Dynamic capture system and method based on virtual digital human

Abstract

The invention relates to a dynamic capture system and method based on a virtual digital human, the dynamic capture system comprises a face capture helmet, a camera, a plurality of posture capture modules, a battery, a microprocessor and a data terminal, the plurality of posture capture modules are respectively connected to each node of human limbs; the microprocessor is connected with the battery, the camera and the posture capturing modules, the data terminal is connected with the microprocessor through the communication module, and each posture capturing module comprises an inertial sensor. According to the design, by arranging the face capturing helmet and the multiple posture capturing modules, face data of a person and dynamic data of all key nodes can be obtained, after the face data and the dynamic data are processed through the microprocessor, corresponding models can be generated and matched with models in an existing model library, and therefore in the subsequent use process, the face capturing effect is greatly improved. When the camera and the posture capturing module collect corresponding data, the data can be directly fed back and displayed through the model, and the cost is low.

Description

Technical Field

[0001] The present invention relates to the technical field of motion capture analysis, and in particular to a motion capture system and method based on virtual digital humans. Background Art

[0002] Digital human is the product of the integration of information science and life science. It uses information science methods to virtually simulate the human body's form and function at different levels. It includes four overlapping development stages: visual human, physical human, physiological human and intelligent human. Ultimately, it establishes a multidisciplinary and multi-level digital model and achieves accurate simulation of the human body from micro to macro.

[0003] Motion capture plays an important role in human-computer interaction in virtual reality. By capturing the user's facial expressions and postures, the virtual reality system can track the user's facial and body movements in real time, enabling the virtual character to better resonate with the user emotionally. Through interaction, the freedom and realism of the virtual reality experience can be increased, and the user's immersion in the virtual environment can be enhanced.

[0004] However, traditional facial expression capture requires sensors placed on the face, which can affect facial movements. Furthermore, motion capture solutions typically require specialized hardware, such as depth cameras and infrared sensors. These devices are expensive and require extensive hardware configuration and precise calibration, resulting in high overall costs. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects and problems of high cost in the prior art and to provide a low-cost motion capture system and method based on virtual digital humans.

[0006] To achieve the above objectives, the technical solution of the present invention is: a motion capture system based on a virtual digital human, the motion capture system comprising a face capture helmet, a camera, multiple gesture capture modules, a battery, a microprocessor, and a data terminal, the multiple gesture capture modules being respectively connected to various nodes of a human limb, the microprocessor being connected to the battery, camera, and multiple gesture capture modules, the data terminal being connected to the microprocessor via a communication module, and the gesture capture modules comprising an inertial sensor;

[0007] The face-catching helmet is used to be worn on the human head;

[0008] The camera is used to obtain a real-time image of the front of the face;

[0009] The inertial sensor is used to obtain various parameters of various key nodes when the human limbs move;

[0010] The microprocessor is used to obtain and process various parameters of the inertial sensor to obtain posture data, and obtain and process the real-time image of the camera to obtain facial expression data information;

[0011] The data terminal is used to import the corresponding action model database according to the facial expression data information, and generate the corresponding action curve and action model library after calculation according to the posture data.

[0012] A motion capture method based on a virtual digital human, the motion capture method being applied to a motion capture system based on a virtual digital human, the motion capture method comprising the following steps:

[0013] Step 1: Import the character model and body movement model of the virtual digital human;

[0014] Step 2: Wear the face capture helmet on the human head and aim the camera at the human face. The camera continuously captures the video stream at a set frame rate. The microprocessor extracts the image frame by frame from the video stream to obtain a real-time image of the front of the face, and obtains the corresponding facial expression data information based on the real-time image.

[0015] Step 3: Match the facial expression data information with the facial action information in the character model library and call it;

[0016] Step 4: Use the posture capture module to collect dynamic data of the key nodes of the human limbs during movement and convert them into various motion models to generate an overall motion model library.

[0017] The step of obtaining corresponding facial expression data information based on the real-time image includes:

[0018] The real-time image of the front face is divided into the left eye area, right eye area, nose area, and mouth area, and each area image is input into the corresponding key point detection network for detection to obtain local key points;

[0019] Perform multiple detections on each area image of the same frame, and calculate multiple local key points detected in the same frame to obtain high-precision key points;

[0020] Compare the position changes of high-precision key points in different frames. If there are abnormal jumps, correct the abnormal high-precision key points.

[0021] The high-precision key points corrected in the four regions are merged into global key points.

[0022] The calculation formula of the high-precision key points is:

[0023]

[0024] Among them, x is the horizontal coordinate of the high-precision key point, y is the vertical coordinate of the high-precision key point, and x i is the horizontal coordinate of the local key point detected for the i-th time, y i is the vertical coordinate of the local key point detected for the i-th time, ∝ i is the confidence of the local key point detected for the i-th time.

[0025] The correction of the high-precision key points of abnormal beating includes:

[0026] Obtain the high-precision key points of the previous continuous frames of the high-precision key points that are abnormally jumping, sum the coordinates of the high-precision key points of the previous continuous frames and the coordinates of the high-precision key points that are abnormally jumping, and take the average to obtain the coordinates of the corrected high-precision key points, and replace the coordinates of the high-precision key points that are abnormally jumping with the coordinates of the corrected high-precision key points.

[0027] The process of collecting dynamic data of key joints of human limbs during movement and converting them into various motion models includes:

[0028] Taking the human head, torso, upper limbs and lower limbs as key nodes, the posture capture modules are respectively tied to the head, torso, upper limbs and lower limbs of the human body. A posture capture module is installed on each shoulder, elbow and wrist of the upper limbs, and a posture capture module is installed on each hip, knee and ankle of the lower limbs.

[0029] The human body moves, and the acceleration parameters, angular velocity parameters and magnetic field parameters of each key node during the movement are obtained through the posture capture module;

[0030] The microprocessor processes the parameters obtained by the posture capture module to obtain posture data and sends it to the data terminal;

[0031] The data terminal calculates the posture data and generates a motion curve graph, which is then converted into various motion models.

[0032] The step of processing the parameters acquired by the posture capture module to obtain posture data includes:

[0033] Calculate the quaternion Q from each posture capture module coordinate system to the bone coordinate system of the wearable part c :

[0034]

[0035] Among them, the symbol Represents quaternion cross product, Q s The quaternion output by each gesture capture module from the geographic coordinate system to the gesture capture module coordinate system, Q0 is the initial quaternion from the geographic coordinate system to the gesture capture module coordinate system measured by each gesture capture module in a stationary state;

[0036] For each posture capture module output quaternion Q c Perform calibration to obtain the calibrated quaternion Q b :

[0037]

[0038] Q b To perform further calibration:

[0039]

[0040] Among them, Q d is the quaternion from the world coordinate system of the virtual model to the coordinate system of each segment of the virtual model in the initial state, f[·] is the conversion operation of the quaternion from the geographic coordinate system to the world coordinate system of the virtual model, Q u The quaternion that drives the virtual model.

[0041] Each gesture capture module outputs the quaternion Q from the geographic coordinate system to the gesture capture module coordinate system. s The calculation formula is as follows:

[0042] Q s =(1-k a -k b )Q g +k a Q a +k m Q m ;

[0043] k a +k m ≤1;

[0044] Among them, k a is the accelerometer weight, k m is the magnetometer weight, Q g is the attitude quaternion estimation based on gyroscope data, Q a is the attitude quaternion estimated based on accelerometer data, Q m is the attitude quaternion estimate based on magnetometer data.

[0045] The attitude quaternion estimation Q obtained based on gyroscope data g The calculation formula is:

[0046]

[0047] Among them, ω x is the angular velocity measured by the gyroscope on the x-axis, ω y is the angular velocity measured by the gyroscope on the y-axis, ωz is the angular velocity measured by the gyroscope on the z-axis, Δt is the sampling interval of the gyroscope, and Q(t) is the attitude quaternion measured by the gyroscope at the current moment;

[0048] Accelerometer-based attitude quaternion Q a The calculation formula is:

[0049]

[0050] θ a =k·||e a ||;

[0051] e a =a×a b ;

[0052]

[0053] a m =(a x , a y , a z );

[0054] a b =(0, 0, -1);

[0055] Among them, θ a is the rotation angle of the accelerometer, e a is the error vector of the accelerometer, ||e a || is the modulus of the accelerometer error vector, k is the proportional coefficient, a m is the acceleration vector measured by the accelerometer, a x is the acceleration of the x-axis measured by the accelerometer, a y is the acceleration of the y-axis measured by the accelerometer, a z is the acceleration of the z-axis measured by the accelerometer, a b is the acceleration in the theoretical gravity direction, a is the normalized acceleration vector;

[0056] Magnetometer-based attitude quaternion Q m The calculation formula is:

[0057]

[0058] θ m =k·||e m ||;

[0059] e m =m a ×m b ;

[0060]

[0061] m=(m x , m y , m z );

[0062] m b =(0, 0, -1);

[0063] Among them, θ m is the rotation angle of the magnetometer, e m is the error vector of the magnetometer, ||e m || is the modulus of the error vector of the magnetometer, k is the proportional coefficient, m is the magnetic field strength vector measured by the magnetometer, m x is the magnetic field strength along the x-axis measured by the magnetometer, m y is the magnetic field strength of the y-axis measured by the magnetometer, m z is the magnetic field strength along the z axis measured by the magnetometer, m b is the magnetic field strength in the theoretical gravity direction, m a is the normalized magnetic field intensity vector.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. The present invention relates to a motion capture system and method for a virtual digital human. By providing a facial capture helmet and multiple gesture capture modules, facial data and dynamic data of key nodes can be acquired. After processing by a microprocessor, a corresponding model can be generated and matched with models in an existing model library. In subsequent use, when the camera and gesture capture modules collect corresponding data, feedback and display can be directly provided through the model. Compared to existing technologies, this system can capture the movements and facial expressions of a virtual digital human model without the need for multiple hardware devices, resulting in lower costs. Therefore, the present invention has a simple structure and low costs.

[0066] 2. The present invention provides a motion capture system and method based on a virtual digital human. By dividing a real-time image into multiple regions and detecting multiple key points in each region, the system can directly drive facial expressions in the model if the detected key points of the image match those in the model library. By setting confidence scores and modifying key points, the detection accuracy is greatly improved, thereby enhancing the reliability of the detection structure. Therefore, the present invention is easy to use and highly reliable.

[0067] 3. In a motion capture system and method based on a virtual digital human, the present invention places gesture capture modules at key nodes of a human limb to acquire dynamic data from these nodes. This data is then converted from a geographic coordinate system to a world coordinate system for the virtual model. This allows for the creation of a three-dimensional model that changes with human motion based on the world coordinate system of the virtual model. After a human body moves, the acquired dynamic data can be directly converted into the virtual three-dimensional model, allowing for direct feedback of the human body's motion through the model, resulting in a high degree of fidelity. Therefore, the present invention is easy to use, has a high degree of fidelity, and is highly reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a structural schematic diagram of a motion capture system based on virtual digital human of the present invention.

[0069] Figure 2 The present invention is a flowchart of a method for capturing motion based on a virtual digital human.

[0070] In the figure: face capture helmet 1, camera 2, gesture capture module 3, battery 4, microprocessor 5, communication module 6, data terminal 7. DETAILED DESCRIPTION

[0071] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] Example 1:

[0073] See also Figure 1 A motion capture system based on a virtual digital human, the motion capture system comprising a face capture helmet 1, a camera 2, a plurality of posture capture modules 3, a battery 4, a microprocessor 5 and a data terminal 7, wherein the plurality of posture capture modules 3 are respectively connected to each node of a human limb, the microprocessor 5 is connected to the battery 4, the camera 2, and the plurality of posture capture modules 3, the data terminal 7 is connected to the microprocessor 5 via a communication module 6, and the posture capture module 3 comprises an inertial sensor;

[0074] The face-catching helmet 1 is used to be worn on the head of a human body;

[0075] The camera 2 is used to obtain a real-time image of the front face of a person;

[0076] The inertial sensor is used to obtain various parameters of various key nodes when the human limbs move;

[0077] The microprocessor 5 is used to obtain and process various parameters of the inertial sensor to obtain posture data, and obtain and process the real-time image of the camera 2 to obtain facial expression data information;

[0078] The data terminal 7 is used to import the corresponding motion model database according to the facial expression data information, and generate the corresponding motion curve and motion model library after calculation according to the posture data.

[0079] In this embodiment, the face capture helmet 1 is made of nylon, and a bracket is provided between the face capture helmet 1 and the camera 2. The bracket adopts a hollow metal hose, which can flexibly adjust the angle, distance and direction of the camera before capturing. After the metal hose is freely adjusted, the camera 2 is facing the front of the human face, covering all expression areas. A Bluetooth module is provided in the inertial sensor for wireless communication. At the same time, the inertial sensor is connected to each key node of the human body through a strap to realize accurate capture of the motion data of the corresponding node. The abdominal belt can adopt a flexible nylon belt with a breathable sponge pad covered on the surface. One end of the abdominal belt is a locking tongue and the other end is an adjustable lock buckle. The locking tongue and the lock buckle are connected in coordination.

[0080] Example 2:

[0081] See also Figure 2 A motion capture method based on a virtual digital human is provided. The motion capture method is applied to a motion capture system based on a virtual digital human as described in Example 1. The motion capture method comprises the following steps:

[0082] Step 1: Import the character model and body movement model of the virtual digital human;

[0083] Step 2: Wear the face capture helmet 1 on the human head and aim the camera 2 at the human face. The camera 2 continuously captures the video stream at a set frame rate. The microprocessor 5 extracts the image frame by frame from the video stream to obtain a real-time image of the front of the human face, and obtains the corresponding facial expression data information based on the real-time image.

[0084] Step 3: Match the facial expression data information with the facial action information in the character model library and call it;

[0085] Step 4: The posture capture module 3 is used to collect dynamic data of key nodes of the human body during movement and convert them into various motion models to generate an overall motion model library.

[0086] In this embodiment, first, the corresponding character model and limb movement model information library are generated through 3DMAX software, and the sensitivity coefficient is adjusted at the same time. Through the received real-time facial expression data information of the human face, the platform is used to quickly match and call the facial movement information in the relevant association mapping information library, thereby realizing real-time driving of the facial expression of the virtual digital human. After the inertial sensor completes the dynamic data collection during the movement of each node, the data captured by the inertial sensor can be transmitted to the software through the wired communication method of the USB interface or the wireless communication method of Bluetooth. These dynamic data are counted and the corresponding action curve diagram is drawn. The data format is first converted to the data format, and then the file at this time is directly imported into the 3DMAX software, so that the above-mentioned action curve diagrams of each key node are converted into each action model, and then the overall action model library is generated.

[0087] Example 3:

[0088] The basic content is the same as Example 2, except that:

[0089] The step of obtaining corresponding facial expression data information based on the real-time image includes:

[0090] The real-time image of the front face is divided into the left eye area, right eye area, nose area, and mouth area, and each area image is input into the corresponding key point detection network for detection to obtain local key points;

[0091] Perform multiple detections on each area image of the same frame, and calculate multiple local key points detected in the same frame to obtain high-precision key points;

[0092] Compare the position changes of high-precision key points in different frames. If there are abnormal jumps, correct the abnormal high-precision key points.

[0093] The high-precision key points corrected in the four regions are merged into global key points.

[0094] The calculation formula of the high-precision key points is:

[0095]

[0096] Among them, x is the horizontal coordinate of the high-precision key point, y is the vertical coordinate of the high-precision key point, and x i is the horizontal coordinate of the local key point detected for the i-th time, y i is the vertical coordinate of the local key point detected for the i-th time, ∝ i is the confidence of the local key point detected for the i-th time.

[0097] The correction of the high-precision key points of abnormal beating includes:

[0098] Obtain the high-precision key points of the previous continuous frames of the high-precision key points that are abnormally jumping, sum the coordinates of the high-precision key points of the previous continuous frames and the coordinates of the high-precision key points that are abnormally jumping, and take the average to obtain the coordinates of the corrected high-precision key points, and replace the coordinates of the high-precision key points that are abnormally jumping with the coordinates of the corrected high-precision key points.

[0099] In this embodiment, the unit of coordinates is pixel. When the pixel of a high-precision key point is greater than a preset threshold, it is considered abnormal jumping. The preset threshold is 5 pixels. When a fast-motion scene occurs, the preset threshold is 8-10 pixels. In scenes with high real-time requirements, the continuous frames are 5 frames. For scenes such as high-precision animation production, the continuous frames are 10 frames. The number of global key points in each frame is 68.

[0100] Example 4:

[0101] The basic content is the same as Example 2, except that:

[0102] The process of collecting dynamic data of key joints of human limbs during movement and converting them into various motion models includes:

[0103] Taking the head, trunk, upper limbs and lower limbs of the human body as key nodes, the posture capture modules 3 are respectively tied to the head, trunk, upper limbs and lower limbs of the human body. One posture capture module 3 is installed on each shoulder, elbow and wrist of the upper limbs, and one posture capture module 3 is installed on each hip, knee and ankle of the lower limbs.

[0104] The human body moves, and the acceleration parameters, angular velocity parameters and magnetic field parameters of each key node during the movement are obtained through the posture capture module 3;

[0105] The microprocessor 5 processes the parameters acquired by the posture capture module 3 to obtain posture data and sends it to the data terminal 7;

[0106] The data terminal 7 calculates the posture data to generate a motion curve diagram, and converts the motion curve diagram into various motion models.

[0107] The process of processing the parameters acquired by the posture capture module 3 to obtain posture data includes:

[0108] Calculate the quaternion Q from each posture capture module coordinate system to the bone coordinate system of the wearable part c :

[0109]

[0110] Among them, the symbol Represents quaternion cross product, Q sis the quaternion output by each gesture capture module 3 from the geographic coordinate system to the gesture capture module coordinate system, and Q0 is the initial quaternion from the geographic coordinate system to the gesture capture module coordinate system measured by each gesture capture module 3 in a stationary state;

[0111] For each quaternion Q output by the posture capture module 3 c Perform calibration to obtain the calibrated quaternion Q b :

[0112]

[0113] Q b To perform further calibration:

[0114]

[0115] Among them, Q d is the quaternion from the world coordinate system of the virtual model to the coordinate system of each segment of the virtual model in the initial state, f[·] is the conversion operation of the quaternion from the geographic coordinate system to the world coordinate system of the virtual model, Q u The quaternion that drives the virtual model.

[0116] Each gesture capture module 3 outputs the quaternion Q from the geographic coordinate system to the gesture capture module coordinate system. s The calculation formula is as follows:

[0117] Q s =(1-k a -k b )Q g +k a Q a +k m Q m ;

[0118] k a +k m ≤1;

[0119] Among them, k a is the accelerometer weight, k m is the magnetometer weight, Q g is the attitude quaternion estimation based on gyroscope data, Q a is the attitude quaternion estimated based on accelerometer data, Q m is the attitude quaternion estimate based on magnetometer data.

[0120] The attitude quaternion estimation Q obtained based on gyroscope data g The calculation formula is:

[0121]

[0122] Among them, ω x is the angular velocity measured by the gyroscope on the x-axis, ω y is the angular velocity measured by the gyroscope on the y-axis, ω z is the angular velocity measured by the gyroscope on the z-axis, Δt is the sampling interval of the gyroscope, Q(t) is the attitude quaternion measured by the gyroscope at the current moment,

[0123] Accelerometer-based attitude quaternion Q a The calculation formula is:

[0124]

[0125] θ a =k·||e a ||;

[0126] e a =a×a b ;

[0127]

[0128] a m =(a x , a y , a z );

[0129] a b =(0, 0, -1);

[0130] Among them, θ a is the rotation angle of the accelerometer, e a is the error vector of the accelerometer, ||e a || is the modulus of the accelerometer error vector, k is the proportional coefficient, a m is the acceleration vector measured by the accelerometer, a x is the acceleration of the x-axis measured by the accelerometer, a y is the acceleration of the y-axis measured by the accelerometer, a z is the acceleration of the z-axis measured by the accelerometer, a b is the acceleration in the theoretical gravity direction, a is the normalized acceleration vector;

[0131] Magnetometer-based attitude quaternion Q m The calculation formula is:

[0132]

[0133] θ m =k·||e m ||;

[0134] e m =ma ×m b ;

[0135]

[0136] m=(m x , m y , m z );

[0137] m b =(0, 0, -1);

[0138] Among them, θ m is the rotation angle of the magnetometer, e m is the error vector of the magnetometer, ||e m || is the modulus of the error vector of the magnetometer, k is the proportional coefficient, m is the magnetic field strength vector measured by the magnetometer, m x is the magnetic field strength along the x-axis measured by the magnetometer, m y is the magnetic field strength of the y-axis measured by the magnetometer, m z is the magnetic field strength along the z axis measured by the magnetometer, m b is the magnetic field strength in the theoretical gravity direction, m a is the normalized magnetic field intensity vector.

[0139] In this embodiment, after wearing the inertial sensor, you need to maintain a static upright posture, and align the posture capture module coordinate system with the geographic coordinate system through the relational sensor data (gravitational acceleration and geomagnetic field direction) in the static state.

Claims

1. A motion capture system based on a virtual digital human, characterized by: The dynamic capture system comprises a face capture helmet (1), a camera (2), a plurality of posture capture modules (3), a battery (4), a microprocessor (5) and a data terminal (7); the plurality of posture capture modules (3) are respectively connected to respective nodes of a human limb; the microprocessor (5) is connected to the battery (4), the camera (2) and the plurality of posture capture modules (3); the data terminal (7) is connected to the microprocessor (5) via a communication module (6); and the posture capture module (3) comprises an inertial sensor; The face-catching helmet (1) is used to be worn on the head of a human body; The camera (2) is used to obtain a real-time image of the front of the human face; The inertial sensor is used to obtain various parameters of various key nodes when the human limbs move; The microprocessor (5) is used to obtain and process various parameters of the inertial sensor to obtain posture data, and to obtain and process the real-time image of the camera (2) to obtain facial expression data information; The data terminal (7) is used to import the corresponding action model database according to the facial expression data information, and generate the corresponding action curve and action model library after calculation according to the posture data.

2. A motion capture method based on a virtual digital human, characterized by: The motion capture method is applied to the motion capture system based on virtual digital human according to claim 1, and the motion capture method comprises the following steps: Step 1: Import the character model and body movement model of the virtual digital human; Step 2: Wear the face capture helmet (1) on the human head and aim the camera (2) at the human face. The camera (2) continuously collects a video stream at a set frame rate. The microprocessor (5) extracts images from the video stream frame by frame to obtain a real-time image of the front of the human face. The corresponding facial expression data information is obtained according to the real-time image. Step 3: Match the facial expression data information with the facial action information in the character model library and call it; Step 4: The dynamic data of the key nodes of the human limbs during movement are collected through the posture capture module (3) and converted into various motion models to generate an overall motion model library.

3. The method for capturing motion based on a virtual digital human according to claim 2, characterized in that: The step of obtaining corresponding facial expression data information based on the real-time image includes: The real-time image of the front face is divided into the left eye area, right eye area, nose area, and mouth area, and each area image is input into the corresponding key point detection network for detection to obtain local key points; Perform multiple detections on each area image of the same frame, and calculate multiple local key points detected in the same frame to obtain high-precision key points; Compare the position changes of high-precision key points in different frames. If there are abnormal jumps, correct the abnormal high-precision key points. The high-precision key points corrected in the four regions are merged into global key points.

4. The method for capturing motion based on a virtual digital human according to claim 3, characterized in that: The calculation formula of the high-precision key points is: Among them, x is the horizontal coordinate of the high-precision key point, y is the vertical coordinate of the high-precision key point, and x i is the horizontal coordinate of the local key point detected for the i-th time, y i is the vertical coordinate of the local key point detected for the i-th time, ∝ i is the confidence of the local key point detected for the i-th time.

5. The method for capturing motion based on a virtual digital human according to claim 4, characterized in that: The correction of the high-precision key points of abnormal beating includes: Obtain the high-precision key points of the previous continuous frames of the high-precision key points that are abnormally jumping, sum the coordinates of the high-precision key points of the previous continuous frames and the coordinates of the high-precision key points that are abnormally jumping, and take the average to obtain the coordinates of the corrected high-precision key points, and replace the coordinates of the high-precision key points that are abnormally jumping with the coordinates of the corrected high-precision key points.

6. The method for capturing motion based on a virtual digital human according to claim 2, characterized in that: The process of collecting dynamic data of key joints of human limbs during movement and converting them into various motion models includes: Taking the head, trunk, upper limbs and lower limbs of the human body as key nodes, the posture capture modules (3) are respectively tied to the head, trunk, upper limbs and lower limbs of the human body, one posture capture module (3) is installed on each shoulder, elbow and wrist of the upper limbs, and one posture capture module (3) is installed on each hip, knee and ankle of the lower limbs; The human body moves, and the acceleration parameters, angular velocity parameters and magnetic field parameters of each key node during the movement are obtained through the posture capture module (3); The microprocessor (5) processes the parameters acquired by the posture capture module (3) to obtain posture data and sends the posture data to the data terminal (7); The data terminal (7) calculates the posture data and generates an action curve diagram, which is then converted into various action models.

7. The method for capturing motion based on a virtual digital human according to claim 6, characterized in that: The parameters acquired by the posture capture module (3) are processed to obtain posture data, including: Calculate the quaternion Q from each posture capture module coordinate system to the bone coordinate system of the wearable part c : Among them, the symbol Represents quaternion cross product, Q s is a quaternion outputted from a geographic coordinate system to a gesture capture module coordinate system by each gesture capture module (3), and Q0 is an initial quaternion from a geographic coordinate system to a gesture capture module coordinate system measured by each gesture capture module (3) in a stationary state; For each quaternion Q output by the posture capture module (3) c Perform calibration to obtain the calibrated quaternion Q b : Q b To perform further calibration: Among them, Q d is the quaternion from the world coordinate system of the virtual model to the coordinate system of each segment of the virtual model in the initial state, f[·] is the conversion operation of the quaternion from the geographic coordinate system to the world coordinate system of the virtual model, Q u The quaternion that drives the virtual model.

8. The method for capturing motion based on a virtual digital human according to claim 7, characterized in that: Each posture capture module (3) outputs a quaternion Q from the geographic coordinate system to the posture capture module coordinate system. s The calculation formula is as follows: Q s =(1-k a -k b )Q g +k a Q a +k m Q m ; k a +k m ≤1; Among them, k a is the accelerometer weight, k m is the magnetometer weight, Q g is the attitude quaternion estimation based on gyroscope data, Q a is the attitude quaternion estimated based on accelerometer data, Q m is the attitude quaternion estimate based on magnetometer data.

9. The method for capturing motion based on a virtual digital human according to claim 8, characterized in that: The attitude quaternion estimation Q obtained based on gyroscope data g The calculation formula is: Among them, ω x is the angular velocity measured by the gyroscope on the x-axis, ω y is the angular velocity measured by the gyroscope on the y-axis, ω z is the angular velocity measured by the gyroscope on the z-axis, Δt is the sampling interval of the gyroscope, and Q(t) is the attitude quaternion measured by the gyroscope at the current moment; Accelerometer-based attitude quaternion Q a The calculation formula is: i a =k·||e a ||; and a =a×α b ; a m =(a x ,a y ,a z ); a b ＝(0,0,-1)； Among them, θ a is the rotation angle of the accelerometer, e a is the error vector of the accelerometer, ||e a || is the modulus of the accelerometer error vector, k is the proportional coefficient, a m is the acceleration vector measured by the accelerometer, a x is the acceleration of the x-axis measured by the accelerometer, a y is the acceleration of the y-axis measured by the accelerometer, a z is the acceleration of the z-axis measured by the accelerometer, a b is the acceleration in the theoretical gravity direction, a is the normalized acceleration vector; Magnetometer-based attitude quaternion Q m The calculation formula is: i m =k·||e m ||; have been m =m a xm b ; m=(m x ,m y ,m z ); m b ＝(0,0,-1)； Among them, θ m is the rotation angle of the magnetometer, e m is the error vector of the magnetometer, ||e m || is the modulus of the error vector of the magnetometer, k is the proportional coefficient, m is the magnetic field strength vector measured by the magnetometer, m x is the magnetic field strength along the x-axis measured by the magnetometer, m y is the magnetic field strength of the y-axis measured by the magnetometer, m z is the magnetic field strength along the z axis measured by the magnetometer, m b is the magnetic field strength in the theoretical gravity direction, m a is the normalized magnetic field intensity vector.

Images