CN108230429A - Real-time whole body posture reconstruction method based on head and two-hand positions and posture - Google Patents

Abstract

A real-time whole-body posture reconstruction method based on the position and posture of the head and hands, using the inverse dynamics scheme to reconstruct the upper body, that is, according to the head position, first estimate the waist position and the shoulder position, and then use the reverse dynamics method according to the input hands position The position of the elbows is calculated by the dynamic method; then the lower body is reconstructed by means of preset animation blending, and finally the positions of all joints are mapped to the reconstructed upper and lower body of the character in turn to complete the reconstruction of the whole body posture.

Description

Real-time Whole Body Pose Reconstruction Method Based on Head and Hands Position and Pose

technical field

The present invention relates to a technology in the field of virtual reality, specifically a method of combining inverse dynamics and animation based on the positions and postures of the user's hands and head, respectively to simulate the user's upper body and lower body in virtual reality. A technique for real-time body pose reconstruction.

Background technique

With the continuous improvement of hardware specifications, virtual reality technology has become hot again, and is gradually favored by people in the industry. Although the current popularization of equipment cannot make users feel the charm of virtual reality, it is believed that when there are enough users and the content is rich enough, virtual reality will radiate its magic.

In various virtual reality applications, the need to reconstruct the human body into the virtual world is often used. For example, in multiplayer virtual reality, each user hopes to see his teammates appear in the virtual world in a human form, instead of a suspended head and two suspended hands currently used due to the limitation of tracking equipment. For the field of virtual reality that pursues immersion, it is a place where progress is needed. Therefore, how to reconstruct human body movements as accurately as possible and in real time based on a small amount of low-dimensional sensor information has become a new research point in the industry.

Before the popularity of virtual reality, there have been a lot of related research work on the human body pose reconstruction algorithm based on a small number of sensors, among which: the more popular scheme is based on the big data-driven method, according to the real-time given sensor pose data to search and match in The existing human body poses in the pose database, and then according to the energy function, the search results, the actual input and the action correlation reach an optimal trade-off condition, and finally the reconstructed pose is presented.

Contents of the invention

In view of the defects in the prior art that the reconstruction results largely depend on the amount of motion data in the database, which cannot be synthesized or poses that do not exist in the motion database, the present invention proposes a real-time whole-body pose based on the position and pose of the head and hands The reconstruction method only needs to provide the position and posture of the hands and the head in real time, and the upper and lower body are reconstructed in real time by means of inverse dynamics and animation mixing respectively.

The present invention is achieved through the following technical solutions:

The present invention adopts the inverse dynamics scheme to reconstruct the upper body, that is, according to the position of the head, first estimate the position of the waist and the position of the shoulders, and then use the inverse dynamics method to calculate the position of the elbows according to the input position of the hands; The lower body is reconstructed in the same way, and finally the positions of all joints are mapped to the upper and lower body of the reconstructed character in turn to complete the reconstruction of the whole body pose.

The upper body reconstruction refers to: after the calibration is completed by the virtual reality controller, the average positive direction of the head and the average direction of the vertical bisector of the line connecting the hands are obtained as feature values by using the data collected in the sensor recording and buffering process to calculate , the angle between the line connecting the two hands and the positive direction of the body, and the angle between the line connecting the two hands and the vertical plane where the head is located, and obtain the current positive direction of the human body; then calculate the waist height, the absolute position of the waist, the left and right The position of the right shoulder and the position of the elbows are calculated according to the inverse dynamics method.

The lower body reconstruction refers to: use the data collected in the process of sensor recording and buffering to calculate the instantaneous speed of the head and the swinging posture of the arms, so as to mix the preset lower body animation, and complete the lower body after legality testing. reconstruction.

The present invention specifically comprises the following steps:

Step 1: The user holds a virtual reality controller, wears a head-mounted display, and poses in a standard calibration posture (T pose) to calibrate the relevant parameters h _s and w _s of the human body, so as to adapt to people of different heights and arm lengths to use the reconstruction scheme.

The standard correction posture refers to: standing upright, raising both arms sideways, and facing the head straight ahead.

The human body-related parameters described Among them: p _h refers to the position of the head, h _m refers to the height of the character model that needs to be mapped to the virtual scene, p _r refers to the position of the right hand, p _l refers to the position of the left hand, and w _m refers to the arm span of the character model that needs to be mapped to the virtual scene.

Step 2, record the collected data through the sensor, and count the auxiliary parameters in real time.

The data includes: the head position and head posture given by the head-mounted display, and the position and posture of both hands given by the hand-held controller.

Preferably, the head pose and hands pose are recorded in the form of quaternions.

The auxiliary parameters include but are not limited to: the velocity and direction of a certain tracking body in the previous several frames can be obtained from the position information and time interval; the acceleration and direction of a certain tracking object in the previous several frames can be obtained from the velocity information and the time interval Find out.

The auxiliary parameters can be manually adjusted according to the scene where the present invention is applied. If a more sensitive judgment needs to be made on whether the head moves, it can be judged by calculating the speed of the head tracking body in the first two frames at this stage. , when you want to make a slower judgment, you can change it to judge according to the speed of the head tracking body in the first 10 frames.

Step 3, the positive direction of the human body, that is, the estimation of the direction facing the chest of the person. Since the same set of head and hands position and posture information may correspond to a variety of positive directions of the human body, when the positive direction of the person remains unchanged , estimated based on the most likely positive direction, including:

3.1) Calculate the average positive direction of the head separately The average direction of the perpendicular bisector of the line connecting the two hands Specifically: in: Refers to the average positive direction of the head accumulated before this frame, hf _p refers to the projection of the positive direction of the head on the xz plane in the world coordinate system, n refers to The statistics are the average positive direction of the head of n frames;

in: Refers to the positive direction of the vertical bisector of the line connecting the two hands accumulated before this frame, lhp and rhp respectively refer to the positions of the left hand and right hand in this frame, (lhp-rhp) _p refers to the line between the two hands in world coordinates The projection on the xz plane in the system, n refers to The statistics are the average positive direction of n frames, and up refers to the upward direction in the world coordinate system, that is, the direction (0, 0, 1).

3.2) Further calculate the angle a _cross between the line connecting the hands and the positive direction of the body and the angle a _linear between the line connecting the hands and the vertical plane where the head is located, specifically: Among them: hp represents the position of the head in this frame, lhp and rhp respectively represent the positions of the left hand and right hand in this frame, and the subscript p represents the projection of the vector on the xz plane in the world coordinate system.

3.3) According to a _cross and a _linear and and Determine the current positive direction of the human body, including:

3.3.1) According to the speed statistics in step 2, it is found that in the first 5 frames, the head and hands are in a static state, it is necessary to confirm whether a _cross is at 85° to 95° or whether a _linear is at Between -5° and 5°, if yes, the positive direction of the body in this frame

3.3.2) When the head is in a static state in recent frames, but the hands are in a non-stationary state, the positive direction of the head will be more reliable than the vertical bisector of the line connecting the hands. Therefore, when and When the angle between is less than 5°, the positive direction of the body in this frame otherwise

3.3.3) When the head is in a non-stationary state in recent frames, but the hands are in a stationary state, the average positive direction of the head It will no longer be reliable. It has been observed that in this case, most of the users are looking around in a static state. Therefore, the positive direction of the body is not arbitrarily changed at this time, that is, the positive direction of the body in this frame is equal to that in the previous frame. The positive direction of the body, bf _c ＝bf _l

3.3.4) When the head and both hands are in a non-stationary state in recent frames, there is no stable input to judge the positive direction of the human body. At this time, the previous positive direction of the human body is still maintained, that is, in this frame The positive direction of the body is equal to the positive direction of the body in the previous frame, bf _c =bf _l .

Preferably, a linear interpolation is performed on the current positive direction of the human body bf _a to ensure that the positive direction of the human body will not be mutated due to unstable input signals, that is, bf _a =bf _l *(1-f)+bf _c * f, where: f is a mixing parameter.

Step 4, construct the left shoulder position lse, the right shoulder position rse and the waist height be _h of the upper body to complete the reconstruction of the upper body torso, specifically including:

4.1) Calculate the waist height according to the position of the head, namely Among them: h _h is the height of the head, h _m is the height of the character model that needs to be mapped to the virtual scene, m _uh refers to the distance from the head of the character model to the waist, h _mh is the minimum head that the character model can squat to height to prevent the waist from being lower than the feet.

4.2) Calculate the absolute position be of the waist according to the horizontal position of the head; preferably, use interpolation to slow down the slight movement of the head, and use the interpolated horizontal position of the head to participate in the calculation.

4.3) Calculate the left shoulder position lse and right shoulder position rse respectively according to the absolute position be of the waist, specifically: lse=me-offset; rse=me+offset, wherein: factor _s is the adjustment parameter of the shoulder height, and bf is the adjustment parameter in step The positive direction of the body calculated in the third step, factor _w is the adjustment parameter of shoulder width, me=be+(hp-be)/factor _s ;

Preferably, the squat is activated when the user's eyes are looking down, that is, the user may be looking at something at his feet and wants to bend down to pick it up; or when the user's head has acceleration in the vertical direction state, that is, the position of the waist in the horizontal direction is frozen; and when the head no longer has vertical acceleration, the offset between the head and the waist is recorded, so as to ensure that the user can accurately walk when squatting. The position of the waist is inferred; when the user stands up again, ie the HMD reaches the calibrated height, the offset is zeroed out.

Step 5, according to the positions lhp and rhp of the handheld controller and the positions lse and rse of the two shoulders, the commonly used inverse dynamics method is used to calculate the positions of the elbows to complete the reconstruction process of the upper body.

Step 6: Use the preset animation mapping of lower limbs as the lower body of the reconstructed model, and check the validity of the mapped results, and then map the positions of all joints to the reconstructed upper and lower body of the character in turn to complete the frame. reconstruction.

The preset animation of the lower limbs refers to: preparing several cyclic animations in advance, including but not limited to: forward, left forward, right forward, left, right, left backward, right backward, backward; According to the direction and speed of the user's movement, the commonly used animation mixing technology is used to mix the loopable animation according to different weights.

The legitimacy detection is to detect the legitimacy of the mapped lower limb position, and judge whether the result of animation mixing is legal, that is, whether the lower limb has penetrated into other objects; The same inverse kinematics calculations are performed to avoid the mistake of elongating the foot when penetrating the ground or jumping.

The loopable animation described above is preferably decided by the swing direction of both hands whether to step the left foot or the right foot first. When the user swings the left arm forward and the right arm backward at the moment of walking, the animation starts with the right foot first; When a forward right arm is detected, a backward left arm is detected, or a legal arm swing is not detected, the animation starts with the left foot out first.

The present invention relates to a system for implementing the above method, including: sensor data cache and statistics module, body forward direction estimation module, reverse dynamics solution module, upper body reconstruction module and lower body reconstruction module, wherein: sensor data cache and statistics module will The data from the sensor is buffered and statistically output to the body positive direction estimation module and the lower body reconstruction module. The body positive direction estimation module estimates the human body's positive direction based on statistical information and outputs it to the upper body reconstruction module. The upper body reconstruction module obtains the waist height according to the human body's positive direction , the absolute position of the waist, and the positions of the left and right shoulders are output to the inverse dynamics solving module for upper body reconstruction, and the lower body reconstruction module uses the animation mixing method to mix the preset cyclic animation according to the statistical information, and mixes the mixed results Output to the inverse dynamics solving module for legality detection and mapping of all joints.

technical effect

Compared with the prior art, the present invention realizes a reconstruction method based on a very small number of sensors that can reconstruct the human body in real time after considering the requirement of the virtual reality system to reconstruct the human body. This invention abandons the tracking information of the feet to adapt to the tracking configuration provided by the popular virtual reality devices. The main reason is that most virtual reality applications do not have a lot of dependence on the movements of the feet, that is to say, natural feet Animation is enough for general virtual reality social applications. The accurate reconstruction of the upper body is more important than the lower body, and in virtual reality applications, everyone's eyes are blocked by the head-mounted display, so as long as the reconstruction scheme can reconstruct Natural enough human motion can be applied to a large number of common virtual reality applications. Therefore, the present invention adopts a relatively accurate inverse dynamics scheme to accurately restore the arm animation, and adopts animation mixing technology to quickly restore the leg animation.

Description of drawings

Fig. 1 is the simulation flowchart that the present invention proposes;

Fig. 2 is a schematic diagram of the average positive direction of the head in the present invention;

Fig. 3 is in the present invention, the schematic diagram to the perpendicular bisector direction of two-hand connection line;

Fig. 4 is in the present invention, the sign diagram of diagonal a _cross ;

Fig. 5 is the label diagram of diagonal a _linear in the present invention;

Fig. 6 is the effect demonstration figure that embodiment makes;

In the figure: a is walking forward, b is walking forward right, c is walking rightward, d is walking backward left, e is walking backward, f is squatting, g is walking in half squat state, h is the jump action.

Detailed ways

In this embodiment, a robot model is used to demonstrate the reconstruction algorithm proposed in the present invention. In this scenario, the user's actions within the specified range will be reconstructed in the virtual reality system in real time. In multi-person applications, the reconstructed results will give other users a stronger immersive experience.

The following steps of this embodiment are all based on the HTC Vive virtual reality device during the experiment, which includes a head-mounted display and two hand controllers as standard, which can capture the position and posture of the user's head and hands, and send sensor signals Through the underlying interface OpenVR, it is converted into the position and posture of the head and hands in the space coordinate system. The sensor configuration of the present invention is currently the standard configuration of high-end head-mounted displays in the field of virtual reality, and the solution is also compatible with Oculus and PSVR.

As shown in Figure 1, this embodiment includes the following steps:

Step 1, create an empty scene, create a ground with a collision body, and import the robot model, and obtain the height, waist height, shoulder width and height information of the robot model, and write it into the configuration file of the present invention together with other relevant information middle.

The rest of the relevant information in the definition step 1 refers to the sensitivity standard of human body motion, such as the speed at which the head moves, which will be judged by the system as the start of walking, etc.

Step 2: The user wears a virtual reality device, that is, a head-mounted display and a handheld motion controller, and poses in a T pose to correct the robot model to suit users of various sizes.

The virtual reality controller adopts but is not limited to a handle device used to obtain hand interaction and absolute position and attitude of the hand, such as Oculus Touch, HTC Vive hand motion controller, PS Move and other equipment. A head-mounted display refers to a display worn on the head in a virtual reality device to present stereoscopic images for both eyes, such as Oculus Rift, HTCVive, PSVR, etc. Head-mounted displays can also track the absolute position and pose of the head.

Step 3: After calibration, start to enter the reconstruction algorithm, record the input sensor data, cache the sensor data within 50 frames, and count some quantities that are helpful for subsequent calculations in real time. For example, the user's head speed, head movement direction, hand speed and hand movement direction in the first 5 frames.

Step 4, estimating the positive direction of the human body, the specific steps include:

4.1) As shown in Figure 3 and Figure 4, calculate the average positive direction of the head The calculation formula is in: Refers to the average positive direction of the head accumulated before this frame. hf _p refers to the projection of the positive direction of the head on the xz plane in the world coordinate system, and n refers to What is counted is the average positive direction of the head of n frames.

4.2) Calculate the average direction of the vertical bisector of the line connecting the hands The calculation formula is in: Refers to the positive direction of the vertical bisector of the accumulated two-hand line before this frame. lhp and rhp respectively refer to the positions of the left hand and right hand in this frame, (lhp-rhp) _p refers to the projection of the line between the two hands on the xz plane in the world coordinate system, and n refers to The statistics are the average positive direction of n frames, and up refers to the upward direction in the world coordinate system, that is, the direction (0, 0, 1).

4.3) As shown in Figure 4, the angle a _cross is calculated to determine the angle between the line connecting the two hands and the positive direction of the body. Among them: hp represents the position of the head in this frame, lhp and rhp represent the positions of the left hand and the right hand in the frame respectively. The subscript p represents the projection of the vector on the xz plane in the world coordinate system.

4.4) As shown in Figure 5, the angle a _linear is calculated to determine the angle between the line connecting the two hands and the vertical plane where the head is located. Among them: the relevant symbols are the same as above.

4.4.1) According to the speed statistics in step 3, if the head and hands are in a static state in the last 5 frames, you need to confirm whether a _cross is at 85° to 95° or whether a _linear is at -5 Between ° and 5°, when yes, the positive direction of the body in this frame

4.4.2) When in recent frames, the head is in a static state, but the hands are in a non-stationary state, when and When the difference between is less than 5°, then otherwise,

4.4.3) When in recent frames, the head is in a non-stationary state, but the hands are in a static state, then bf _c = bf _l

4.4.4) When the head, that is, both hands are in a non-stationary state in recent frames, then bf _c =bf _l

In the above steps, bf _c is the positive direction of the body in this frame, and bf _l is the positive direction of the body in the previous frame.

4.5) Finally, the actual positive direction bf _a of the body needs to do a linear interpolation to ensure that the positive direction of the human body will not change suddenly due to unstable input signals, that is, bf _a = bf _l *(1-f)+ bf _c *f, where: f is a mixing parameter. According to experiments, the mixing result looks the most natural when the value is 0.1.

Step 5: Reconstruct the upper body torso, that is, the left shoulder position lse, the right shoulder position rse and the waist position be.

5.1) Calculate the waist height according to the position of the head Among them: h _h is the height of the head, h _m is the height of the character model that needs to be mapped to the virtual scene, m _uh refers to the distance from the head of the character model to the waist, h _mh is the minimum head that the character model can squat to height to prevent the waist from being lower than the feet.

5.2) After the waist height be _h is obtained, the position of the waist on the horizontal plane needs to be obtained, so as to know the absolute position be of the waist. Since it can only be estimated by the position of the head, and because the position of the head will not be static, there will always be a slight drift, so when the horizontal position of the waist is completely calculated according to the horizontal position of the head, the character will always be The problem of fluttering, therefore, the interpolation method is used to slow down the small movement of the head to stabilize the position of the waist.

4.3) After the waist position be is known, the calculation formulas of lse and rse are as follows: me=be+(hp-be)/factor _s ; lse=me-offset; rse=me+offset; where factor _s is the adjustment parameter of the shoulder height, bf is the positive direction of the body calculated in step 3, and factor _w is the adjustment parameter of the shoulder width.

Step 6: Use the inverse dynamics algorithm to calculate the elbow position with the left and right shoulder positions as the roots and the left and right hand motion sensors as the end nodes.

Step 7: According to the direction and speed of the user's movement, the animation mixing technology commonly used in China is used to mix the 8 pre-prepared animations according to different weights, and finally map the mixed results to the lower body of the reconstructed model.

In the above steps, for the forward and backward two animations, it is decided according to the swing direction of the hands whether to step the left foot or the right foot first. When the user swings the left arm forward and the right arm backward at the moment of walking , then move the right foot out first, otherwise, when it is detected that the right arm is swinging forward, the left arm is swinging backward or no legal arm swing action is detected, the left foot is first moved out.

Step 8: Check the legality of the foot position to determine whether the result of animation mixing is legal. If it is not legal, use the same inverse dynamics method as the hand to calculate it, so as to avoid penetrating the ground or jumping An error where the foot is elongated.

In steps 3 to 8 mentioned above, the computer continuously performs loop operations to reconstruct the human animation of each frame in real time.

The final effect of this embodiment is shown in Figure 6, wherein Figure 6a is the forward effect, Figure 6b is the right forward effect, Figure 6c is the left effect, Figure 6d is the left backward effect, Figure 6e is the backward effect, Figure 6f It is the effect of squatting, Figure 6g is the effect of walking with knees bent, and Figure 6h is the effect of jumping

The present invention can be applied to applications that require human body reconstruction in the field of virtual reality, and the application should not require excessive accuracy for human body reconstruction, such as VR virtual reality social applications, multi-person applications, and the like.

The above specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The scope of protection of the present invention is subject to the claims and is not limited by the above specific implementation. Each implementation within the scope is bound by the invention.

Claims (10)

1. A real-time whole-body posture reconstruction method based on the position and posture of the head and both hands, characterized in that the upper body reconstruction is carried out using an inverse dynamics scheme, that is, according to the head position, the waist position and the shoulder position are first estimated, and then according to the input The position of the hands is estimated by the reverse dynamics method to obtain the position of the elbows; then the lower body is reconstructed through the preset animation blending method, and finally the positions of all joints are mapped to the reconstructed character's upper and lower body in turn to complete the reconstruction of the whole body posture ; The upper body reconstruction refers to: after the calibration is completed by the virtual reality controller, the average positive direction of the head and the average direction of the vertical bisector of the line connecting the hands are obtained as feature values by using the data collected in the sensor recording and buffering process to calculate , the angle between the line connecting the two hands and the positive direction of the body, and the angle between the line connecting the two hands and the vertical plane where the head is located, and obtain the current positive direction of the human body; then calculate the waist height, the absolute position of the waist, the left and right The position of the right shoulder and the position of the elbows are calculated according to the inverse dynamics method. 2. The method according to claim 1, wherein the reconstruction of the lower body refers to: the instantaneous velocity of the head and the swing posture of both arms are calculated by using the data collected in the process of sensor recording and buffering, so as to The preset lower body animation is blended, and the lower body reconstruction is completed after legality testing. 3. The method according to claim 1, characterized in that, the current positive direction of the human body, that is, the direction in which the chest of the person is facing, is obtained by the following method: 3.1) Calculate the average positive direction of the head separately The average direction of the perpendicular bisector of the line connecting the two hands Specifically: in: Refers to the average positive direction of the head accumulated before this frame, hf _p refers to the projection of the positive direction of the head on the xz plane in the world coordinate system, n refers to The statistics are the average positive direction of the head of n frames; in: Refers to the positive direction of the vertical bisector of the line connecting the two hands accumulated before this frame, lhp and rhp respectively refer to the positions of the left hand and right hand in this frame, (lhp-rhp) _p refers to the line between the two hands in world coordinates The projection on the xz plane in the system, n refers to The statistics are the average positive direction of n frames, and up refers to the upward direction in the world coordinate system, that is, the direction (0, 0, 1); 3.2) Further calculate the angle a _cross between the line connecting the hands and the positive direction of the body and the angle a _linear between the line connecting the hands and the vertical plane where the head is located, specifically: Among them: hp represents the position of the head in this frame, lhp and rhp respectively represent the positions of the left hand and right hand in this frame, and the subscript p represents the projection of the vector on the xz plane in the world coordinate system; 3.3) According to a _cross and a _linear and and Determine the current positive direction of the human body. 4. The method according to claim 3, wherein said step 3.3 specifically comprises: 3.3.1) When a _cross is close to 90° or a _linear is close to 0°, the positive direction of the body in this frame 3.3.2) When and The difference between is not big, then the positive direction of the body in this frame otherwise 3.3.3) When in recent frames, the head is in an unstable state, but the hands are in a stable state, the average positive direction of the head It will no longer be reliable. It has been observed that in this case, most of the users are looking around in a static state. Therefore, the positive direction of the body is not arbitrarily changed at this time, that is, the positive direction of the body in this frame is equal to that in the previous frame. The positive direction of the body, bf _c = bf _l ; 3.3.4) When the head and both hands are in an unstable state in recent frames, there is no stable input to judge the positive direction of the human body. At this time, the previous positive direction of the human body is still maintained, that is, in this frame The positive direction of the body is equal to the positive direction of the body in the previous frame, bf _c =bf _l . 5. The method according to claim 3 or 4, characterized in that linear interpolation is performed on the current positive direction of the human body bf _a to ensure that the positive direction of the human body will not cause a sudden change due to an unstable input signal, that is, bf _a =bf _l *(1-f)+bf _c *f, wherein: f is a mixing parameter. 6. The method according to claim 1, wherein the height of the waist, the absolute position of the waist, and the positions of the left and right shoulders are obtained in the following manner: 4.1) Calculate the waist height according to the position of the head, namely Among them: h _h is the height of the head, h _m is the height of the character model that needs to be mapped to the virtual scene, m _uh refers to the distance from the head of the character model to the waist, h _mh is the minimum head that the character model can squat to height to prevent the waist from being lower than the feet; 4.2) Calculate the absolute position be of the waist according to the horizontal position of the head; preferably, use interpolation to slow down the tiny movement of the head, and use the interpolated horizontal position of the head to participate in the calculation; 4.3) Calculate the left shoulder position lse and right shoulder position rse respectively according to the absolute position be of the waist, specifically: lse=me-offset; rse=me+offset, wherein: factor _s is the adjustment parameter of the shoulder height, and bf is the adjustment parameter in step The positive direction of the body calculated in the third step, factor _w is the adjustment parameter of shoulder width, me=be+(hp-be)/factor _s ; 7. The method according to claim 6, characterized in that, when the user's eyes look down, the user may be looking at the things at his feet and want to bend down to pick it up; or the user's head has When accelerating in the vertical direction, the squatting state is activated, that is, the position of the waist in the horizontal direction is frozen; and when the head no longer has vertical acceleration, the offset between the head and the waist is recorded, Therefore, it is ensured that the position of the waist can be accurately inferred when the user walks in a squatting state; when the user stands up again, that is, when the head-mounted display reaches the calibrated height, the offset is reset to zero. 8. The method according to claim 2, characterized in that the preset animation of the lower limbs refers to: preparing several cyclic animations in advance, including: forward, left forward, right forward, left, right, Backward left, backward right, backward; then according to the direction and speed of the user's movement, the animation mixing technology commonly used in China is used to mix the cyclic animation according to different weights. 9. The method according to claim 2, characterized in that the legitimacy detection is to detect the legitimacy of the mapped lower limb position, and judge whether the result of animation mixing is legal, that is, whether the lower limb has penetrated into the What happens inside other objects; the lower body is calculated using the same inverse kinematics as the upper body when it is not legal, to avoid errors where the foot is elongated when penetrating the ground or jumping up. 10. A system for realizing the method according to any one of the above claims, characterized in that it comprises: a sensor data cache and statistics module, a body forward direction estimation module, an inverse dynamics solution module, an upper body reconstruction module and a lower body reconstruction module, Among them: the sensor data cache and statistics module outputs the data from the sensor to the body positive direction estimation module and the lower body reconstruction module after the statistics are cached. The reconstruction module obtains the waist height, the absolute position of the waist, and the positions of the left and right shoulders according to the positive direction of the human body, and outputs them to the inverse dynamics solving module for upper body reconstruction. The loop animation is mixed, and the mixed result is output to the inverse dynamics solving module for legality detection and mapping of all joints.