virtual reality device. This achieves synchronous visualization of real actions in the virtual space of the virtual reality device, making data quality observable. Furthermore, while driving the virtual model to follow the user's movement, it determines whether the current virtual posture data meets preset action quality requirements, transforming quality assessment from post-event analysis to real-time monitoring. This instantly identifies joint abnormalities, sudden posture changes, and other issues. If the data does not meet the requirements, it is immediately identified as re-recorded posture data, and corresponding action re-recording guidance information is generated. The virtual reality device guides the user to re-execute only the re-recorded task segment corresponding to the re-recorded data, replacing the inefficient method of relying on human experience or overall retries. This directly corrects erroneous actions during the acquisition process. Finally, only the real posture data corresponding to the virtual posture data that meets the action quality requirements is stored, completing source filtering before data is entered into the database. This ensures that each frame of data collected meets the preset robot capability range and task logic requirements, fundamentally eliminating the generation and accumulation of invalid data, improving the overall quality and usability of the dataset, and achieving efficient and high-quality posture data acquisition.

[0017] The above description is merely an overview of the technical solutions of the embodiments of this application. In order to better understand the technical means of the embodiments of this application and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of this application more obvious and understandable, specific implementation methods of this application are described below. Attached Figure Description

[0018] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic diagram of the structure of the posture data acquisition system based on virtual reality technology provided in an embodiment of this application is shown; Figure 2 This invention provides a schematic diagram of the structure of another posture data acquisition system based on virtual reality technology, according to an embodiment of this application. Figure 3 A flowchart illustrating the posture data acquisition method based on virtual reality technology provided in an embodiment of this application is shown. Figure 4 A schematic diagram of the structure of the posture data acquisition device based on virtual reality technology provided in an embodiment of this application is shown; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0019] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein.

[0020] Figure 1 A schematic diagram of the structure of the posture data acquisition system based on virtual reality technology provided in an embodiment of this application is shown, as follows: Figure 1 The system shown includes a server, a joint tracking device, and a virtual reality device. The server is communicatively connected to both the joint tracking device and the virtual reality device. The joint tracking device is worn by the user and collects the user's current real-time posture data when performing a preset task, and sends the current real-time posture data to the server. The server executes a posture data acquisition method based on virtual reality technology. The server generates driving data for driving the virtual human body model based on the current real-time posture data, and sends the driving data and the corresponding motion re-recording guidance information to the virtual reality device. The virtual reality device displays the motion state of the virtual human body model and the motion re-recording guidance information, and drives the virtual human body model to follow the user's movements based on the driving data.

[0021] The server can be a high-performance workstation deployed locally or a server cluster located in the cloud. The specific configuration depends on the data volume and real-time requirements.

[0022] The server executes a posture data acquisition method based on virtual reality technology through a built-in computer program. First, as a data processing center, the server receives raw motion data from the joint tracking device in real time. Second, as a posture calculation center, the server performs fusion calculations on the received data to generate calculation results that accurately describe the posture of each joint and the overall motion state of the human body. Third, as a drive data generation center, the server generates drive data that can drive the virtual human body model to move synchronously based on the calculation results. Finally, as a quality assessment and guidance decision center, the server performs real-time quality analysis on the calculated posture data, and generates corresponding action re-recording guidance information when the data does not meet the preset action quality requirements.

[0023] The server establishes a stable data communication link with the joint tracking device and virtual reality equipment through wireless communication networks (such as Wi-Fi, Bluetooth, or 5G) to ensure low latency and high reliability of data transmission.

[0024] The joint tracking device comprises multiple joint trackers worn on key body parts of the user. Each joint tracker is a hardware unit used to directly collect the user's motion data. Joint trackers are typically wearable, miniaturized sensor nodes designed for easy attachment to key joints of the body, such as via straps, clips, or adhesive tape to the head, torso, arms, hands, pelvis, and lower limbs. Each joint tracker integrates an Inertial Measurement Unit (IMU), which consists of a three-axis gyroscope, a three-axis accelerometer, and an optional three-axis magnetometer.

[0025] Preferably, the key body positions include at least one of the following: head, torso, both arms, both hands, pelvis, and lower limbs. The key joint data is the data corresponding to the key body positions. The data collected and processed by the system corresponding to these key body positions is the key joint data. By deploying trackers at these specific locations, the system can obtain sufficiently rich and accurate motion information.

[0026] When a user wears a joint tracking device and performs a preset task, the IMU in each joint tracking device continuously senses and collects information on the angular velocity, acceleration, and changes in the direction relative to the Earth's magnetic field of the corresponding joint during movement. This raw sensor data is initially packaged and preprocessed by the device's internal microprocessor, and then transmitted to the server in real time via its built-in wireless communication module.

[0027] The number and location of joint tracking devices can be flexibly configured according to the accuracy requirements of data acquisition and the needs of human body's degrees of freedom of movement, ensuring that comprehensive joint movement information of the whole body can be captured.

[0028] Virtual reality devices (VR devices) are display terminals used to provide users with immersive visual feedback and interactive guidance. VR devices are head-mounted displays that include a display screen for presenting stereoscopic images, an optical lens system, and head motion tracking sensors.

[0029] Virtual reality devices have two main ways of driving the movement of virtual human models: The first way is indirect server-driven, where the virtual reality device is responsible for display and presentation. The server completes all posture calculations, drive data generation, and virtual human model motion calculations, and then sends the calculated form and posture data (i.e., rendering data) of each frame of the virtual human model to the virtual reality device. The virtual reality device is only responsible for receiving this data and rendering it on the screen to achieve visual display. The second way is direct virtual reality device-driven, where the server sends the calculated joint angles, displacements, and other drive data to the virtual reality device. The graphics processing unit (GPU) inside the virtual reality device calculates and drives the virtual human model to update its motion in real time based on the received drive data. The server is mainly responsible for sending drive data and receiving possible status feedback. Regardless of the driving method used, the core functions of virtual reality devices are: firstly, to display a virtual human body model that moves in real time, enabling synchronous visualization of real movements in virtual space, allowing users to intuitively perceive their own movement state and posture details; secondly, to synchronously display motion re-recording guidance information sent from the server. This information can be visually overlaid on the virtual human body model or displayed in a suitable position in the field of view (e.g., highlighting virtual joints that need adjustment, displaying text prompts, indicating the re-recording path in virtual space), thereby guiding users to make targeted adjustments and re-records. In addition, virtual reality devices can also receive simple interactive commands from users via controllers, gestures, or eye contact, and feed these commands back to the server, such as confirming the start of data acquisition or marking task segments.

[0030] In this application, "user" refers to a human body participating in the posture data acquisition task. During the operation of this system, the user needs to perform a series of actions according to the preset task specifications.

[0031] Users wear joint tracking devices, allowing the system to sense and record their joint movements in real time. Simultaneously, by wearing virtual reality devices, users can observe the movement of their digital avatar—a virtual human model—within the virtual reality environment during task execution, and receive quality assessment feedback and re-recording instructions from the system.

[0032] The user's role is both the executor of actions and the real-time monitor and adjuster of data quality.

[0033] Among them, the current true posture data refers to the set of raw or preliminary processed data collected by the joint tracking device at any time to characterize the current motion state of each joint of the user's body.

[0034] In terms of data content, the current real-time posture data includes sensor data such as angular velocity and acceleration obtained from various joint tracking devices, as well as joint angle and angular velocity change trends or relative displacement information obtained through preliminary fusion calculations. In terms of data dimensions, the current real-time posture data covers the motion information of multiple joints throughout the body, collectively describing the overall posture of the human body at that moment. The current real-time posture data serves as the foundational input data for all subsequent processing (solution, visualization, quality assessment, and storage) of this system.

[0035] A pre-designed task refers to a sequence of actions or specific skills that are pre-designed and required from the user to train a robot or build a behavioral dataset. For example, a pre-designed task might be "moving an object from point A to point B and placing it there," "completing a set of gymnastic movements," or "simulating the fine motor skill of tightening a screw." Pre-designed tasks define the goals and scope of data collection, typically including the requirements for the start, intermediate, and end states of the actions. During the data collection process, this system consistently uses the user's correct and high-quality completion of the pre-designed task as a crucial criterion for data quality assessment.

[0036] The driving data consists of information generated by the server based on the current real-time posture data, used to precisely control the movement of the virtual human model. Driving data can take different forms, depending on whether a server-indirect driving mode or a virtual reality device-direct driving mode is used. In the server-indirect driving mode, the driving data might be the vertex positions, skeletal transformation matrices, or complete frame rendering data of each frame of the server-rendered virtual human model. In the virtual reality device-direct driving mode, the driving data mainly includes calculated and optimized information such as the rotation angles (Euler angles or quaternions) and displacements of various joints (e.g., shoulder, elbow, hip, knee) of the human body, as well as information on skeletal hierarchy relationships. The core function of the driving data is to ensure that the movement state of the virtual human model can be synchronized with the user's real-time movement accurately and in real time, providing a foundation for visual feedback and interaction.

[0037] A virtual human model is a digital 3D human body model constructed in the virtual environment of a virtual reality device to represent the user. This model has a hierarchical system of joints and bones corresponding to human anatomy and can generate movement based on input actuation data. The virtual human model is rendered and displayed in the virtual reality device, and its motion state (i.e., the angles of each joint, the position and orientation of the body) changes in real time following the user's current real-world posture data. The virtual human model is not only the core carrier of visual feedback, but its motion trajectory and state also serve as a direct object for data quality assessment and the generation of re-recording guidance.

[0038] Motion state refers to the posture and pose information of a virtual human model in virtual space as it changes over time. Specifically, it is reflected in the current rotation angle of each joint, the relative positions between bones, and the overall position coordinates and orientation of the model in virtual space. Motion state is the direct result of driving data acting on the virtual human model, presented to the user in a dynamic and continuous visual format through virtual reality devices, allowing the user to intuitively see the digital representation of their own movements.

[0039] When the server determines that the currently collected posture data has quality issues, it generates and transmits information to the virtual reality device to guide the user to correct or re-execute the relevant actions; this is called motion re-recording guidance information. This information takes various forms, including visual cues such as highlighting joints with abnormal postures on the virtual human model (e.g., marking abnormal joints in red), indicating the trajectory or position of the correct action with dotted lines or arrows in virtual space, and displaying text prompts such as "Please readjust your left hand posture to horizontal." It can also be auditory cues, such as emitting specific sound effects; or more complex guidance, such as replaying a slow-motion demonstration of the correct action in the virtual scene or generating a "shadow model" for the user to follow. The specific content and form of the motion re-recording guidance information are determined by the server's quality assessment logic.

[0040] Pose data that is marked by the server as not meeting the action quality requirements during the data acquisition process and requires the user to re-execute the corresponding action to obtain replacement data is called re-recorded pose data. Re-recorded pose data contains a pose sample at a specific point in time, which is the data to be replaced identified by the system during source filtering in the acquisition phase.

[0041] A re-recorded task segment refers to a sequence of actions within a pre-defined task that corresponds to the re-recorded posture data and requires the user to repeat. It doesn't necessarily have to be the entire pre-defined task; it could be a specific step, a holding phase of a particular posture, or a range of motion within the task. For example, in the complete task of "picking up a cup from the table," if an anomaly is detected in the posture data at the instant of "hand grasping," then the short segment corresponding to that "grabbing" action is the re-recorded task segment. The user only needs to re-execute this grasping action, rather than repeating the entire process of picking up the cup.

[0042] For example: First, the user wears multiple joint tracking devices on key joints of their body and puts on a virtual reality device. Then, they start the system and select a preset task of "delivering an item." During task execution, the joint tracking devices continuously collect the user's motion information and send the current real-time posture data to the server. After receiving the data, the server immediately performs posture calculations, generating a calculation result describing the full-body posture. The server generates driving data based on this calculation result and sends it to the virtual reality device. The virtual reality device drives the virtual human model to move in real time based on the driving data (whether it's rendering frame data from the server or joint angle data). The user sees their virtual avatar synchronously performing the "delivering an item" action on their head-mounted display. Simultaneously, the server's quality assessment module continuously analyzes the calculated posture data. For example, during the arm-extending delivery phase, if the server detects unreasonable and drastic fluctuations in the elbow joint angle of the virtual human model (potentially corresponding to joint tracker tremors or joint abnormalities in reality), it determines that the posture data for that period does not meet the preset smoothness and continuity quality requirements. The server then generates motion re-recording guidance information, such as highlighting the elbow joint in orange in the virtual view and displaying the text prompt "Please smoothly straighten your arm." This guidance information is sent to the virtual reality device along with the driving data. After seeing the prompt, the user realizes that there is a problem with the motion and readjusts their arm accordingly, re-performing a smooth motion only for the "straighten arm and deliver" re-recording task segment. The server then performs a quality assessment on this newly acquired segment of data. Only after confirming that it meets the requirements is the segment of data marked as valid data. Finally, after the entire task is completed, the server only stores all posture data marked as meeting the motion quality requirements (including the parts that passed the initial acquisition and the parts that passed the re-recording), thus ensuring that invalid information is filtered out from the source of the dataset before it is entered into the database, and the overall quality is guaranteed.

[0043] This system makes the quality of data acquisition observable through real-time visualization using virtual reality devices, allowing users to instantly identify problems in their actions. By enabling real-time quality assessment and dynamic guidance during acquisition via the server, quality control shifts from post-event inspection to in-process intervention, effectively preventing the generation and accumulation of large amounts of invalid data. By guiding re-recording of specific re-recording task segments, rather than retries of the entire dataset, the system significantly improves data acquisition efficiency and user experience. Finally, the system stores only data that meets quality requirements, enhancing the overall quality and usability of the constructed pose dataset. This provides high-value raw data for subsequent robot training or behavior analysis, fundamentally solving the problems of low data acquisition quality and high post-processing cleaning costs associated with traditional methods.

[0044] This embodiment also provides another implementation scheme for a posture data acquisition system based on virtual reality technology. This scheme focuses on utilizing virtual reality devices with computing capabilities to realize the core functions of the system, thereby constructing a lightweight, low-latency edge computing acquisition system. Specifically, Figure 2 This application provides a schematic diagram of the structure of another posture data acquisition system based on virtual reality technology, as shown in the embodiment of the present application. Figure 2 The system shown includes a joint tracking device and a virtual reality device with a communication connection; the joint tracking device is worn by the user and collects the user's current real posture data when performing preset tasks, and sends the current real posture data to the virtual reality device; the virtual reality device is used to execute a posture data acquisition method based on virtual reality technology.

[0045] like Figure 2 As shown, the system includes a joint tracking device and a virtual reality device, with a communication connection established between the joint tracking device and the virtual reality device. Under this architecture, the system no longer relies on an external, independent server for data processing, but instead pushes computing power down to the user-worn edge device, achieving a high degree of integration between data acquisition and processing.

[0046] The function of the joint tracking device in this embodiment is consistent with that in the previous embodiment, but its data transmission target has changed. The joint tracking device is worn by the user and collects the user's current real-time posture data when performing a preset task. The specific structural form, internal sensor composition (such as the inertial measurement unit), and wearing method of the joint tracking device have been described in detail in the previous embodiment and will not be repeated here. The main difference of the joint tracking device in this embodiment is that it directly sends the collected current real-time posture data to the virtual reality device, rather than sending it to an external server. This direct connection method reduces intermediate nodes in data transmission and helps to reduce transmission latency.

[0047] The virtual reality device is the core execution unit of the system in this embodiment. It not only has display functions but also integrates powerful computing capabilities. The virtual reality device integrates a high-performance processor (such as a central processing unit (CPU) and a graphics processing unit (GPU)) and a storage module, enabling it to independently run complex algorithms, perform real-time data processing, and 3D rendering.

[0048] The virtual reality device (VRD) acts as the server in the aforementioned embodiments, executing the posture data acquisition method based on VR technology. Specifically, after receiving the current real posture data from the joint tracking device, the VRD first performs localized posture calculation to generate driving data for the virtual human model. Secondly, based on the driving data, it drives the virtual human model to move in its local rendering engine and displays the virtual human model's motion state in real time. Simultaneously, the VRD also runs data quality assessment logic to determine in real time whether the current virtual posture data meets preset motion quality requirements. If the data does not meet the requirements, the VRD immediately identifies it as posture data to be re-recorded and uses its local computing power to generate corresponding motion re-recording guidance information. Finally, the VRD integrates the motion re-recording guidance information with the motion image of the virtual human model, guiding the user to make adjustments.

[0049] For example: After a user puts on a joint-tracking device and a computing-capable virtual reality (VR) device, they begin performing a preset task of "grabbing an object with one hand." During this process, the joint-tracking device collects real-time data on the user's arm joints' current posture and transmits it directly to the VR device via wireless communication (such as Bluetooth or Wi-Fi). Upon receiving the data, the VR device immediately uses its built-in processor to fuse and calculate the angles and displacements of each joint in the user's arm—the driving data. Based on this driving data, the VR device updates the arm posture of the virtual human model on its display screen in real time, allowing the user to see their virtual arm synchronously performing a grasping motion. Simultaneously, the VR device's internal quality assessment module continuously monitors the smoothness of the grasping motion's trajectory and whether the joint angles are within the normal range of human movement. If the device detects that the user's movement is too fast, causing data jitter (i.e., abrupt posture changes), it determines that the data is re-recorded posture data and immediately displays a red warning box on the virtual arm on the screen, indicating "Please slow down and smooth the movement"—a motion re-recording guidance message. After seeing the prompt, the user slows down and re-performs the grasping motion. Once the virtual reality device detects that a new action meets the quality requirements, it confirms the validity of the data segment and stores it in its local memory.

[0050] This system integrates server functionality into virtual reality devices, achieving unified acquisition, processing, evaluation, and feedback. This significantly reduces data transmission latency between devices, enabling more real-time synchronous display of virtual models and anomaly feedback, thus greatly enhancing user immersion and interactive experience. Furthermore, this architecture eliminates reliance on external high-performance servers, reducing hardware deployment costs and complexity, and improving system portability, making it applicable to data acquisition tasks in various complex scenarios such as outdoor and mobile environments. The edge computing-based implementation ensures the privacy and security of data processing; all raw posture data and evaluation results are processed locally and uploaded to the cloud for archiving only when needed, effectively protecting user privacy.

[0051] Furthermore, the method for acquiring posture data based on virtual reality technology is explained. Specifically, Figure 3 The diagram illustrates a flowchart of a posture data acquisition method based on virtual reality technology provided in an embodiment of this application. This method is executed by a posture data acquisition device. This posture data acquisition device can be a server with powerful data processing capabilities, as described in the previous embodiments, i.e., ... Figure 1 The server shown can also be a virtual reality device that integrates computing functions in the aforementioned embodiments, i.e., as... Figure 2 The specific form of the posture data acquisition device in the virtual reality device shown depends on the deployment choice of the system architecture. For example... Figure 3 As shown, the method includes the following steps S110~S170: S110 determines the user's current actual posture data when performing a preset task based on the joint tracking device worn by the user.

[0052] The posture data acquisition device receives raw sensor data in real time from joint tracking devices at various key joints of the user's body via wireless or wired communication interfaces. This data includes information such as the angular velocity, acceleration, and magnetic field strength of each joint.

[0053] After receiving the raw data, the attitude data acquisition device first performs preliminary data cleaning and timestamp alignment to eliminate time errors between different tracking devices. Then, using a pre-set attitude calculation algorithm (such as an inertial navigation algorithm based on complementary filtering or Kalman filtering), the device fuses and calculates the three-dimensional attitude angles of each joint relative to its own coordinate system. Furthermore, it uses a human skeletal model to transform these local joint angles into a unified attitude representation of the entire body. The results of this series of calculations constitute the current true attitude data representing the user's motion state at the current moment.

[0054] A preset task refers to a specific sequence of actions set for data acquisition, such as "carrying a heavy object," "waving," or "obstacle walking." Users need to perform the corresponding limb movements according to the requirements of the task. For example, during the process of a user performing the preset task of "waving," the posture data acquisition device continuously calculates the joint angle and position of the user's arm at every moment of the arm's movement, thereby obtaining a continuous stream of current real posture data.

[0055] S120, mapping the current real posture data to the current virtual posture data of the virtual human body model in the virtual reality device, wherein the virtual reality device is worn on the user's head and is used to display the virtual human body model to the user.

[0056] Virtual reality devices are worn on a user's head and used to display virtual human models to the user.

[0057] In S120, the posture data acquisition device performs a coordinate transformation from physical space to virtual space. Based on the joint angles, bone length ratios, and anthropometry parameters contained in the current real-world posture data, the device calculates the rotation matrix and translation vector of the corresponding skeletal nodes in the virtual human body model. This process needs to consider the perspective projection parameters and field of view of the virtual reality device to ensure that the mapped model visually conforms to the perspective rules observed by the human eye. Through this mapping, a physical action in the real world (e.g., raising an arm 30 degrees) is accurately converted into a mathematical description in virtual space (i.e., a virtual skeletal node rotating 30 degrees), which is the current virtual posture data. This data can be directly used to drive the virtual human body model, ensuring that the model's posture in the virtual environment maintains a highly consistent synchronization with the user's real posture.

[0058] S130 drives the virtual human body model in the virtual reality device to follow the user's movements based on the current virtual posture data.

[0059] The posture data acquisition device transmits the generated current virtual posture data to the graphics rendering engine of the virtual reality device. Based on the received data, the rendering engine updates the state of each bone in the virtual human model in real time and recalculates the position of the model's vertices, thereby generating dynamically changing images on the virtual reality device's display screen.

[0060] Because the data is typically updated at a rate of 60 frames per second or higher, the virtual human models that users see in virtual reality devices can move with extremely low latency, following their own real movements and creating a strong sense of immersion.

[0061] This real-time feedback not only allows users to visually see their own actions, but also provides an intuitive visual benchmark for subsequent quality assessment. For example, when a user raises their right hand, the right hand of the virtual human model on the virtual reality device screen will also rise synchronously, with the movement trajectories completely overlapping.

[0062] S140, determine whether the current virtual posture data meets the preset motion quality requirements.

[0063] If the current virtual pose data does not meet the motion quality requirements, then jump to S150 and execute S150~S170.

[0064] If the current virtual posture data meets the motion quality requirements, then jump to S110 and execute S110~S170 to continuously collect posture data.

[0065] Preset motion quality requirements refer to a series of judgment criteria pre-set according to the purpose of data acquisition, robot kinematic constraints, or task logic requirements. These criteria may specifically include: data continuity (whether the amplitude of posture changes between adjacent frames is within a reasonable threshold and there are no jumps), posture rationality (whether the joint angle exceeds the physiological limits of the human body or robot, such as the knee cannot bend backward), motion smoothness (whether the acceleration is too large to avoid data jitter), and consistency with task logic (for example, in the "grasping" task, whether the hand posture closes to contact the object).

[0066] The attitude data acquisition device uses a built-in quality assessment module to perform algorithmic detection on the current virtual attitude data in real time. For example, it calculates the Euclidean distance or angle difference between two adjacent frames of attitude data. If the difference exceeds a preset abrupt change threshold, it is determined to be discontinuous; or it checks whether the joint angles exceed a preset range of [-180 degrees, 180 degrees]. This judgment process is performed in real time during data acquisition, rather than through offline analysis afterward.

[0067] For data that does not meet the requirements, an error correction process is immediately initiated; for data that meets the requirements, it is considered valid and high-quality data, and continues into the next round of data collection, thus ensuring a continuous acquisition of valid data. This real-time filtering mechanism avoids the accumulation of invalid data and ensures that the data flow throughout the entire collection process remains at a high quality level.

[0068] S150, the current virtual posture data is determined as the re-recorded posture data, and action re-recording guidance information corresponding to the re-recorded posture data is generated.

[0069] When the attitude data acquisition device detects an anomaly in step S140, it immediately marks the current attitude data as re-recorded attitude data. The re-recorded attitude data refers to data points that are determined to be substandard and need to be discarded and re-acquired.

[0070] The posture data acquisition device records the data point and its timestamp in the error log, but does not immediately store it in the final qualified database. Subsequently, the guidance generation logic is initiated to generate instruction content to help the user correct errors, resulting in motion re-recording guidance information. The generation process of this guidance information relies on the analysis of the error type: if a joint angle exceeds the limit, the guidance information may include "Please reduce the arm swing amplitude"; if a data jump is detected (possibly caused by sensor signal loss), the guidance information may include "Please check the tracker and recalibrate"; if a movement deviates from the task logic (e.g., a grasping action was not performed when it should have), the guidance information may include "Please try to grasp the action." This guidance information is encapsulated as text prompts, voice commands, or virtual visual markers (e.g., drawing the correct motion trajectory in virtual space) and prepared for sending to the user.

[0071] S160: Based on the virtual reality device, display motion re-recording guidance information to the user to guide the user to re-execute the re-recording task segment corresponding to the re-recording posture data in the preset task.

[0072] The posture data acquisition device sends the generated motion re-recording guidance information to the virtual reality device for display.

[0073] When virtual reality devices display to users, they can use a variety of augmented reality or virtual reality visual presentation methods, such as popping up a text prompt box in the center of the field of vision, highlighting virtual joints with abnormal postures with a specific color (such as red), or generating a "ghost model" in the virtual scene to demonstrate standard movements for users to imitate.

[0074] Once users see these instructions in the virtual reality device, they can clearly understand what problems existed in their previous actions. For example, in a preset task of "walking from point A to point B and opening the door," if an abnormal posture occurred only in the "opening the door" segment, the user only needs to be guided to repeat the "opening the door" action segment, without having to walk from point A again. This targeted re-recording mechanism greatly reduces user fatigue and improves the efficiency and accuracy of data collection.

[0075] After the user re-executes the action segment according to the instructions, the process will re-enter S110 to S140 to re-evaluate the newly collected data until the data meets the requirements.

[0076] S170 stores the real posture data corresponding to the virtual posture data that meets the action quality requirements during the user's execution of a preset task, in order to complete the posture data acquisition.

[0077] The posture data acquisition device only stores the current virtual posture data that is determined to meet the motion quality requirements in step S140. It is worth noting that the stored object is the real posture data corresponding to the current virtual posture data, that is, the most original and most valuable human joint motion data after calculation and calibration.

[0078] The storage medium can be the device's local storage unit or a cloud database. To ensure data traceability and availability, the actual attitude data can be encapsulated and formatted together with corresponding metadata such as timestamps, task tags, and user identifiers for storage.

[0079] This source-level filtering method ensures that the final dataset does not contain any invalid or low-quality samples, thus guaranteeing the overall purity and usability of the dataset. The attitude data acquisition task is considered complete once all steps of the pre-defined task have been correctly executed and the corresponding data have passed quality checks and been stored.

[0080] During data acquisition, when the posture data acquisition device detects a quality problem in a certain segment of motion, simply prompting the user to repeat it is often insufficiently accurate, as the user may not be clear which part of the motion trajectory has deviated. Therefore, a playback and pinpoint prompting mechanism based on re-recorded task segments can be further introduced. By reproducing the previous motion process in the virtual reality device and accurately presenting prompt information at the specific posture node where the problem occurred, the user can intuitively locate the source of the error and thus make targeted corrections. Preferably, the re-recorded task segments correspond to multiple virtual posture data; after S160, the method further includes S161~S162: S161, Generate prompt information corresponding to the re-recorded posture data, wherein the prompt information is used to indicate that the current posture of the virtual human body model is an unqualified posture.

[0081] In this step, the attitude data acquisition device generates specific prompt information based on the re-recorded attitude data determined in step S150.

[0082] The aforementioned prompt information refers to the feedback signal generated by the posture data acquisition device to convey data quality issues to the user. Its forms include, but are not limited to, visual markers (such as highlighting, color changes, and arrow indicators), text labels (such as "joint abnormality" and "posture out of limit"), or auditory prompts (such as specific beeping sounds).

[0083] The unqualified posture refers to the specific posture state that is determined in step S140 to not meet the preset action quality requirements.

[0084] The generation of prompts typically involves associating abnormal features (such as joint angles outside the range or discontinuous motion trajectories) in the re-recorded posture data with pre-designed visual or auditory resources. For example, if a joint angle is detected to be outside the normal physiological range of human activity, a text prompt "abnormal joint angle" can be generated, and the corresponding color parameter can be set to a red warning color.

[0085] S162, based on the multiple virtual posture data corresponding to the re-recorded task segment, drive the virtual human body model in the virtual reality device to move again, and when the virtual human body model in the virtual reality device moves to the posture corresponding to the re-recorded posture data, display a prompt message in the virtual reality device.

[0086] The posture data acquisition device retrieves a series of virtual posture data (i.e., previously collected action sequences containing errors) that belong to the re-recording task segment, and uses this data to drive the virtual human model to reproduce the previous movement trajectory in virtual space. This process is similar to video playback, allowing the user to see how they just performed the action.

[0087] During playback, the virtual human model's posture continuously changes. When its posture at a certain moment matches the specific posture marked as re-recorded posture data in step S150 (e.g., through timestamp matching or posture parameter matching), the posture data acquisition device determines that the error has occurred. At this time, the posture data acquisition device sends the prompt information generated in step S161 to the virtual reality device for presentation. The presentation method can be a dynamic effect superimposed on the virtual human model. For example, when the model moves into an incorrect twisting posture, the specific joint flashes red light momentarily, and a "Posture Error" label pops up next to it, to prompt the user at this instant that the posture has a problem.

[0088] By replaying multiple virtual posture data corresponding to the re-recording task segments, users can objectively examine the details of their previous actions. By presenting prompts at specific moments when the movement reaches an unqualified posture, the spatiotemporal precision of the problem point is achieved, avoiding user ambiguity or misunderstanding of the error occurrence point, greatly reducing the difficulty of error correction and cognitive load. Furthermore, this intuitive and precise feedback mechanism effectively shortens the trial and error time required for users to adjust their movements, improves the success rate of single re-recording, and further enhances the efficiency and quality of the entire posture data acquisition.

[0089] In real-world data acquisition scenarios, users often need to repeatedly execute the same action or task segment multiple times to obtain sufficient training samples. Simply retaining all data generated from repeated executions during storage would not only consume a large amount of storage space but also increase the difficulty of subsequent data cleaning. Therefore, an intelligent storage strategy based on filtering repetitive task segments can automatically select the optimal data segments for storage by quantifying the data quality of each execution, thereby ensuring high quality and low redundancy in the final dataset. Preferably, S170 includes sub-steps S171~S173: S171, identify the task segments that are repeatedly executed during the user's execution of a preset task, and obtain at least one group of repeated task segments, wherein the group of repeated task segments includes task segments that are repeatedly executed multiple times, and each executed task segment corresponds to multiple virtual pose data.

[0090] The attitude data acquisition device analyzes the acquired complete task flow to identify specific action units that are repeatedly performed by the user.

[0091] The task segment refers to a sub-action within a preset task that has a clear start and end boundary, such as "swinging an arm once," "taking a step," or "grabbing an object once." The repetitive task segment group refers to a data set formed by multiple consecutive or discontinuous executions of the same task segment. By analyzing the action characteristics on the time series (e.g., periodic zeroing of actions, repetition of key posture states) or according to the preset task script definition, multiple executed action segments are categorized into the same group. For example, if the preset task is "squat training," and the user may have performed 10 squats consecutively, then the data from these 10 squats is identified as a repetitive task segment group containing 10 repetitively executed task segments. Each task segment contains corresponding, continuously changing virtual posture data over time, which records the complete process of that action.

[0092] S172, the number of re-recorded attitude data in the virtual attitude data corresponding to each executed task segment in the repetitive task segment group is determined as the re-recording amount of the task segment executed in that time.

[0093] The attitude data acquisition device scores the quality of each individual execution within a group of repetitive task segments.

[0094] The re-recording quantity is a quantitative indicator used to measure the degree of data defects during the execution of a task segment. The specific calculation method is as follows: the attitude data acquisition device retrieves all virtual attitude data generated during the execution process and counts the number of data points marked as re-recorded attitude data. The definition of re-recorded attitude data (i.e., data that does not meet the motion quality requirements) has been explained in steps S140 and S150 and will not be repeated here.

[0095] A higher re-recording rate indicates more abnormal postures, joint over-limits, or data jumps during the execution, resulting in poorer movement quality. Conversely, a lower re-recording rate, or even zero, indicates a smooth execution that fully meets the preset movement quality requirements. For example, in the squat training example above, the first squat might result in 3 re-recorded posture data points due to instability, thus its re-recording rate is 3; the second squat, however, is performed perfectly without triggering any re-recording markers, resulting in a re-recording rate of 0.

[0096] S173, store the real pose data corresponding to the task segment with the least amount of re-recording in the repetitive task segment group, as well as the real pose data corresponding to the task segments that were not repeatedly executed during the user's execution of the preset task, in order to complete the pose data acquisition.

[0097] For the identified repetitive task segments, the attitude data acquisition device compares the number of re-recorded tasks in each segment within the group and selects the segment with the fewest re-recorded tasks (i.e. the one with the best quality). Only the actual attitude data corresponding to this segment is stored in the final database, while other segments with high re-recorded tasks within the group are discarded.

[0098] For task segments that are not executed repeatedly, that is, action units that appear only once in the entire preset task (such as the transition from the ready position to the first squat), the corresponding real posture data is stored according to the normal logic (provided that the data itself is not marked as invalid as a whole).

[0099] With this strategy, the attitude data acquisition device automatically completes the process of removing false data and selecting the best from multiple versions of data at the same time as the data acquisition is completed, and directly outputs the highest quality attitude dataset.

[0100] For example, the user's preset task is "continuous obstacle crossing," which includes four stages: "approach run," "jump," "landing," and "reset." The user repeats the "jump-landing" action cycle three times. In step S171, the posture data acquisition device identifies that the first to third cycles belong to the same repetitive task segment group. In step S172, the posture data acquisition device counts the re-recording amount of these three cycles: the lower leg angle is abnormal during the first jump, with a re-recording amount of 2; the posture is unstable during the second landing, with a re-recording amount of 5; the third action is perfect, with a re-recording amount of 0. In step S173, the posture data acquisition device determines that the re-recording amount of the third cycle is the least (0), so only the actual posture data corresponding to the third "jump-landing" cycle is stored in the database, while the data of the non-repeated "approach run" and "reset" stages are also stored. The data of the other two flawed cycles are not saved in the final high-quality dataset.

[0101] By introducing re-recording volume as an evaluation metric, the posture data acquisition device can objectively and accurately quantify the quality differences of each repeated action, avoiding the subjectivity and inefficiency of manual screening. By storing only the task segment data with the least re-recording volume, the posture data acquisition device completes the automatic screening of multiple versions of data at the data source, greatly improving the purity and consistency of the final data stored, ensuring that the data used for robot training or behavior modeling are the optimal samples. Furthermore, this intelligent storage mechanism effectively eliminates redundant low-quality duplicate data, saves storage space, and significantly reduces the workload of subsequent data cleaning and preprocessing, further improving the overall efficiency of posture data acquisition.

[0102] In complex posture data acquisition scenarios, the quality judgment logic in step S140, if relying solely on rigid rule-based judgment, may lead to misjudgments, such as misclassifying certain special but reasonable actions as unqualified, or failing to identify some complex non-regular errors. Therefore, the inventors of this application propose a hierarchical judgment mechanism that integrates preset rules and artificial intelligence models. This mechanism first uses low-cost, low-latency rules for rapid screening. For edge cases where the rules cannot determine the outcome, a pre-trained model with advanced semantic understanding capabilities is introduced for secondary judgment, thereby significantly improving the accuracy and robustness of quality assessment while ensuring data acquisition efficiency. Specifically, S140 includes sub-steps S141~S143: S141, perform a preliminary judgment on the current virtual posture data according to the preset motion rationality rules, and obtain a preliminary judgment result; if the preliminary judgment result is that the current virtual posture is reasonable, then execute S142; if the preliminary judgment result is that the current virtual posture is unreasonable, then execute S143.

[0103] The attitude data acquisition device first calls the preset motion rationality rules in the attitude data acquisition device to perform a first round of rapid screening of the input current virtual attitude data.

[0104] The preset motion rationality rules refer to a set of hard constraints and thresholds predefined based on the principles of human kinematics, biomechanics, or robot kinematics. These rules can be set to include restrictions on joint angles (e.g., the knee joint bending angle does not exceed 0 degrees), restrictions on joint angular velocity (e.g., the instantaneous angular velocity does not exceed X degrees per second to prevent jumps caused by sensor noise), constraints on bone length (to prevent bone stretching and contraction caused by calculation errors), and constraints on posture continuity (the change in posture between adjacent frames does not exceed a threshold).

[0105] The specific meaning and composition of the current virtual attitude data have been explained in detail in step S120, and will not be repeated here.

[0106] The posture data acquisition device compares each parameter in the current virtual posture data with the thresholds in the aforementioned rules to make a preliminary judgment. If all parameters are within the allowed range of the rules, the preliminary judgment result is reasonable; if any parameter violates the rule constraints, the preliminary judgment result is unreasonable. For example, if a user's arm joint angle is detected to have a sudden 180-degree reversal, this is physiologically impossible, so the rule judgment result is directly unreasonable.

[0107] S142, Determine that the current virtual posture data meets the motion quality requirements.

[0108] When the preset rules in step S141 determine that the current posture is reasonable, it means that the posture data fully meets the acquisition standards in terms of basic aspects such as geometry, range of motion, and continuity, and there are no obvious hard errors. Since the calculation amount for rule determination is extremely small and the response speed is fast, the posture data acquisition device can directly accept the determination result without performing subsequent complex calculations. At this time, the posture data acquisition device directly marks the current virtual posture data as meeting the motion quality requirements and allows it to enter the subsequent storage process (such as storage in step S170). This design ensures that the vast majority of normal and standardized motion data can pass the quality check in real time and efficiently, guaranteeing the smoothness of the acquisition process.

[0109] S143, input the current virtual pose data into the pre-trained visual language model to obtain the model's judgment result. If the model's judgment result is that the current virtual pose is reasonable, then the current virtual pose data is determined to meet the action quality requirements. If the model's judgment result is that the current virtual pose is unreasonable, then the current virtual pose data is determined to not meet the action quality requirements.

[0110] Visual language models are large-scale artificial intelligence models trained using deep learning techniques, capable of simultaneously understanding both image (visual) and text (linguistic) information. Typical visual language models include CLIP, GPT-4V (Generative Pre-trained Transformer 4 with Vision), or LLaVA. These models are trained on large amounts of image-text pairs, enabling them to understand the semantic content of images and perform reasoning and judgment based on natural language instructions.

[0111] The initial input is the current virtual pose data (usually joint coordinates or angle data). However, to adapt to the input requirements of the visual language model, the pose data acquisition device typically renders this pose data in real-time in the background as a pose image or keyframe screenshot of a virtual human model, or extracts the corresponding visual feature map. Subsequently, the pose data acquisition device inputs this image data along with specific prompts into a pre-trained visual language model. The prompts guide the model in making judgments, such as: "Please determine whether the pose of the person in the image is natural and reasonable? Are there any obvious joint distortions or abnormalities?" Next, the model performs the inference process and outputs the model's judgment result. Visual language models, leveraging their powerful generalization and semantic understanding capabilities, can identify complex scenarios that rules struggle to cover. For example, a particular yoga pose, even if its joint angles are close to the limits set by the rules, might be judged as reasonable by the visual language model through observing overall limb coordination and context. Conversely, some movements that seem to conform to angle rules might be judged as unreasonable if they visually present unnatural distortions or violate ergonomics.

[0112] Finally, the posture data acquisition device makes a final decision based on the model's judgment result: if the model's judgment result is reasonable, it means that although the posture may violate some preset rule boundaries (such as being at an extreme angle), it is still a valid and realistic human posture at a higher-dimensional semantic and visual level. To avoid accidental deletion due to overly rigid rules, the posture data acquisition device adopts the model's judgment and determines that the current virtual posture data meets the action quality requirements; if the model's judgment result is unreasonable, it means that the posture not only has problems at the rule level, but is also confirmed as incorrect or abnormal at the visual and semantic levels. Based on this, the posture data acquisition device determines that the current virtual posture data does not meet the action quality requirements and marks it as re-recorded posture data, triggering the subsequent re-recording guidance process (such as step S150).

[0113] For example: A user is collecting posture data containing a set of high-difficulty martial arts movements. When performing a "back grappling" movement, the arm rotates at a large angle. The preset rules in step S141 detect that the shoulder joint rotation angle exceeds a preset safety threshold (e.g., 120 degrees), and the judgment result is unreasonable. The posture data acquisition device does not directly reject the data, but proceeds to step S143. Based on the current virtual posture data, the posture data acquisition device quickly generates a screenshot of the virtual human body model's posture at this moment and inputs it into a pre-trained visual language model, asking "Is this movement normal?". After analyzing the image, the visual language model finds that this is a common anti-joint technique in martial arts, and the overall posture is coherent and conforms to the logic of human movement. Therefore, it outputs a reasonable judgment result. Based on the intelligent judgment of the model, the posture data acquisition device finally determines that the posture data meets the movement quality requirements, avoiding the loss of high-quality movement samples due to rigid rules.

[0114] By employing a cascaded decision architecture of pre-defined rules and pre-trained models, the speed and accuracy of data quality assessment are effectively balanced. Rule-based decision-making ensures real-time performance and the interception of obvious errors, while the visual language model provides semantic-level understanding of complex and edge cases. Furthermore, this dual-verification mechanism significantly reduces the false positive rate, preventing the incorrect rejection of special but valid actions due to overly strict rules, thereby improving the richness and diversity of the collected posture data. Moreover, by utilizing the pre-trained visual language model, posture data acquisition devices do not need to manually write complex decision algorithms for each specific action, possessing strong generalization and adaptability, and can support a wider range of more challenging posture data acquisition tasks.

[0115] The specific implementation of the preset motion rationality rules in the aforementioned step S141 is explained. Optionally, a dual constraint mechanism based on geometric kinematics, namely joint movement rules and joint distance rules, can be introduced to accurately verify the rationality of the current virtual posture data. By quantifying the rate of change of joints in the time dimension and the geometric relationship in the spatial dimension, posture abrupt changes and deformation data caused by sensor noise, solution drift or signal loss can be effectively eliminated, ensuring that the collected posture data conforms to the basic physical laws of human kinematics.

[0116] Specifically, the motion rationality rules include joint movement rules and joint distance rules; S141 includes sub-steps S141a~S141c: S141a, determine the degree of joint movement in the current virtual posture based on the key joint data in the previous virtual posture data and the key joint data in the current virtual posture data.

[0117] This step focuses primarily on the changes in attitude data over time.

[0118] The attitude data acquisition device retrieves the virtual attitude data from the previous sampling time (i.e., time T-1) as the previous virtual attitude data and compares it with the current virtual attitude data at the current time (i.e., time T).

[0119] Key joint data refers to data points extracted from whole-body posture data to characterize the state of major joints in human movement, such as the spatial coordinates or rotation angles of elbows, wrists, and knees. The device calculates the spatial displacement or angular change of the same key joint between two moments to determine the degree of joint movement. This degree of movement reflects the instantaneous velocity or acceleration of the joint movement and is used to determine whether the movement is too violent or has abnormal jumps.

[0120] S141b, determine the joint distance between multiple key joints in the current virtual posture based on the key joint data in the current virtual posture data.

[0121] This step primarily focuses on the geometric characteristics of the posture data in the spatial dimension. Based on the current virtual posture data, the posture data acquisition device selects two or more key joints and calculates the Euclidean distance or a specific geometric metric between them to obtain the joint distance. This distance reflects the spatial topological relationships of the human skeletal structure, such as upper arm length, shoulder width, or the relative distance between the hands and feet. During human movement, although joint angles constantly change, the physical length of the bones usually remains constant, and specific distances between limbs are often within a reasonable physiological range. Therefore, by monitoring these distances, abnormal phenomena such as bone stretching or limb penetration caused by calculation errors can be effectively identified.

[0122] S141c, if the degree of joint movement satisfies the joint movement rule, and the joint distance between multiple key joints satisfies the joint distance rule, then the current virtual posture is determined to be reasonable.

[0123] The attitude data acquisition device compares the degree of joint movement calculated in S141a with the joint movement rules, and at the same time compares the joint distance calculated in S141b with the joint distance rules.

[0124] Specifically, the joint movement rule includes ensuring that the degree of joint movement does not exceed a preset joint movement limit. This means that if the displacement or angle change of a critical joint between two frames exceeds the maximum threshold set by the attitude data acquisition device (e.g., a single-frame movement exceeding 1 meter), the attitude is determined to have undergone an unreasonable instantaneous shift and is therefore deemed unreasonable.

[0125] The joint distance rule has two implementation methods.

[0126] The first method is static threshold determination, which means that the joint distances between multiple key joints are all within a preset joint distance range. For example, the distance between a person's left and right shoulders is set to be between 0.3 meters and 0.6 meters. If the calculated distance exceeds this range, it is determined to be abnormal.

[0127] The second method is dynamic judgment based on proportion, meaning that the current distance proportion falls within a preset reasonable proportion range corresponding to the distance proportion item. This judgment method can adapt to users of different body types (such as children or adults), and evaluates the reasonableness of the posture through relative proportion rather than absolute distance.

[0128] To achieve proportion-based determination, the attitude data acquisition device executes steps S144a~S144d: S144a, for a preset distance ratio term, based on the current virtual posture data, determine the first distance of the first joint set required to constitute the distance ratio term, and the second distance of the second joint set, wherein the first distance is a geometric metric calculated based on the spatial coordinates of the multiple joints included in the first joint set, and the second distance is a geometric metric calculated based on the spatial coordinates of the multiple joints included in the second joint set.

[0129] Here, the attitude data acquisition device extracts two sets of key joints from the current virtual attitude data according to the definition of the preset distance ratio.

[0130] The distance ratio term defines two geometric quantities used to calculate the ratio, such as the ratio of "hand width" to "shoulder width," or the ratio of "upper limb length" to "torso length." The posture data acquisition device acquires a first set of joints (e.g., left and right fingertips) and a second set of joints (e.g., left and right acromion). Then, based on the spatial coordinates of the joints in the first set, the posture data acquisition device calculates the geometric metric (e.g., Euclidean distance) between them to obtain the first distance; similarly, it calculates the second distance based on the coordinates of the second set of joints.

[0131] S144b, the ratio of the first distance to the second distance is determined as the current distance ratio of the distance ratio term.

[0132] The posture data acquisition device performs a division operation, dividing the first distance by the second distance to obtain a dimensionless ratio, i.e., the current distance ratio. This ratio can reflect the geometric proportions of human posture. For example, when a person's arms are raised laterally, the ratio of the distance between the fingertips to the distance between the acromions is usually greater than 1.

[0133] S144c, determine whether the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio item; if the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio item, then execute S144d.

[0134] If the current distance ratio is outside the preset reasonable ratio range corresponding to the distance ratio item, then it is determined that the joint distance between multiple key joints does not meet the joint distance rule.

[0135] The posture data acquisition device compares the calculated current distance ratio with a pre-defined reasonable range (e.g., [1.5, 3.0]) for that distance ratio. If the ratio is within the range, it means that the posture conforms to human geometric proportions and is judged as reasonable; conversely, if the ratio exceeds the range (e.g., the hand is suddenly calculated to an extremely far position, causing the ratio to increase abnormally), it is judged as unreasonable and does not meet the joint distance rules.

[0136] S144d, determine that the joint distances between multiple key joints satisfy the joint distance rule; wherein, the distance ratio term defines the first set of joints and the second set of joints on which it is based, and the relationship between the first set of joints and the second set of joints includes at least one of the following: the adjacency relationship of ipsilateral limbs, the symmetry relationship about the body midline, or the geometric relationship that together form a specific functional triangle.

[0137] To enable joint distance rules to adapt to different human body shapes and complex movement scenarios, the distance ratio term defines a specific topological relationship between the first and second joint sets upon which it is based. This relationship is built upon a deep understanding of human anatomy and sports biomechanics and is primarily used to quantify the geometric consistency between joints.

[0138] Specifically, the relationship between the first joint set and the second joint set includes at least the adjacency relationship of ipsilateral limbs, the symmetry relationship about the body midline, or the geometric relationship of jointly forming a specific functional triangle. The following will explain these three relationships in detail with specific application scenarios.

[0139] First, the adjacency relationship of ipsilateral limbs focuses on the proportion between adjacent segments on the same limb link, mainly used to assess the coordination of the limb itself or the rationality of a specific posture. This relationship determines whether the movement conforms to the laws of human kinematics by comparing the projection or actual length ratio of adjacent bones in the same limb (such as thigh and lower leg, upper arm and forearm) in the current posture. For example, in the scenario of assessing the degree of arm flexion, the distance ratio term is set as the ratio of the upper arm length to the forearm length. At this time, the first joint set includes the left shoulder and left elbow, used to calculate the first distance D1 (i.e., the geometric length of the upper arm in the current posture); the second joint set includes the left elbow and left wrist, used to calculate the second distance D2 (i.e., the geometric length of the forearm in the current posture). The distance ratio term R1 is the ratio of D1 to D2. This ratio term can be used to determine whether the arm remains in an extended state during movement (R1 deviates little from the preset static model ratio) or has undergone severe flexion (D2 is significantly shortened due to joint folding, causing R1 to increase). If R1 exceeds the reasonable range, the posture data acquisition device can determine that the arm posture is abnormal. For example, in assessing the degree of thigh and calf folding during a squat, the first joint set includes the left hip and left knee, calculating the first distance D3 (the projected distance of the thigh length in the current three-dimensional space); the second joint set includes the left knee and left ankle, calculating the second distance D4 (the projected distance of the calf length in the current three-dimensional space). The distance ratio R2 is the ratio of D3 to D4. During the squat, although both D3 and D4 shorten with changes in joint angle, the change pattern of their ratio R2 has specific physiological characteristics. By monitoring whether R2 is within a preset reasonable ratio range, the posture data acquisition device can accurately determine whether the squat depth is up to standard and whether there are incorrect postures such as excessive forward tilting of the knee joint. This judgment method based on the ratio of adjacent limb segments can effectively filter out abnormal limb extension and contraction caused by calculation errors, ensuring the local continuity of the movement.

[0140] Secondly, regarding the symmetry of the body's midline, the focus is on the distance relationship between corresponding parts on the left and right sides of the body. This is mainly used to assess postural balance and whether there is lateral tilting or trunk twisting. In normal human movement, the postures on the left and right sides are usually highly symmetrical. If the proportions are unbalanced, it often indicates movement disorder or data acquisition errors. For example, in assessing shoulder level and trunk stability, the first joint set includes the left and right shoulders, calculating the first distance D5 (i.e., shoulder width); the second joint set includes the left and right hips, calculating the second distance D6 (i.e., hip width). The distance ratio term R3 is the ratio of D5 to D6. In an upright or normal movement state, the ratio of shoulder width to hip width R3 should remain approximately constant, within a small, preset reasonable ratio range. If the posture data acquisition device detects a significant change in R3 during movement, such as a sudden change in value, this usually indicates lateral bending or twisting of the trunk, or data drift from one side of the joint tracker. For example, in assessing arm swing symmetry during walking or running, the first joint set includes the left shoulder and left wrist, calculating the first distance D7 (the extension length of the left arm in space); the second joint set includes the right shoulder and right wrist, calculating the second distance D8 (the extension length of the right arm in space). The distance ratio R4 is the ratio of D7 to D8. In periodic gait movements, the amplitude of the left and right arm swings should be basically consistent, therefore the ratio R4 should fluctuate regularly around the value 1. If R4 deviates significantly from the value 1 for a long time, or if the fluctuation pattern is chaotic, it indicates to the user that the left and right arm swings are asymmetrical, the movements are uncoordinated, or that there is missing or abnormal hand tracking data on one side. By verifying the symmetry relationship, the posture data acquisition device can effectively identify systemic posture imbalance problems.

[0141] Finally, the geometric relationship that constitutes a specific functional triangle refers to using the side length ratio of the geometric shape (usually a triangle) formed by multiple joints in space to precisely quantify compound angles or body orientation. Compared to simple two-point distances, triangular relationships can capture more complex spatial geometric features. For example, in assessing the forward or backward tilt angle of the body in the sagittal plane, the joints constituting the triangle include the left shoulder, right shoulder, and hip center (this point is usually calculated from the midpoint of the spatial coordinates of the left and right hip joints). In this case, the first joint set includes the left and right shoulders, and the first distance D9 (i.e., the base of the triangle, corresponding to shoulder width) is calculated; the second joint set includes the left shoulder and hip center, and the second distance D10 (i.e., one side of the triangle) is calculated. The distance ratio term R5 is the ratio of D9 to D10. When the body is in an upright position, R5 is a specific baseline value; when the body tilts forward, the vertical projection position of the hip center relative to the shoulder changes, causing the vertical component of the side D10 to decrease, thus increasing the ratio R5. The ratio R5 has a clear trigonometric function correspondence with the body's forward tilt angle, allowing the posture data acquisition device to accurately infer whether the torso tilt angle exceeds the standard. For example, in the complex scenario of evaluating the angle between the upper limb and torso during a shooting or pushing motion, the joints forming the triangle include the left shoulder, left elbow, and left wrist. The first joint set includes the left shoulder and left elbow, yielding a first distance D11; the second joint set includes the left elbow and left wrist, yielding a second distance D12. Although the ratio of D11 to D12 alone cannot directly determine the specific elbow joint angle, by introducing a third distance—the distance D13 from the left shoulder to the left wrist—and combining it with the triangle cosine theorem, the posture data acquisition device can accurately calculate the actual elbow joint angle. This example demonstrates the broad meaning of geometric measurement; it is not limited to simple distance ratios but can also include angles or other derived geometric quantities based on distance. This functional triangle-based determination method enables posture data acquisition devices to perform high-precision quality assessments of complex composite postures, making it suitable for challenging motion data acquisition scenarios.

[0142] In summary, this embodiment constructs multi-dimensional joint distance rules by defining geometric relationships such as same-side adjacency, midline symmetry, or functional triangles between the first joint set and the second joint set. These rules utilize the inherent kinematic constraints of the human body and can efficiently and universally judge the rationality and accuracy of posture data without setting complex numerical thresholds for each specific action. This significantly improves the robustness of posture data acquisition equipment and the accuracy of data quality control.

[0143] For example, a user is collecting posture data with both arms extended laterally. At a certain moment, due to strong electromagnetic interference, the tracker worn on the left arm causes a momentary jump in the calculated joint coordinates of the left arm. In step S141a, the posture data acquisition device detects that the movement of the left wrist joint reaches 50 cm in a single frame, far exceeding the preset joint movement threshold (e.g., 10 cm), triggering an alarm for the joint movement rule. Although a distance ratio check may be initiated at this point, assuming the movement rule is not triggered, the posture data acquisition device continues to execute steps S141b and S144a. The posture data acquisition device calculates the ratio (i.e., the distance ratio term) between the distance between the wrists (first distance) and the distance between the shoulders (second distance). Under normal circumstances, the ratio should be slightly greater than 2 during lateral raise. However, due to the jump of the left wrist coordinate, the first distance increased abnormally, and the calculated current distance ratio reached 5.0. Step S144c judged that the ratio exceeded the preset reasonable ratio range (such as 1.5 to 3.0), so it was determined that the joint distance did not meet the rules. Combining the results of S141a and S141c, the posture data acquisition device determined that the current virtual posture data was unreasonable, and thus marked it as re-recorded posture data to avoid storing this frame of erroneous posture data in the database.

[0144] By combining joint movement rules and joint distance rules, the posture data acquisition device performs dual verification of posture data from both temporal and spatial dimensions, greatly improving the comprehensiveness and accuracy of data quality assessment. By introducing distance ratio determination based on joint set relationships (such as symmetry and adjacency), the posture data acquisition device can utilize the inherent geometric topological features of the human body to identify errors. This method is more robust than simply relying on absolute coordinate thresholds and can adapt to users of different body types. Furthermore, by using these efficient rules to remove obvious noise and abnormal data in the preliminary judgment stage (S141), the posture data acquisition device reduces its dependence on expensive computing resources (such as visual language models) and improves the real-time performance and overall efficiency of posture data acquisition.

[0145] This application embodiment determines the user's current real posture data when performing a task and maps this data in real time to the current virtual posture data of the virtual human model in the virtual reality device. This achieves synchronous visualization of real actions in the virtual space of the virtual reality device, making data quality observable. Furthermore, while driving the virtual model to follow the user's movement, it determines whether the current virtual posture data meets preset action quality requirements, transforming quality assessment from post-event analysis to real-time monitoring. This instantly identifies joint abnormalities, sudden posture changes, and other issues. If the data does not meet the requirements, it is immediately identified as re-recorded posture data, and corresponding action re-recording guidance information is generated. The virtual reality device guides the user to re-execute only the re-recorded task segment corresponding to the re-recorded data, replacing the inefficient method of relying on human experience or overall retries. This directly corrects erroneous actions during the acquisition process. Finally, only the real posture data corresponding to the virtual posture data that meets the action quality requirements is stored, completing source filtering before data is entered into the database. This ensures that each frame of data collected meets the preset robot capability range and task logic requirements, fundamentally eliminating the generation and accumulation of invalid data, improving the overall quality and usability of the dataset, and achieving efficient and high-quality posture data acquisition.

[0146] Figure 4 A schematic diagram of the structure of a posture data acquisition device based on virtual reality technology provided in an embodiment of this application is shown. Figure 4 As shown, the device 200 includes: a determination module 210, a mapping module 220, a driving module 230, a judgment module 240, a processing module 250, a display module 260, and a storage module 270.

[0147] The determination module 210 is used to determine the current real posture data of the user when performing a preset task based on the joint tracking device worn by the user.

[0148] The mapping module 220 is used to map the current real posture data to the current virtual posture data of the virtual human body model in the virtual reality device, wherein the virtual reality device is worn on the user's head and is used to display the virtual human body model to the user.

[0149] The drive module 230 is used to drive the virtual human body model in the virtual reality device to follow the user's movements based on the current virtual posture data.

[0150] The judgment module 240 is used to determine whether the current virtual posture data meets the preset motion quality requirements.

[0151] The processing module 250 is used to determine the current virtual posture data as re-recorded posture data if the current virtual posture data does not meet the motion quality requirements, and to generate motion re-recording guidance information corresponding to the re-recorded posture data.

[0152] The display module 260 is used to display motion re-recording guidance information to the user based on the virtual reality device, so as to guide the user to re-execute the re-recording task segment corresponding to the re-recording posture data in the preset task.

[0153] The storage module 270 is used to store the real posture data corresponding to the virtual posture data that meets the action quality requirements during the user's execution of a preset task, so as to complete the posture data acquisition.

[0154] The posture data acquisition device 200 based on virtual reality technology in this application embodiment also includes other modules for performing the steps of the above method embodiments, which will not be described in detail here.

[0155] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. The electronic device can be a server or a VR device.

[0156] like Figure 5 As shown, the electronic device may include a processor 302 and a memory 304.

[0157] The memory 304 is used to store the computer program 306. The memory 304 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device. The computer program 306 may include computer-executable instructions.

[0158] The processor 302 is used to execute the computer program 306 to implement the above-described embodiment of the posture data acquisition method based on virtual reality technology.

[0159] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Electronic device 300 may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0160] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described posture data acquisition method embodiment based on virtual reality technology.

[0161] This application provides a computer program that can be executed by a processor to implement the above-described posture data acquisition method embodiment based on virtual reality technology.

[0162] This application provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described posture data acquisition method embodiment based on virtual reality technology.

[0163] In the several embodiments provided in this application, any function, if implemented as a software functional module / unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the technical solution of this application can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or other electronic device) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing computer program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0164] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of this application are not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of this application.

[0165] It should be noted that the above embodiments are illustrative of this application and not restrictive, and those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In claims enumerating several means, several units or modules of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

[0166] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for acquiring posture data based on virtual reality technology, characterized in that, The method includes: Based on the joint tracking device worn by the user, the current real posture data of the user when performing a preset task is determined; The current real posture data is mapped to the current virtual posture data of the virtual human body model in the virtual reality device, wherein the virtual reality device is worn on the user's head and is used to display the virtual human body model to the user. The virtual human body model in the virtual reality device is driven to follow the user's movements based on the current virtual posture data. Determine whether the current virtual posture data meets the preset motion quality requirements; If the current virtual posture data does not meet the action quality requirements, the current virtual posture data is determined as re-recorded posture data, and action re-recording guidance information corresponding to the re-recorded posture data is generated; Based on the virtual reality device, the action re-recording guidance information is displayed to the user to guide the user to re-execute the re-recording task segment in the preset task that corresponds to the re-recording posture data; The system stores the actual posture data corresponding to the virtual posture data that meets the action quality requirements during the user's execution of the preset task, in order to complete posture data acquisition.

2. The method according to claim 1, characterized in that, The re-recording task segment corresponds to multiple virtual posture data; after displaying the action re-recording guidance information to the user based on the virtual reality device to guide the user to re-execute the re-recording task segment corresponding to the re-recorded posture data in the preset task, the method further includes: Generate a prompt message corresponding to the re-recorded posture data, wherein the prompt message is used to indicate that the current posture of the virtual human body model is an unqualified posture; The virtual human model in the virtual reality device is driven to move again based on multiple virtual posture data corresponding to the re-recorded task segment, and the prompt information is presented in the virtual reality device when the virtual human model in the virtual reality device moves to the posture corresponding to the re-recorded posture data.

3. The method according to claim 1, characterized in that, The storage of virtual posture data corresponding to the virtual posture data that meets the action quality requirements during the user's execution of the preset task, in order to complete posture data acquisition, includes: Identify the repeatedly executed task segments during the user's execution of the preset task to obtain at least one group of repeated task segments, wherein the group of repeated task segments includes multiple repeatedly executed task segments, and each executed task segment corresponds to multiple virtual pose data; The number of re-recorded attitude data in the virtual attitude data corresponding to each executed task segment in the repetitive task segment group is determined as the re-recording amount of the executed task segment; The system stores the actual posture data corresponding to the task segment with the least re-recording in the repetitive task segment group, as well as the actual posture data corresponding to the task segments that were not repeatedly executed during the user's execution of the preset task, in order to complete the posture data acquisition.

4. The method according to claim 1, characterized in that, The step of determining whether the current virtual posture data meets the preset motion quality requirements includes: The current virtual posture data is preliminarily judged according to the preset motion rationality rules to obtain a preliminary judgment result; If the preliminary judgment result indicates that the current virtual posture is reasonable, then the current virtual posture data is determined to meet the action quality requirements. If the preliminary judgment result is that the current virtual posture is unreasonable, the current virtual posture data is input into the pre-trained visual language model to obtain the model judgment result. If the model judgment result is that the current virtual posture is reasonable, the current virtual posture data is determined to meet the action quality requirements. If the model judgment result is that the current virtual posture is unreasonable, the current virtual posture data is determined to not meet the action quality requirements.

5. The method according to claim 4, characterized in that, The motion rationality rules include joint movement rules and joint distance rules; the preliminary determination of the current virtual posture data based on the preset motion rationality rules to obtain a preliminary determination result includes: The degree of joint movement in the current virtual posture is determined based on the key joint data in the previous virtual posture data and the key joint data in the current virtual posture data. Determine the joint distances between multiple key joints in the current virtual posture based on the key joint data in the current virtual posture data; If the degree of joint movement satisfies the joint movement rule, and the joint distance between the multiple key joints satisfies the joint distance rule, then the current virtual posture is determined to be reasonable.

6. The method according to claim 5, characterized in that, The joint movement rule includes the condition that the degree of joint movement does not exceed a preset joint movement degree; The joint distance rule includes the fact that the joint distances between the multiple key joints are all within a preset joint distance range; Alternatively, the joint distance rule includes the condition that the current distance ratio is within a preset reasonable ratio range corresponding to the distance ratio item; The determination process for determining whether the current distance ratio falls within the preset reasonable ratio range corresponding to the distance ratio item includes: For a preset distance ratio, based on the current virtual posture data, a first distance of a first joint set and a second distance of a second joint set required to constitute the distance ratio are determined, wherein the first distance is a geometric metric calculated based on the spatial coordinates of multiple joints included in the first joint set, and the second distance is a geometric metric calculated based on the spatial coordinates of multiple joints included in the second joint set. The ratio of the first distance to the second distance is determined as the current distance ratio of the distance ratio item; Determine whether the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio item; If the current distance ratio is within the preset reasonable ratio range corresponding to the distance ratio item, then it is determined that the joint distance between the multiple key joints satisfies the joint distance rule. The distance ratio term defines the first joint set and the second joint set on which it is based. The relationship between the first joint set and the second joint set includes at least one of the following: the adjacency relationship of ipsilateral limbs, the symmetry relationship about the body midline, or the geometric relationship that together form a specific functional triangle.

7. The method according to claim 5, characterized in that, The joint tracking device includes multiple joint trackers for wearing on key body positions of the user. The key body positions include at least one of the head, torso, arms, hands, pelvis, and lower limbs. The key joint data is data corresponding to the key body positions.

8. A posture data acquisition system based on virtual reality technology, characterized in that, The system includes a server, a joint tracking device, and a virtual reality device, wherein the server is communicatively connected to both the joint tracking device and the virtual reality device. The joint tracking device is worn by the user and collects the user's current real posture data when performing a preset task, and sends the current real posture data to the server. The server is configured to perform the method according to any one of claims 1 to 7, wherein the server generates driving data for driving the virtual human body model based on the current real posture data, and sends the driving data and the action re-recording guidance information corresponding to the re-recorded posture data to the virtual reality device. The virtual reality device is used to display the motion state of the virtual human body model and the motion re-recording guidance information, and to drive the virtual human body model to follow the user's movements according to the driving data.

9. A posture data acquisition system based on virtual reality technology, characterized in that, The system includes a joint tracking device and a virtual reality device with a communication connection; The joint tracking device is worn by the user and collects the user's current real posture data when performing a preset task, and sends the current real posture data to the virtual reality device. The virtual reality device is used to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the posture data acquisition method based on virtual reality technology as described in any one of claims 1 to 7.

Images