JP2019159348A - Skinning decomposition acceleration method considering locality of weight map and skinning decomposition acceleration program of the same - Google Patents

Abstract

To provide a skinning decomposition acceleration technique of conversion with low memory consumption and high speed and high accuracy.SOLUTION: A skinning decomposition acceleration method applies an initial process that performs a cluster allocation processing unit 12 allocating a cluster of a bone for operating an object by applying different methods according to characteristics of the object, a system matrix reduction processing unit 13 clustering the cluster of the neighbor bone for the set bone cluster to reduce a size of a system matrix of two dimensional plan problem in a weight map update calculation so as to maintain it to a constant size, and an operation number reduction processing 14 introducing an evaluation function representing posture similarity between the objects in an animation and setting the object in a key frame in repetition operation or a resting state to a weight map update calculation target only once so as to limit a weight map update calculation number.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a skinning decomposition speed-up method in consideration of locality of weight maps in a skinning decomposition technique for converting an animation of a three-dimensional object (person, animal, object, etc.) created on a vertex basis into a 3D skin animation, and the same skinning. It relates to a high-speed decomposition program.

  Skinning (processing) is one of the techniques for deforming a three-dimensional object. For a stationary object, a bone (control point) for manipulating the object and a weight map indicating how much the bone affects each vertex are set. Thereby, the whole object can be deformed by manipulating the posture of the bone. For skinning processing of CG characters in recent real-time applications, a linear blend skinning (LBS) method is widely used. Furthermore, techniques for improving the quality of skin animation by introducing redundant bones called auxiliary bones and expressing complex skin deformations such as skin and muscle bulges have been put into practical use. In general, the number of bones is much smaller than the number of vertices of the object, so rather than memorizing the object's vertex coordinates for every frame in the animation, remember the object's weight map and the bone's pose for each frame. It is more efficient from the viewpoint of memory consumption. In addition, the skinning process has a low calculation cost and can be accelerated by hardware. However, in order to perform the skinning process, it is necessary to appropriately set bones, corresponding weight maps, and postures of the bones in each frame, which is a heavy burden on the artist.

Therefore, there is a high demand for a technology for automatically converting vertex-based animation into skin animation that consumes less memory and has a low computation cost. The technique for converting vertex-based animation to skin animation is called skinning decomposition. Representative examples of this skinning decomposition include James et al. Skinning Mesh Animations, Hasler et al. Learning Skeletons for Shape and Pose, Li et al., Smooth Skinning Decomposition SS, and DR. From the viewpoint of sex, it is one of the methods that are currently attracting the most attention.

  Skinning decomposition by SSDR solves the inverse problem of the linear blend skinning method, and receives a set of example data of an object and outputs skinning parameters (the posture of each bone and weight map) that best approximates each example data. . Specifically, the shape of the object for each frame of the object animation is input, and a rigid transformation matrix of each bone that best approximates the shape is calculated for each frame. However, the weight map calculates one parameter for the entire animation. As shown in FIG. 3A, the SSDR includes three processes: a clustering initialization process, a bone rigid body transformation matrix update process (Update Bone Transformations) process, and a weight map update process (Update Bone Transformations) process. Consists of. First, the initial value of the bone rigid transformation matrix is calculated by clustering similar moving parts of the example data of the object. Next, the final skinning parameters are calculated by repeatedly updating the weight map and the bone rigid body transformation matrix using the Block Coordinate Descent method.

  Another method for deforming a three-dimensional object is a blend shape. This is a technique for creating a new shape by preparing a plurality of models having the same number of vertices and connection information between vertices but different shapes and blending each shape. It is widely used especially when creating facial (facial expression) animations. However, the blend shape has a problem that it requires a large amount of calculation (to operate in real time) as compared with skinning. Thus, there is an increasing demand for technology for converting animations created with blend shapes into skin animations.

  In addition, the following patent documents 1 and 2 are mentioned as a literature in which the conventional skinning process was described, and the following non-patent literature 1 is mentioned as a non-patent literature in which SSDR and skinning decomposition | disassembly were described.

JP 2002-63595 A JP 2004-334661 A

Binh Hul Le, Zhigan Deng, “Smooth Skinning Decomposition with Rigid Bones”, University of Heathton, 2012, [December 25, 2016, TP / S. CS. UH. EDU / BLE / PAPERS / 2012SA-SSDR / 2012_SA_SSDR_PREPRINT. PDF

The above-described skinning decomposition by SSDR makes it possible to calculate a skinning parameter with less error for an object whose movable part is determined by a skeleton or the like. Etc.) has a problem that the accuracy is reduced. Furthermore, SSDR does not target animations created with blend shapes.

  Further, the above-described skinning decomposition by SSDR has a problem that the amount of calculation increases exponentially as the number of bones to be set increases. This is because the size of the system matrix of the quadratic programming problem solved when calculating the weight map is proportional to the number of bones to be set. In particular, this problem appears remarkably when it is necessary to set a large number of bones, such as approximating an animation of a large deformation object (cloth, clothes, etc.) with high accuracy by skinning.

  A first object of the present invention is to provide a skinning decomposition speed-up method and a skinning decomposition speed-up program in consideration of the locality of a weight map that can be applied to animation and blend shapes of objects with large deformations.

  A second object of the present invention is to provide a skinning decomposition speed-up method and a skinning decomposition speed-up program in consideration of the locality of a weight map in which the amount of calculation does not increase explosively even when the number of bones to be set increases. That is.

To achieve the above object, the present invention provides a vertex-based animation for storing vertex coordinates of objects of all frames in an animation, and a skin animation for calculating a shape from the posture of a bone set for the object and a corresponding weight map. In addition, in the skinning decomposition acceleration method and program considering the locality of the weight map that is automatically converted using a computer,
To the computer,
Different methods for stationary objects depending on the characteristics of the object, such as objects whose movable parts are determined by the skeleton etc., such as animals, objects that undergo large deformations such as cloth, objects created with blend shapes, etc. Assign a cluster of bones to manipulate the object by applying [Cluster Assignment Process]
Only the neighboring bone clusters are clustered with respect to the bone cluster set in the [cluster assignment processing step], and the size of the system matrix of the quadratic programming problem in the weight map update operation is reduced to a constant size. Keep the [system matrix reduction processing step],
For the bone cluster set in the [Cluster Allocation Processing Step], an evaluation function expression representing the pose similarity between objects in the animation is introduced. After the initialization process of executing [calculation frequency reduction processing] that limits the number of weight map update calculations by setting the weight map update calculation target only once,
A weight map update step of updating the weight map of the stationary object by the initialization step;
A bone rigid body transformation matrix update step for updating a rigid transformation matrix of a bone of an object in the animation in which the weight map is updated by the weight map updating step;
The main feature is that it is repeatedly executed.
In particular, according to the present invention, in order to achieve the first object, three methods are selectively used according to the object to be processed in the [cluster allocation processing step] of the initialization step.
(1) For an object such as an animal whose movable part is determined by a skeleton or the like, an error value is calculated for each cluster, and a new cluster is assigned to a cluster whose error value is equal to or greater than a threshold value. However, the entire object is initially set as one cluster.
(2) Clustering is performed for an object that has a large deformation whose movable part is not determined, such as cloth, based on the result of determining the position of the bone using the farthest point sampling method (Farest Point Sampling).
(3) In blend shape application, clustering is performed based on preset bones (such as those set by an artist).
In order to achieve the second object, the present invention introduces a technique for reducing the amount of calculation of the weight map in the initialization process. In this weight map calculation amount reduction method, when calculating the weight map of each bone, the system matrix reduction process that approximates the solution using only the animation of the neighborhood of the bone instead of the animation of the entire object, When obtaining the weight map, the object shape of all frames is not used, but a frame having a similar shape is thinned out [calculation frequency reduction processing step].

  The skinning decomposition acceleration method and the skinning decomposition acceleration program considering the locality of the weight map according to the present invention can be applied to a wide range of animations by properly using three methods according to the object to be processed in the initialization process. On the other hand, it is possible to apply high-precision skinning decomposition. Specifically, by devising the allocation of clusters, skinning decomposition with higher accuracy than SSDR can be performed on an object such as an animal whose movable part is determined by a skeleton or the like. By using the farthest point sampling method (Farest Point Sampling), it is possible to perform highly accurate skinning decomposition even for an object that is deformed largely, such as a cloth, whose movable part is not fixed. In addition, by introducing a function of performing SSDR using a bone set in advance, it is possible to apply to a blend shape.

  In addition, the skinning decomposition acceleration method and skinning decomposition acceleration program considering the locality of the weight map according to the present invention introduces a method for reducing the amount of calculation of the weight map, thereby increasing the number of bones to be set. However, it achieves skinning decomposition that does not increase explosively. Specifically, considering the locality of the weight map, when obtaining the weight map of each bone, the solution is approximately obtained using the animation only in the vicinity of the bone, not the animation of the entire object. As a result, the system matrix of the quadratic programming problem to be solved when calculating the weight map can be reduced, and the system matrix can be kept substantially constant even when the number of bones to be set increases. . In addition, when obtaining the weight map, a process of thinning out similarly shaped frames is introduced instead of using the object shapes of all the frames in the animation. This makes it possible to reduce the number of calculations for calculating the weight map. By combining these two approaches, the calculation time in the weight map is greatly reduced, and the calculation time of the entire SSDR is reduced. Since the proposed method is a method for obtaining an approximate solution of SSDR, the restoration error becomes slightly larger than that of the conventional SSDR. However, in combination with the above-described cluster allocation process, speedup and accuracy can be improved. Enables compatibility.

It is a figure which shows the structural example of the computer system which performs the skinning decomposition speed-up method and the same skinning decomposition speed-up program which considered the locality of the weight map by this invention. It is a figure for demonstrating operation | movement of the skinning decomposition | disassembly speed-up program by a present Example. It is a figure for demonstrating the outline | summary of the skinning decomposition | disassembly by a prior art. It is a figure for demonstrating the arithmetic expression used for [the rigid body transformation matrix update process of a bone] by a present Example. It is a figure for demonstrating the principle of the [cluster allocation process process] by a present Example. It is a figure for demonstrating the computing equation of the cost function used in the case of allocation of a cluster. It is a figure for demonstrating the secondary planning problem which should be solved in the [weight map update process] by a present Example. It is a figure for demonstrating the calculation amount reduction process principle by the [system matrix reduction process process] by a present Example. It is a figure for demonstrating the change of the system matrix by the [system matrix reduction process process] by a present Example. It is a figure for demonstrating the processing time shortening effect in the [system matrix reduction process process] by a present Example. It is a figure for demonstrating the change of the error in the [system matrix reduction process process] by a present Example. It is a figure for demonstrating the principle of [the calculation frequency reduction process process] by a present Example. It is a figure for demonstrating the arithmetic expression and result which are used for [the calculation frequency reduction process process] by a present Example.

  The details of the skinning decomposition acceleration program to which the skinning decomposition acceleration method considering the locality of the weight map according to one embodiment of the present invention is applied will be described below with reference to the drawings.

[Constitution]
As shown in FIG. 1, the hardware configuration of a computer system that executes a skinning decomposition acceleration program to which the skinning decomposition acceleration method according to the present invention according to the present embodiment is applied is shown in FIG. For a memory 17 that is a storage means for storing data, an input section 19 that is an input means such as a keyboard operated by an operator, a display section 18 that displays image data such as 3D animation, and a stationary object A cluster allocation processing unit 12 for setting bones called control points for manipulating objects, and a cluster of neighboring bones for a plurality of clusters for which bones are set by the cluster allocation processing unit 12 System matrix reduction for clustering The number-of-operations reduction processing unit 14 that performs thinning to limit the number of weight map update computations for repetitive motion and stationary objects among the objects in the animation clustered by the system matrix reduction processing unit 13. A weight map update processing unit 16 for updating a weight map of a stationary object whose amount of calculation has been reduced by the system matrix reduction processing unit 13 and the calculation frequency reduction processing unit 14, and an update by the weight map update processing unit 16 A bone rigid transformation matrix update processing unit 15 for updating the rigid transformation matrix of the bone of the object being animated with respect to the weight map, a vertex-based animation information database (DB) 10 for storing vertex-based animation information, and a skin Anime It includes a skin animation information database (DB) 11 for storing ® down information, and CPU20 for controlling the respective portions.

  The computer system that executes the skinning decomposition acceleration program according to the present embodiment executes a process that will be described later, whereby a vertex-based animation that stores vertex coordinates of objects of all frames in the animation is set in the object. Skinning decomposition is performed to convert the pose and the corresponding weight map into a skin animation that calculates the shape with high accuracy and high speed.

The configuration example shown in FIG. 1 is merely an example, and the configuration is such that the operation of the processing unit such as the cluster allocation processing unit 12 is executed by a program stored in the memory 17 or other peripheral devices are connected. You may comprise as follows.
[Description of skinning decomposition method]

The proposed method consists of [cluster allocation processing], [system matrix reduction processing], and [calculation frequency reduction processing]. The proposed method is an extension of the SSDR initialization process and can be easily incorporated into a conventional framework. For processing other than the initialization step, “Smooth Skinning Decomposition with Rigid Bones” (http: //graphics.cs.uh_ed12_20/sap20_dus/sap20_dus_dr/dr/dap20/Sr20_dus_rap_dr/dap_sr20_dus_dr12 .Pdf) is applied as it is.
[Cluster allocation process]

  First, in the conventional skinning decomposition, in the case of a “horse” whose movable part is an object determined by a skeleton or a joint, as shown in FIG. 5A, the entire “horse” is converted into one cluster (stationary state). The multiple vertices that make up the object are considered as a unit, and a new cluster is assigned to the part with the highest cost function value in the cluster (the circle in the figure), and so on. The task of assigning a new cluster to the part with a large value was repeated. In the case of this conventional method, a cluster is newly assigned to a portion where the value of the cost function is already sufficiently small and well clustered, and there is a problem that a useless cluster is generated, and the number of clusters varies depending on the number of powers of 2. Therefore, there is a problem that the user cannot set the number of favorite clusters.

  Note that clustering described in this application is used in two meanings. In “cluster assignment processing”, it is a set of vertices that move in a similar manner. Specifically, it means a set of vertices that are given the same rigid transformation matrix in the initialization process. To do. [System matrix reduction processing] means a cluster and a set of clusters existing within the second adjacency from the cluster (cluster clustering). The cost function will be described later.

  Therefore, in the [cluster allocation process] of the present invention, as shown in FIG. 5B, a new cluster is not allocated to all clusters, but a cluster having a large cost function value (for example, a cost function) A method is adopted in which a new cluster is assigned only to a portion of which the value of is higher than 1 / n of the entire cluster, two or more portions in the illustrated example, and a circle portion in the drawing). In other words, the present invention initially considers the whole as one cluster, assigns a new cluster to the cluster with the cost function value of the upper 1 / n, and the number of clusters is (the value specified by the user) × ( n / n + 1) are allocated one by one. As a result, the present invention can assign a cluster where the cluster is to be assigned, and can also reduce the final restoration error. For example, when the object is a horse, the conventional method concentrates the bone (marked with a circle in the figure) on the tail or leg where the amount of operation is large due to shaking, but in the present invention it does not concentrate on the part and operates You can place bones all over the part.

  Note that the cost function is “Robust and Accurate Skeletal Rigging from Mesh Sequences” (URL: http: //graphics.cs/u-s/s-u/sd. .Pdf) is an arithmetic expression shown in FIG. 6 as (1a) and (1b) of the third page “Cluster splitting”. In the present invention, this cost function is used as it is. This cost function is a function representing how much the vertices in the cluster follow the movement of the bone. The value of the cost function of a vertex is the sum of the shape reconstructed using the parameters calculated by skinning decomposition and the error of the example data for each frame, the centroid of the cluster to which the vertex belongs in the initial shape, and its It is calculated as the product of vertex distances. The higher the value of this cost function, the more different the vertex will be from the cluster to which it belongs. Therefore, in the present invention, a vertex having the maximum cost function value in each cluster is set as a candidate for a location to which a new cluster is allocated.

  [Cluster allocation processing] is for an object whose movable part is determined by a skeleton or the like. In the case of an object whose moving part is not defined, for example, an object such as a cloth swaying in the wind, Kavan et al. Kavan et al. Determine the bone position using farest point sampling (hereinafter referred to as the farthest point sampling method), set the weight map, and perform skinning decomposition based on the dual quaternion concept. . In the present invention, for an object for which a movable part has not been determined, the position of the bone is determined using the farthest point sampling method, and then the weight map and the rigid transformation matrix of the bone are obtained by SSDR, so that SAD and SSDR are obtained. Convert soft objects with higher accuracy to skinning-based animations.

<URL>
http: // www. tnt. uni-hanover. de / papers / data / 815 / 815_1. pdf

<Research by Kavan et al.>
KAVAN, L.A. McDonnell, R .; , DOBBYN, S. , ZARA, J. et al. , AND O'SULLIVAN, C.I. 2007. Skinning arbitrary deformations. In I3D'07: Proc. of Symp. on Interactive 3D Graphics and Games, 53-60.

  The farthest point sampling method is a technique in which sample bones are arranged in space so that the minimum distance from a sample bone having a stationary pose object to the nearest sample bone is maximized. URL <https: // www. cs. utah. edu / ~ ladislav / kavan07 skin. arbitrary / kavan07 skin. arbitrary. pdf>.

  In the case of an object to which a blend shape is applied, clustering is performed based on preset bones (such as those set by an artist). In this embodiment, the initial value of the weight map was calculated and clustered using the smooth skin function of Autodesk Maya (Autodesk is a registered trademark). There are many methods for calculating weight maps from the positional relationship between the bones and vertices of an object, so you can use your own method here.

As described above, in the [cluster assignment process] according to this embodiment, high-precision skinning decomposition is applied to a wide range of target animations by appropriately assigning clusters according to the properties of the target object. Make it possible.
[System matrix reduction processing]

  Next, a description will be given of [system matrix reduction processing] for reducing the size of the system matrix of the quadratic programming problem in the weighting update process (Update Bone-Vertex Weight Map) of skinning decomposition.

  First, in the weight map update process according to the prior art, it is necessary to solve the quadratic programming problem shown in FIG. 7 in which the arithmetic expressions of the linear blend skinning method are arranged in terms of skin weights (weighting). This quadratic programming problem is that the skin weight is not negative as a constraint, the skin weight of each vertex is 1 in total, and the number of bones affecting each vertex is K (generally 4) The following condition is added. Also, this optimization problem needs to be solved for a dense system matrix.

  The conventional method solves the quadratic programming problem for all vertices for each vertex to calculate the skin weight, then takes the top four bones with the largest skin weight values, and again quadratic plans for them Solve the problem and determine the final skin weight. For this reason, the size of the system matrix for the quadratic programming problem is proportional to the number of bones set, and the amount of calculation increases explosively as the number of bones to be set increases.

  Here, as shown on the left side of FIG. 8A, for each cluster set in the human body model, a value indicating that the cluster is the nth neighboring cluster of the cluster to which the target vertex belongs can be defined. I can do it. As shown by the arrow on the left side of FIG. 8B, the movement of the surface vertex a is strongly influenced by the bone of the cluster having a small value, and is not significantly influenced by the bone of the cluster having a large value (for example, the hand) (weight). Map locality). In skinning, since the number of bones affecting a certain vertex is usually K (generally 4) or less, the influence from bones having a large value can be ignored.

  Therefore, the present invention proposes a method for calculating the skin weight under the assumption that each vertex of the object is not affected by a bone far from the object. The method includes a first step of constructing adjacent information between the clusters using the clustering result of [cluster assignment processing], a second step of obtaining a cluster number to which a vertex for obtaining a skin weight belongs, A cluster whose cluster number has been acquired in two steps, a third step for acquiring bones within the second neighborhood of the cluster (cluster clustering), and a skin weight for only the bones acquired in the third step are calculated. And a fifth step of taking out the top four bones having the largest value among the skin weights calculated in the fourth step and solving the quadratic programming problem for them again to determine the final skin weight. .

  By clustering this cluster, as shown in the left side of FIG. 8B, the weight calculation is performed on all the bones with respect to the surface vertex a, as shown in the right side of FIG. 8B. The calculation target is limited to only the bones of the cluster within the second adjacency of the cluster to which the vertex a belongs. As a result, the size of the system matrix of the quadratic programming problem can be reduced and kept constant regardless of the number of bones to be set.

  The effect of clustering this cluster will be described next. According to the clustering of this cluster, as shown in the left side of FIG. 9B, the number of elements of the system matrix is the number of bones × the number of bones in the conventional method. b) As shown on the right side, the number of bones within the second neighborhood of each vertex × the number of bones within the second neighborhood can be set, and the size of the system matrix can be reduced ([system matrix reduction processing] ).

  By this [system matrix reduction processing], the present invention can shorten the processing time. For example, in the case where the object is a cloth having a large displacement, as shown in FIG. When the number becomes 20 or more, the weight calculation time (weight map update processing calculation time) suddenly increases. However, in the present invention, the weight calculation time increases even when the number of bones exceeds 20 as shown in the left side of the figure. This is the same for the weight calculation time for the horse shown in FIG. 10 (b) and the weight calculation time for the human face shown in FIG. 10 (c).

Further, according to this [system matrix reduction processing], an increase in restoration error after updating the weight map can also be suppressed. For example, when the object is a cloth having a large displacement, as shown in FIG. Compared with the conventional method shown in FIG. 6 and the present invention shown on the left side of the figure, the restoration error does not increase. This is the restoration error for the horse shown in FIG. 11B and the restoration shown in FIG. The same applies to errors. Note that this process alone is not more accurate than the conventional method, but even if an approximate solution is found, the recovery error hardly increases, and when combined with the cluster allocation process, it is more accurate than the conventional method. Will improve.
[Calculation count reduction processing]

  Next, the present invention employs [calculation number reduction processing] for reducing the number of times of solving the quadratic programming problem following [system matrix reduction processing]. This [number-of-calculations reduction processing] pays attention to the fact that there are scenes in which the object repeatedly moves or is stationary during the animation. Reduce the number of operations by performing map update operations (thinning processing). For example, in the animation in which a horse travels as shown in FIG. 12 (a), there is a frame with a similar running posture of the horse, and there is a frame with a similar shape fluttering clothes as shown in FIG. 12 (b). Focusing on the above, the object in the repetitive motion and stationary key frame is selected as a weight map update calculation target only once.

In order to perform this thinning process, the present invention introduces an evaluation function expression representing the posture similarity between objects in the animation as shown in FIG. 13 (a), for example, the neck of the horse shown at the left end of FIG. 13 (b). When the conversion from the rest (stationary) pose at the time t shown on the right side to t = t1 is R _j1 ^t1 and the conversion to t = t2 is R _j1 ^t2 with respect to the partial bone j1,
When the Frobenius norm value of the difference between R _j1 ^t1 and R _j1 ^t2 is equal to or smaller than a predetermined threshold value, the two postures are evaluated to be similar, and the weight map is only once for a key frame having the similar posture. Select as update calculation target.

With this thinning process, as shown in FIG. 13C, for example, the number 48 of key frames that are subject to weight map update calculation for horse motion can be reduced to the number 11 of key frames.
[Operation]

Next, the operation of the skinning decomposition acceleration program employing the skinning decomposition acceleration method according to the present embodiment will be described.
The skinning decomposition acceleration program according to the present embodiment is stored in the memory 17 shown in FIG. 1, and the CPU 20 performs skinning decomposition on the vertex-based animation information into skin animation information. As shown in FIG. An initialization process and a skinning decomposition process are executed in each step.

The initialization process is executed by the following steps.
(1) Step S201:
Reading the vertex-based animation information to be processed from the vertex-based animation information DB 10 into a memory;
(2) Step S210:
Objects whose movable parts such as animals are determined by skeletons, objects such as cloth, which are largely deformed and objects created by blend shapes, with respect to the stationary object of the vertex-based animation information read in step S201 [Cluster allocation processing step] step of assigning bone clusters for manipulating the object by applying different methods depending on the characteristics of the object.

Step S210 includes the following steps.
(3) Step S211:
Determining whether a blend shape has been applied to the object and a bone has been set by the artist.
(4) Step S212:
When it is determined in step S211 that the blend shape is not applied to the object, it is determined whether or not the movable part is an object that is determined by a skeleton or the like, or an object that undergoes a large deformation in which the movable part is not determined. Determining.
(5) Step S213:
A step of calculating an error value for each cluster and assigning a new cluster to a cluster having an error value equal to or greater than a threshold when it is determined in step S212 that the movable part is an object determined by a skeleton or the like. However, the entire object is initially set as one cluster.
(6) Step S214:
When it is determined in step S212 that the object (cloth or the like) undergoes a large deformation with no movable part determined, clustering is performed based on the result of determining the bone position using the farthest point sampling method (Farest Point Sampling). Step to perform.
(7) Step S215:
When it is determined in step S211 that a blend shape is applied to the object, clustering is performed based on preset bones (such as those set by an artist).

Following the step S213 value step S214 and step S215, the initialization process according to this embodiment executes the following steps.
(8) Step S220:
Only the neighboring bone clusters are clustered with respect to the bone clusters set in the step S210, and the size of the system matrix of the quadratic programming problem in the weight map update operation is reduced and kept constant [System Matrix reduction process step].
(9) Step S230:
For the bone cluster set in step S210, an evaluation function formula representing posture similarity between objects in animation is introduced, and the weight map is only once for objects in repetitive motion or stationary key frames. [Calculation number reduction process] step of limiting the number of weight map update calculations by setting an update calculation target.

Following the initialization step, the next skinning decomposition step is performed.
(10) Step S240:
A step of a weight map update step of updating the weight map of the object in the animation selected as the weight map update calculation target in step S230;
(11) Step S250:
A bone rigid transformation matrix updating step for updating the bone rigid transformation matrix of the object bone in the animation in which the weight map is updated in step S240.

(12) Step S260:
Subsequent to step S250, it is determined whether the skinning decomposition process has been repeated for the number of times set by the user, and the weight map is determined when it is determined that the process has not been performed for the number of times set by the user. Returning to step S220 of the update process.
(13) Step S270:
A step of outputting skin animation information to the skin animation information DB 11 when it is determined in step S260 that the skinning decomposition process has been repeated for the number of times set by the user.

  In this example, whether the movable part in step S212 is an object determined by the skeleton or the like or an object that undergoes a large deformation in which the movable part is not determined may be set by the operator, The CPU may determine by pattern matching based on the shape or outline.

  Further, bone setting for operating the object in step S210 may be performed by an operator operation.

  Furthermore, in the step S220, an example has been described in which bones within the second adjacency are subjected to calculation of the quadratic programming problem in the skin weight update calculation. However, the present invention is not limited to this, and the nth Bones within the adjacent range (n is an integer) may be the target of calculation of the quadratic programming problem, or bones that are close to each other may be simply calculated as the target of calculation of the quadratic programming problem.

10 vertex-based animation information database (DB),
11 Skin animation information database (DB),
12 initialization processing unit, 14 thinning processing unit, 15 skeleton conversion update processing unit,
16 weight map update processing unit, 17 memory, 18 display unit,
19 input unit, 20 CPU

To achieve the above object, the present invention provides a vertex-based animation for storing vertex coordinates of objects of all frames in an animation, and a skin animation for calculating a shape from the posture of a bone set for the object and a corresponding weight map. In addition, in the skinning decomposition acceleration method and program considering the locality of the weight map that is automatically converted using a computer,
To the computer,
The object of the stationary state, the movable portion such as an animal varies depending on the characteristics of the object the object, such as that created objects definite by skeleton like, objects to large variations not determined moving parts such as cloth, a blend Shape Assign a cluster of bones for manipulating the object by applying a method [cluster assignment process]
Only the neighboring bone clusters are clustered with respect to the bone cluster set in the [cluster assignment processing step], and the size of the system matrix of the quadratic programming problem in the weight map update operation is reduced to a constant size. Keep the [system matrix reduction processing step],
For the bone cluster set in the [Cluster Allocation Processing Step], an evaluation function expression representing the pose similarity between objects in the animation is introduced. After an initialization process of executing [calculation frequency reduction processing] that limits the number of weight map update calculations by setting the weight map update calculation target to solve the secondary planning problem only once,
A weight map update step of updating the weight map of the stationary object by the initialization step;
A bone rigid body transformation matrix update step for updating a rigid transformation matrix of a bone of an object in the animation in which the weight map is updated by the weight map updating step;
The main feature is that it is repeatedly executed. In particular, according to the present invention, in order to achieve the first object, three methods are selectively used according to the object to be processed in the [cluster allocation processing step] of the initialization step. (1) For an object such as an animal whose movable part is determined by a skeleton or the like, an error value is calculated for each cluster, and a new cluster is assigned to a cluster whose error value is equal to or greater than a threshold value. However, the entire object is initially set as one cluster. (2) Clustering is performed for an object that has a large deformation whose movable part is not determined, such as cloth, based on the result of determining the position of the bone using the farthest point sampling method (Farest Point Sampling). (3) In blend shape application, clustering is performed based on preset bones (such as those set by an artist). In order to achieve the second object, the present invention introduces a technique for reducing the amount of calculation of the weight map in the initialization process. In this weight map calculation amount reduction method, when calculating the weight map of each bone, the system matrix reduction process that approximates the solution using only the animation of the neighborhood of the bone instead of the animation of the entire object, When obtaining the weight map, the object shape of all frames is not used, but a frame having a similar shape is thinned out [calculation frequency reduction processing step].

Claims (8)

A vertex-based animation that stores the vertex coordinates of objects in all frames in the animation is automatically converted using a computer to a skin animation that calculates the shape from the posture of the bone set in the object and the corresponding weight map. A skinning decomposition speedup method considering the locality of the weight map to be performed,
To the computer,
Depending on the characteristics of the stationary object, depending on the characteristics of the object, such as an object whose movable part is determined by a skeleton, such as an animal, an object that undergoes a large deformation such as a cloth, an object created by a blend shape, etc. Assign a cluster of bones for manipulating the object by applying a method [cluster assignment process]
Only the neighboring bone clusters are clustered with respect to the bone cluster set in the [cluster assignment processing step], and the size of the system matrix of the quadratic programming problem in the weight map update operation is reduced to a constant size. Keep the [system matrix reduction processing step],
For the bone cluster set in the [Cluster Allocation Processing Step], an evaluation function expression representing the pose similarity between objects in the animation is introduced. After the initialization process of executing [calculation frequency reduction processing] that limits the number of weight map update calculations by setting the weight map update calculation target only once,
A weight map update step of updating the weight map of the stationary object by the initialization step;
A bone rigid body transformation matrix update step for updating a rigid transformation matrix of a bone of an object in the animation in which the weight map is updated by the weight map updating step;
A skinning decomposition acceleration method that takes into account the locality of the weight map that is repeatedly executed. In the [cluster allocation processing step],
In the computer,
A first step of determining whether a blend shape has been applied to the object and a bone has been set by the artist;
A second step of performing clustering based on preset bones (such as those set by an artist) when it is determined that a blend shape is applied to the object in the first step;
When it is determined that the blend shape is not applied to the object in the second step, it is determined whether the movable part is an object determined by a skeleton or the like, or an object undergoing a large deformation with no movable part determined. A third step to perform,
A fourth step of calculating an error value for each cluster and allocating a new cluster to a cluster having an error value equal to or greater than a threshold when it is determined in the third step that the movable part is an object determined by a skeleton or the like; ,
A fifth step of performing clustering based on the result of determining the position of the bone using the farthest point sampling method when it is determined in the third step that the object is a large deformation whose movable part is not determined;
The skinning decomposition speed-up method in consideration of the locality of the weight map according to claim 1, wherein: In the [system matrix reduction processing step],
In the computer,
A sixth step of constructing adjacent information between the clusters using the clustering result of the [cluster allocation process];
A seventh step of acquiring a cluster number to which the vertex for obtaining the skin weight belongs;
An eighth step of obtaining a cluster whose cluster number has been obtained in the seventh step and a bone within the second neighborhood of the cluster (cluster clustering);
A ninth step of calculating skin weights only for bones acquired in the eighth step;
A tenth step of taking out the top four bones having a large value among the skin weights calculated in the ninth step and solving the quadratic programming problem for them again to determine a final skin weight;
The skinning decomposition speed-up method in consideration of the locality of the weight map according to claim 1, wherein: In the [Calculation frequency reduction process],
In the computer,
An eleventh step of introducing an evaluation function expression representing a posture similarity between objects in the animation to the cluster of bones set in the [cluster allocation processing step];
A twelfth step of calculating a value of an evaluation function equation between any two key frames using the evaluation function equation introduced in the eleventh step; and a value of the evaluation function equation calculated in the twelfth step is equal to or less than a threshold value. A thirteenth step for determining whether or not
A fourteenth step in which, when it is determined in the thirteenth step that the value of the evaluation function formula is equal to or less than a threshold, only one of the key frames is a weight map update calculation target;
The skinning decomposition speed-up method in consideration of the locality of the weight map according to claim 1, wherein: A vertex-based animation that stores the vertex coordinates of objects in all frames in the animation is automatically converted using a computer to a skin animation that calculates the shape from the posture of the bone set in the object and the corresponding weight map. A skinning decomposition acceleration program that takes into account the locality of the weight map to be
To the computer,
Depending on the characteristics of the stationary object, depending on the characteristics of the object, such as an object whose movable part is determined by a skeleton, such as an animal, an object that undergoes a large deformation such as a cloth, an object created by a blend shape, etc. Assign a cluster of bones for manipulating the object by applying a method [cluster assignment process]
Only the neighboring bone clusters are clustered with respect to the bone cluster set in the [cluster assignment processing step], and the size of the system matrix of the quadratic programming problem in the weight map update operation is reduced to a constant size. Keep the [system matrix reduction processing step],
For the bone cluster set in the [Cluster Allocation Processing Step], an evaluation function expression representing the pose similarity between objects in the animation is introduced. After the initialization process of executing [calculation frequency reduction processing] that limits the number of weight map update calculations by setting the weight map update calculation target only once,
A weight map update step of updating the weight map of the stationary object by the initialization step;
A bone rigid body transformation matrix update step for updating a rigid transformation matrix of a bone of an object in the animation in which the weight map is updated by the weight map updating step;
A fast skinning decomposition program that takes into account the locality of the weight map to execute repeatedly. In the [cluster allocation processing step],
In the computer,
A first step of determining whether a blend shape has been applied to the object and a bone has been set by the artist;
A second step of performing clustering based on preset bones (such as those set by an artist) when it is determined that a blend shape is applied to the object in the first step;
When it is determined that the blend shape is not applied to the object in the second step, it is determined whether the movable part is an object determined by a skeleton or the like, or an object undergoing a large deformation with no movable part determined. A third step to perform,
A fourth step of calculating an error value for each cluster and allocating a new cluster to a cluster having an error value equal to or greater than a threshold when it is determined in the third step that the movable part is an object determined by a skeleton or the like; ,
A fifth step of performing clustering based on the result of determining the position of the bone using the farthest point sampling method when it is determined in the third step that the object is a large deformation whose movable part is not determined;
The skinning decomposition acceleration program in consideration of the locality of the weight map according to claim 5 for executing In the [system matrix reduction processing step],
In the computer,
A sixth step of constructing adjacent information between the clusters using the clustering result of the [cluster allocation process];
A seventh step of acquiring a cluster number to which the vertex for obtaining the skin weight belongs;
An eighth step of obtaining a cluster whose cluster number has been obtained in the seventh step and a bone within the second neighborhood of the cluster (cluster clustering);
A ninth step of calculating skin weights only for bones acquired in the eighth step;
A tenth step of taking out the top four bones having a large value among the skin weights calculated in the ninth step and solving the quadratic programming problem for them again to determine a final skin weight;
The skinning decomposition acceleration program in consideration of the locality of the weight map according to claim 5 for executing In the [Calculation frequency reduction process],
In the computer,
An eleventh step of introducing an evaluation function expression representing a posture similarity between objects in the animation to the cluster of bones set in the [cluster allocation processing step];
A twelfth step of calculating a value of an evaluation function equation between any two key frames using the evaluation function equation introduced in the eleventh step; and a value of the evaluation function equation calculated in the twelfth step is equal to or less than a threshold value. A thirteenth step for determining whether or not
A fourteenth step in which, when it is determined in the thirteenth step that the value of the evaluation function formula is equal to or less than a threshold, only one of the key frames is a weight map update calculation target;
The skinning decomposition acceleration program in consideration of the locality of the weight map according to claim 5 for executing