CN104463788B - Human motion interpolation method based on movement capturing data - Google Patents

Abstract

The invention discloses a human body motion interpolation method based on motion capture data. Firstly, the loaded two-segment motion sequence files are analyzed and the coordinate system conversion operation is performed; then, based on the motion data in the world coordinate system, the characteristics of the front and rear positional relationship of both feet are extracted. and step timing features, and according to the law of human motion, according to the above two features, the long motion sequence is divided into short motion segments; finally, the most similar frame pair is selected in the same short segment and the key frame pair is determined based on this, Then based on the key frame, the quaternion spherical interpolation algorithm is used to perform angle rotation interpolation, and the linear interpolation method is used to perform translation interpolation of the root node, thereby connecting a new movement. The present invention determines the key frame pair based on the same short segment after division, ensures that the interpolation transition sequence conforms to the visual logic sequence of human eyes, and has good visual effect.

Description

Human motion interpolation method based on motion capture data

technical field

The invention belongs to the field of computer vision, and in particular relates to a human motion interpolation method based on motion capture data.

Background technique

With the rapid development of computer image technology and its related derivatives, motion capture technology has gradually become a means of data acquisition in the fields of virtual reality, computer vision, film and television production, game entertainment and computer animation. However, the price of the system is expensive; the site environment is strictly required when collecting data; animation production has high requirements for the quality of actors' movements, which makes the research on the reusability of motion data a very meaningful research direction. Reusability research is to use the motion data in the existing database to build a motion network through editing, fusion, splicing, synthesis, etc., to generate rich and varied new motion sequences, and to generate virtual actions that meet their respective needs, making computer animation production, virtual The work efficiency in various fields such as reality has been greatly improved, and production costs have been saved.

Keyframe interpolation is an important interpolation technique widely used in the research of human motion data reusability. Its basic principle is to first obtain or make several key frames in the animation sequence, and then use interpolation technology to generate intermediate transition frames based on the key frames. The extraction of key frames mainly considers two aspects of motion features and key frame quantitative analysis. Since human motion contains multiple degrees of freedom and is motion data in high-dimensional space, the traditional method directly analyzes and extracts features from the original data, and there is a problem of excessive calculation. Therefore, the current method is to reduce the dimensionality of the original data first, and extract key frames from the dimensionality-reduced data, which can greatly reduce the amount of calculation. Key frame interpolation technology is to directly generate intermediate frames by using interpolation algorithm given several key frames. Common interpolation algorithms include: linear interpolation, quaternion interpolation, cubic spline interpolation, and double interpolation. The representative ones using key frame interpolation technology are: Ashraf and Wong use bilinear interpolation method to generate new motion from more than two given motions. Rose et al. proposed an inverse kinematics approach incorporating motion constraints. By dividing the human body model into three different parts, a new human body motion is synthesized by frame interpolation. The quaternion spherical interpolation technology maps the rotation of the quaternion to the unit four-dimensional sphere, gradually reduces the angle between the two quaternions, and gradually transitions between the given two poses to generate a new pose, which is often used in key frames In interpolation research. However, in the production of 3D human body animation, due to the complexity of the 3D human body model, high-dimensional motion data, and the sensitivity of human vision to human body motion, relying on pure interpolation algorithms can reduce the error of key frame pair extraction and generate highly realistic images using interpolation algorithms. Sports are challenging and often require a combination of kinematic or other knowledge.

Contents of the invention

The purpose of the present invention is to provide a human motion interpolation method based on motion capture data, which is a new key frame interpolation method based on feature analysis, motion cycle segmentation, and transition frame pair matching.

The technical solution adopted in the present invention is, a human body motion interpolation method based on motion capture data, firstly analyze the loaded two-segment motion sequence file (BVH file), and perform coordinate system conversion operation; then based on the world coordinate system Motion data, extracting the characteristics of the front and rear position relationship of the feet and the timing characteristics of the step distance, and according to the law of human motion, according to the above two characteristics, the long motion sequence is divided into short motion segments; finally, the most similar motion sequence is selected in the same short segment Frame pairs and determine the key frame pairs based on this, and then use the quaternion spherical interpolation algorithm to perform angle rotation interpolation based on the key frame pairs, and use the linear interpolation method to perform translation interpolation of the root node, thereby connecting a new movement.

The present invention is also characterized in that,

Specifically include the following steps:

Step 1, load and analyze the motion sequence file (BVH file), and convert the relative position information of the motion data of the BVH file in the local coordinate system into the absolute position information in the world coordinate system;

Step 2, based on the absolute position information in the world coordinate system obtained in step 1, extract the spatial position relationship features and temporal relationship features of the joint points, and use the front and rear position relationship features and step distance of the feet according to the periodic law of human motion The timing feature divides the long motion sequence into multiple short motion sequences;

Step 3, based on the segmentation result obtained in step 2, align the segmented short segments based on the time frame sequence, and determine the frame pair with the shortest Euclidean distance as the most similar pose in the same short segment, so as to determine the key frame pair ;

Step 4, based on the key frame pair in step 3, use the quaternion spherical interpolation algorithm to generate the angle rotation value of the intermediate transition frame, and use the linear interpolation algorithm to generate the translation value of the root node intermediate transition frame.

In step 1, the motion sequence file is composed of two parts: the skeleton part and the motion data; firstly, the human skeleton part is parsed by the token passing method: by gradually reading each keyword, integer character, floating-point character, String parsing, and then parse the motion data part according to the sequence of the skeleton structure;

Use a recursive method to find the absolute position information of each joint point of the human body motion data in the world coordinate system; the conversion formula is as follows (1):

p _i ^(j) = T _i ^(root) R _i ^(root) ... R _i ^(k) ... p ₀ ^(j) (1)

Among them, p _i ^(j) represents the coordinates of the joint point N _j in the world coordinate system at the i-th moment of the motion sequence; T _i ^(root) and R _i ^(root) represent the translation and rotation transformation matrices of the root node respectively; R _i ^(k) represents the rotation transformation matrix of the joint N _k relative to its direct parent node in the skeleton structure; N _k is any node from the root node to the node N _j in the tree-shaped human skeleton; p ₀ ^(j) represents the initial time , the offset of N _j in the local coordinate system of its parent node.

In step 2, according to the nature of human body movement in a periodic law, especially for mobile sports, no matter what the style is, the feet move forward alternately; based on this law, use the forward spatial position relationship of the feet The features and step timing features are used to segment long motion sequences; the two segmentation functions are expressed as follows:

Among them, the Pace_changed function indicates whether the forward step distance of both feet at a certain moment changes from increasing to decreasing, or from decreasing to increasing. If so, assign the function value to 1, indicating that the step distance at this moment is the The maximum or minimum step distance of a short segment is to define this moment as the segmentation point of this segment of motion; otherwise, it is assigned a value of 0, indicating that this moment is a non-segmentation point of a short segment; the Front_foot function indicates whether the right foot at a certain moment In front of the left foot, when the right foot is in front, assign a value of 1, otherwise assign a value of 0.

In step 3, based on the segmentation results obtained in step 2, use the formula (4) to align the time frame sequence in the short segment to obtain matching frame pairs: f ₁ and f ₂ , f ₁ ^′ and f ₂ ^′ represent the two segments of Start and end frames, calculate the matching frame pair (f _i , f _i ′); then select the most similar frame pair as the key frame pair based on the matching frame pair calculation; use the most commonly used Euclidean distance to determine the most similar frame Yes, assuming that the calculated minimum value of the Euclidean distance is D(P _i , P _j ), then select the next frame as a key frame pair (P _i , P _j+1 ); among them, the formula for determining the minimum value of the Euclidean distance As shown in (5) below:

Among them, p _i ^k , p _j ^k represent the position information of the k-th joint point in the i-th frame and the j-th frame of the two motion sequences respectively; w _k represents the weight of the k-th joint point in the human skeleton, generally In other words, the weight of the node closer to the root node is greater.

In step 4, use formula (6) and formula (7) to convert Euler angle rotation angle and quaternion: Euler angle group is the rotation angle around Z, Y, X, and the corresponding quaternion after conversion is q=[wabc];

Use the formula (8) to generate the transitional rotation angle based on the quaternion spherical interpolation algorithm: p ₀ and p ₁ are the rotation quaternions of a joint point in two key frames, Ω is the angle difference, t is the interpolation parameter, and To control the smooth transition speed during the interpolation process; as the value of t changes, change the interpolation angle, when t is close to 1, the angle rotation of interpolation p is closer to p ₁ ; when t is close to 0, the angle rotation of interpolation p is closer to p ₀ ;

Use formula (9) to perform linear interpolation based on root node translation information: where, the root node coordinates of the start frame and end frame are C ₀ , C ₁ , respectively, and u is an interpolation parameter. When u is close to 0, it is close to C ₀ , close to 1 is close to C ₁ ;

C _i (x _i ,y _i ,z _i )=uC ₀ (x ₀ ,y ₀ ,z ₀ )+(1-u)C ₁ (x ₁ ,y ₁ ,z ₁ ) (9).

The beneficial effect of the present invention is that the human body motion interpolation method based on the motion capture data of the present invention extracts and divides the long motion sequence file (BVH file) into a plurality of short motion sequence segments according to the feature based on the periodic law of moving human body motion. The key frame pair is determined based on the segmented short segment, and a long motion sequence segment is interpolated and connected by using a quaternion spherical interpolation algorithm and a linear interpolation algorithm. The present invention determines the key frame pair based on the same short segment after division, ensures that the interpolation transition sequence conforms to the visual logic sequence of human eyes, and has good visual effect.

The invention has important significance for the reusability research of motion capture data. Compared with the traditional key frame interpolation method, the selection of the key frame pair in the present invention avoids the error of unreasonable interpolation sequence, and the motion of the interpolation connection can be more natural and seamless compared with the traditional interpolation method.

Description of drawings

Fig. 1 is the structural diagram of human skeleton;

Fig. 2 is a flow chart of parsing a BVH file;

Fig. 3 is the human body posture of the 102nd frame of walking motion under the local coordinate system;

Fig. 4 is the human body posture of the 102nd frame of walking motion under the world coordinate system;

Fig. 5 is a posture sequence diagram of one cycle of walking motion;

Fig. 6 is a schematic timing diagram of walking motion;

Fig. 7 is the time-series relationship diagram of the forward step distance of walking motion;

Fig. 8 is a sequence diagram after the walking motion is divided according to the segmentation point;

Fig. 9 is a schematic timing diagram of walking motion segmentation;

Fig. 10 is a schematic diagram of quaternion spherical interpolation;

Fig. 11 is a sequence diagram of key frame pairs and interpolation gestures of the present invention;

Fig. 12 is a key frame pair and interpolation posture sequence diagram of the traditional method;

Fig. 13 is a sequence diagram of an original walking posture;

Fig. 14 is a time sequence diagram of an original striding motion posture;

Fig. 15 is a sequence diagram of motion interpolation connection for walking and striding.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

The invention is a human motion interpolation method based on feature analysis, motion segmentation, and key frame pair matching, and combines the feature analysis and motion segmentation results to determine the key frame pair. The high-dimensional problem of human motion sequence is solved by feature extraction and dimensionality reduction. The present invention proposes for the first time the use of a coordinate system conversion method for feature extraction, and achieves the purpose of accurately extracting human body motion features while reducing the dimensionality of high-dimensional human body motion sequences. In addition, the present invention determines key frame pairs based on the segmented short segments to ensure the order of human vision in the interpolation transition.

The human body motion interpolation method based on the motion capture data of the present invention first parses the loaded two-segment motion sequence files (BVH files) and performs feature analysis; then, according to the periodic law of human motion movement, the long motion sequence is segmented, and based on the segmented Multiple short motion sequence segments are aligned in frame timing to obtain matching frame pairs, and the most similar pose of the same short segment is determined from the matching frame pairs to determine key frame pairs; finally, quaternion spherical interpolation is used based on the rotation angle data of the motion sequence Algorithm, based on the translation data of the motion sequence, the linear interpolation algorithm is used to generate the transition motion sequence and connect the key frame pairs to generate a new long motion sequence.

It mainly includes the following steps:

Step 1, load and analyze two motion sequence files (BVH files). The motion sequence file (BVH file) is stored in text mode and consists of two parts: skeleton part and motion data. The skeleton structure of the BVH file is not uniform, but the structure is similar. Figure 1 is a typical example: the skeleton is stored in a tree structure, the middle hip node (Hip) of the human body is the root node, the upper back node (Upperback), left and right The hips (L_Hip, R_Hip) are its child nodes, and nodes such as the upper back, left and right hips also have their own child nodes, and so on until the end point. The motion data stores the translation information of the root node and the angular rotation information of all related nodes relative to their parent nodes in the order of the skeleton structure. The present invention analyzes the loaded motion sequence file (BVH file), and the analysis process is as shown in Figure 2 below: load the BVH file, if the file is not empty, first analyze the bone part of the BVH file to obtain the frame structure, initial posture and the number of frames of the BVH file and frame frequency; then analyze the data block part of the BVH file based on the skeleton structure, total frame number, and frame frequency, and calculate the offset and rotation of the joint points of each frame until the last frame; finally, based on the information of the data block part, draw each Frames of human motion poses.

However, because each joint point in the human motion sequence is stored in its own relative coordinate system, it is not convenient for analysis. In order to accurately and efficiently extract the spatial position relationship characteristics and timing characteristics of the joint points in the motion sequence, the present invention proposes for the first time that the motion sequence The relative motion information of the joint points in the local coordinate system is converted into the absolute position information in the world coordinate system. The formula for transforming the relative position information of the N _j joint points of the human skeleton from the local coordinate system to the world coordinate system is shown in (1): p _i ^(j) represents the world coordinates of the joint point N _j at the i-th moment of the motion sequence The coordinates of the system; T _i ^(root) and R _i ^(root) represent the translation and rotation transformation matrix of the root node respectively, and R _i ^(k) represents the joint N _k in the skeleton structure (N _k is the tree-shaped human skeleton, starting from the root Any node between node and node N _j ) relative to the rotation transformation matrix of its direct parent node; p ₀ ^(j) represents the initial offset of N _j in the local coordinate system where its parent node is located.

p _i ^(j) = T _i ^(root) R _i ^(root) ... R _i ^(k) ... p ₀ ^(j) (1)

Fig. 3 is the 102nd frame of the walking motion before the conversion, and Fig. 4 is the 102nd frame of the walking motion after the conversion, and they are the same postures by comparison, which shows that the conversion formula adopted by the present invention is effective.

According to the absolute position information in the world coordinate system obtained in step 1, the spatial position relationship and time series characteristics of the joint points are extracted, and according to the periodic law of human motion, the movement is divided into multiple short segments by using the front and rear position relationship characteristics of the feet ;

Because most of the human body motions are periodic motions, especially the motions of movement, both feet are periodically exchanged to walk forward. After the present invention has analyzed the law of human body movement, utilizes for the first time the relationship formula of whether the forward step distance of both feet shown in the following formula (2) increases and the relationship formula of the front and back positions of both feet shown in formula (3), to convert a period of motion Divided into four short motion segments;

Among them, Pace_changed in formula (2) is based on the function to judge whether the forward step distance of both feet at a certain moment is changed from increasing to starting to decrease, or from decreasing to starting to increase. If so, assign the function value to 1 , indicating that the step distance at this moment is the maximum or minimum step distance of the short segment, that is, this moment is defined as the segmentation point of the segment; otherwise, it is assigned a value of 0, indicating that this moment is a non-segmentation point of a short segment. Front_foot in formula (3) indicates whether the right foot is in front of the left foot at a certain moment according to the function. If the right foot is in front, it is assigned a value of 1, otherwise it is assigned a value of 0.

The following describes the segmentation process according to the specific walking motion: Figure 5 is a timing diagram of the input walking motion posture. Figure 6-Figure 8 illustrates the principle of motion segmentation, and Figure 9 is the motion effect after segmentation. FIG. 6 is a timing diagram of the walking motion sequence. In order to facilitate the description of the segmentation process, the frame timing is represented by a grid. Fig. 7 is a diagram showing the relationship between the forward step distance of both feet and the frame timing of the walking movement, in which the four moments A, B, C, and D are the moments of the maximum and minimum step distances of the walking movement sequence in one cycle, as the movement Sequence split point. The schematic diagram of the walking motion sequence after segmentation is shown in Figure 8: the numbers in the squares represent the sequence numbers of the segmented motion segments. It can be seen from the figure that the human motion sequence of each cycle is divided into 4 different motion sequence shorts. part. Fig. 9 is a motion sequence diagram represented by the sequence of arrows according to the segmentation time points. The motion posture of the human body in the figure is the posture of the segmentation time points, and the motion sequence is represented according to the sequence of arrows according to the segmentation points. It can be seen from the figure that the segmentation time point is determined accurately, and a walking movement is segmented in line with human visual logic.

Step 3, because the number of frames in the segmented short segment is different, in order to accurately determine the key frame pair, the present invention first performs an alignment operation based on the timing of time frames in the same short segment, and uses the distance formula to determine the most similar frame pair in the same short segment . If the most similar frame pair is (P _i , P _j ), then the transition frame pair is determined as: (P _i , P _j+1 ).

Suppose the start frame and end frame of m short segment are f ₁ and f ₂ respectively, and the start frame and end frame of n short segment are f ₁ ′ and f ₂ ′ respectively, then calculate m and n according to the following formula (4) Matching frame pair (f _i , f _i ') for the motion segment.

Based on the matched frame sequence pair, calculate the distance of the corresponding frame according to the following formula (5) to determine the most similar frame pair:

Among them, p _i ^k and p _j ^k represent the position information of the kth joint point in the i-th frame and the j-th frame of the two motion sequences respectively. w _k represents the weight of the kth joint point in the human skeleton. Generally speaking, the weight of the node closer to the root node is greater.

Figure 11 is a timing diagram of the transition frame sequence generated by the interpolation algorithm to determine the key frame pair by the traditional method: the right foot of the human body in the first two poses in the figure represents the sequence of stretching forward, but the pose of the key frame pair in the traditional method shows the right foot of the human body. Foot back to insert frame. Therefore, the interpolation sequence does not match the original human motion sequence. Fig. 12 is a key frame pair determined by the method of the present invention, and a sequence diagram of transition frames generated by an interpolation algorithm: the second posture and the last posture from the left are the key frame pairs determined by the present invention, and the left foot of the human body in the figure is formed by The step-by-step sequential transition interpolation from back to front is in line with the visual logic of the human eye. It shows that the method for determining the key frame pair of the present invention avoids the error of unreasonable interpolation sequence.

Step 4: Given two segments of motion, based on the key frame pair obtained in Step 3, use the quaternion spherical interpolation algorithm to generate the transition value of the angle rotation, and use the linear interpolation algorithm to generate the transition value of the translation information, thereby connecting a new long motion sequence .

_Quaternion spherical interpolation is widely used in the study _of rigid body rotation due to its simple principle and other advantages. , the angle between the two frames is Ω, and the interpolation formula is used to insert the frame P so that the joint point k gradually transitions from point P ₀ on the spherical surface to point P ₁ on the spherical surface step by step along the spherical surface. The quaternion spherical interpolation formula is shown in (8): t is a parameter of the interpolation formula, which is mainly used to control the speed of the smooth transition of interpolation frames during the interpolation process. When t is close to 1, the rotation angle of the inserted frame P is close to P ₁ , corresponding to the value of θ in FIG. 4 close to Ω. When t is close to 0, the rotation angle of interpolation P is close to P0, corresponding to the value of θ in Figure 4 close to 0.

Linear interpolation is the most commonly used technique, and it is an interpolation algorithm with high computational efficiency. For the translation information of the root node, the present invention adopts a linear interpolation method, and the formula is shown in (9) below. The root node coordinates of the start frame and the end frame are C ₀ and C ₁ respectively, and u is an interpolation parameter. When u is close to 0, it is close to C ₀ , and when it is close to 1, it is close to C ₁ .

C _i (x _i ,y _i ,z _i )=uC ₀ (x ₀ ,y ₀ ,z ₀ )+(1-u)C ₁ (x ₁ ,y ₁ ,z ₁ ) (9)

Human body animation is the most challenging research in the field of animation. On the one hand, human eyes are very sensitive and direct in judging the most familiar sports, and any flaws will become particularly obvious. On the other hand, due to the complex structure of the human skeleton and the high-dimensional human motion sequence, it is a great challenge to accurately analyze the human motion data. Therefore, so far there is no objective standard that can replace the subjective perception of the human eye to measure the quality of motion interpolation. Fig. 13 is a timing diagram of a motion sequence of a normal walk. Fig. 14 is a timing diagram of a motion sequence of striding. Fig. 15 is a new walking motion sequence after interpolation and connection of two motion sequences. It can be seen from the figure that the walking movement is interpolated and connected to the stride walking movement, which naturally and seamlessly connects the two motion sequences together.

Claims (3)

1. a kind of human motion interpolation method based on movement capturing data, it is characterised in that parse two sections of fortune of loading first Dynamic sequential file, and carry out coordinate system conversion operation；The exercise data being then based under world coordinate system, extract double-legged anteroposterior position Relationship characteristic and step pitch temporal aspect are put, and according to human motion rule, two features according to more than, by long motion sequence It is divided into short motor segment；Most like frame pair is finally chosen in same short section and is based on this determination key frame pair, is then based on Key frame carries out the flat of root node to carrying out angle rotation interpolation with quaternary number sphere interpolation algorithm with linear interpolation method Interpolation is moved, so as to connect into one section of new motion；

Specifically include following steps：

Step 1, loading parsing motion sequence file, and the relative position to the exercise datas of BVH files under local coordinate system Information is converted to the absolute location information under world coordinate system；

Motion sequence file is made up of two parts in step 1：Skeleton part and exercise data；Parsed first in alternative space method Human skeleton part：By progressively reading each keyword in motion sequence file, shaping character, floating-point ocra font ocr, word Symbol string parsing, then parses exercise data part according to skeleton structure order；

Absolute location information of each artis of human body movement data under world coordinate system is obtained using recursive mode；Conversion is public Shown in formula following (1)：

p_i ^(j)=T_i ^(root)R_i ^(root)...R_i ^(k)...p₀ ^(j) (1)

Wherein, p_i ^(j)Represent the i-th moment artis N of motion sequence_jThe coordinate in world coordinate system；T_i ^(root),R_i ^(root)Point Not Biao Shi root node translation and rotational transformation matrix；R_i ^(k)Represent joint N in skeleton structure_kWith respect to the rotation of its direct father node Turn transformation matrix；N_kFor in tree-like human skeleton, from root node to node N_jBetween arbitrary node；p₀ ^(j)When representing initial, N_j Offset under the local coordinate system of its father node；

Step 2, the absolute location information under world coordinate system obtained based on step 1, the locus for extracting artis are closed It is the feature of feature and sequential relationship, and according to human motion periodic regularity, using double-legged front and back position relationship characteristic and step Long motion sequence carried out away from temporal aspect to be divided into multiple short motion sequences；In step 2, according to human motion in periodically The property of rule, all it is that both feet are alternately advanced forward forward regardless of style especially for the motion of mobile class；Base It is according to the long motion sequence of segmentation with double-legged forward spatial position relationship feature and step pitch temporal aspect in this rule；Two Individual segmentation function represents as follows respectively：

<mrow> <mi>P</mi> <mi>a</mi> <mi>c</mi> <mi>e</mi> <mo>_</mo> <mi>c</mi> <mi>h</mi> <mi>a</mi> <mi>n</mi> <mi>g</mi> <mi>e</mi> <mi>d</mi> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mi>i</mi> <mi>f</mi> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <mrow> <mi>D</mi> <mi>i</mi> <mi>s</mi> <mi>t</mi> <mo>_</mo> <mi>f</mi> <mi>e</mi> <mi>e</mi> <mi>t</mi> </mrow> </mtd> <mtd> <mrow> <mi>c</mi> <mi>h</mi> <mi>a</mi> <mi>n</mi> <mi>g</mi> <mi>e</mi> <mi>d</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>

<mrow> <mi>F</mi> <mi>r</mi> <mi>o</mi> <mi>u</mi> <mi>n</mi> <mi>t</mi> <mo>_</mo> <mi>f</mi> <mi>o</mi> <mi>o</mi> <mi>t</mi> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mi>i</mi> <mi>f</mi> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <mrow> <mi>r</mi> <mi>i</mi> <mi>g</mi> <mi>h</mi> <mi>t</mi> <mo>_</mo> <mi>f</mi> <mi>o</mi> <mi>o</mi> <mi>t</mi> </mrow> </mtd> <mtd> <mrow> <mi>f</mi> <mi>r</mi> <mi>o</mi> <mi>u</mi> <mi>n</mi> <mi>t</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>

Wherein, whether step pitch reduces the both feet at Pace_changed function representations a certain moment being transformed to by increase forward, Or increase being transformed to by reduction, if it is, functional value is entered as into 1, represent the maximum that this moment step pitch is the short section Step pitch or minimum step, i.e., be defined as to the cut-point of this section motion this moment；Otherwise 0 is entered as, represents that this moment is short for certain The non-cut-point of section；Whether the right crus of diaphragm at Frount_foot function representations a certain moment is in front of left foot, when right crus of diaphragm is in front, 1 is entered as, is otherwise entered as 0；

Step 3, the segmentation result obtained based on step 2, the short section after segmentation is carried out to be based on time frame sequence alignment, and same In one short section, the most short frame of Euclidean distance is determined to for most like posture, so that it is determined that key frame pair；

Step 4, the key frame pair based on step 3, middle transition frame angle rotational value is generated with quaternary number sphere interpolation algorithm, Root node middle transition frame shift value is generated with linear interpolation algorithm.

2. the human motion interpolation method according to claim 1 based on movement capturing data, it is characterised in that step 3 In, the segmentation result that is obtained based on step 2, formula (4) carries out the time frame sequence alignment in short section, obtains matching frame It is right：f₁With f₂、f₁' and f₂' respectively represent two sections starting and end frame, be calculated matching frame to (f_i,f_i′)；It is then based on Match frame and choose most like frame to as key frame pair to calculating；Most like frame is determined using the most frequently used Euclidean distance It is right, it is assumed that the Euclidean distance minimum value being calculated is D (P_i,P_j), then selection next frame is key frame to i.e. (P_i,P_j+1)；Its In, determine shown in Euclidean distance minimum value formula following (5)：

<mrow> <msup> <msub> <mi>f</mi> <mi>i</mi> </msub> <mo>&amp;prime;</mo> </msup> <mo>=</mo> <msup> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>&amp;prime;</mo> </msup> <mo>+</mo> <mfrac> <mrow> <msup> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>&amp;prime;</mo> </msup> <mo>-</mo> <msup> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>&amp;prime;</mo> </msup> </mrow> <mrow> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>-</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> </mrow> </mfrac> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>

<mrow> <mi>D</mi> <mrow> <mo>(</mo> <msub> <mi>P</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>P</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>w</mi> <mi>k</mi> </msub> <mo>|</mo> <mo>|</mo> <msup> <msub> <mi>p</mi> <mi>i</mi> </msub> <mi>k</mi> </msup> <mo>-</mo> <msup> <msub> <mi>p</mi> <mi>j</mi> </msub> <mi>k</mi> </msup> <mo>|</mo> <mo>|</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>

Wherein, p_i ^k, p_j ^kRepresent two sections of motion sequences, k-th of artis respectively in the positional information of the i-th frame and jth frame；w_kRepresent K-th of artis weight shared in human skeleton, in general, the node weights near from root node are bigger.

3. the human motion interpolation method according to claim 1 based on movement capturing data, it is characterised in that step 4 In, formula (6) and formula (7) carry out the Eulerian angles anglec of rotation and mutually changed with quaternary number：Eulerian angles groupFor around Z, Y, the X anglec of rotation, corresponding quaternary number is q=[w a b c] after changing；

Formula (8) is carried out based on the quaternary number sphere interpolation algorithm generation transition anglec of rotation：p₀、p₁For two key frames The rotation quaternary number of certain artis, Ω are its differential seat angle, and t is interpolation parameter, for controlling the speed seamlessly transitted in Interpolation Process Degree；With the change of t values, change interpolation angle, when t is close to 1, interpolation p angle rotation is closer to p₁；When t is close to 0, Interpolation p angle is rotated close to p₀；

<mrow> <mi>p</mi> <mo>=</mo> <mi>S</mi> <mi>L</mi> <mi>E</mi> <mi>R</mi> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mn>0</mn> </msub> <mo>;</mo> <msub> <mi>p</mi> <mn>1</mn> </msub> <mo>;</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>s</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>t</mi> <mo>)</mo> </mrow> <mi>&amp;Omega;</mi> </mrow> <mrow> <mi>s</mi> <mi>i</mi> <mi>n</mi> <mi>&amp;Omega;</mi> </mrow> </mfrac> <msub> <mi>p</mi> <mn>0</mn> </msub> <mo>+</mo> <mfrac> <mrow> <mi>sin</mi> <mi> </mi> <mi>t</mi> <mi>&amp;Omega;</mi> </mrow> <mrow> <mi>s</mi> <mi>i</mi> <mi>n</mi> <mi>&amp;Omega;</mi> </mrow> </mfrac> <msub> <mi>p</mi> <mn>1</mn> </msub> <mo>,</mo> <mi>t</mi> <mo>&amp;Element;</mo> <mo>&amp;lsqb;</mo> <mn>0</mn> <mo>,</mo> <mn>1</mn> <mo>&amp;rsqb;</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>8</mn> <mo>)</mo> </mrow> </mrow>

Formula (9) carries out the linear interpolation based on root node translation information：Wherein, start frame and the root node of end frame are sat Mark is C respectively₀、C₁, u is interpolation parameter, when u is close to 0 close to C₀, close to when 1 close to C₁；

C_i(x_i,y_i,z_i)=uC₀(x₀,y₀,z₀)+(1-u)C₁(x₁,y₁,z₁) (9)。