Research on 3D simulation design and dynamic virtual display of clothing flexible body

3.1.1 Body modeling and clothing module database

Human modeling is the basis of the garment simulation model, without the acquisition of human physiological feature points, it cannot accurately establish the human parameter model. Human modeling is to form a virtual 3D model of a real human body through parameterization. At present, there are many methods to obtain the 3D model of a human body. The common methods are as follows: the first is to capture the visual points of a human body from multiple angles with high-definition cameras in a closed space, obtain the spatial point cloud coordinates of the human body through image processing technology, and then use the splicing algorithm to form a human model. This requires high quality of environment and camera, as shown in Figure 7. The second is to use Kinect photosensitive capture camera equipment to convert scanned images into body size data using data cleaning and compression technology and use multiple algorithms such as ray casting algorithm to obtain the coordinates of spatial point cloud, and finally build a human model [4,5] as shown in Figure 8. The third method is to establish a point cloud model by acquiring data of multiple human parts and, at the same time, compare it with the existing human model database, and select a closer human model for fine-tuning reconstruction, as shown in Figure 9.

Figure 7

3D Human scanning room.

Figure 8

Kinect photosensitive capture scanner.

Figure 9

3DCLO design software models with data.

At present, the three methods are the best and have the highest utilization rate. In enterprises, the first and second methods are commonly used, and the third method is mostly used for college research. However, some problems have also been encountered. From this topic, the researchers believe that integrating the second and third methods to extract and model human body data is the best method with high accuracy, and can fine-tune relevant data in the later stage. Specifically, Mao and Wang [6] used the ray cast by the non-contact three-dimensional scanning equipment (photosensitive capture equipment) to the human body, used the sensor to receive the reflected ray, used the photoelectric conversion element to obtain the digital signal, and then used the image processing technology to obtain the spatial point cloud coordinates, which are converted into human body size data. According to the 24 physiological parameters of the human body obtained, different human skeleton models are constructed using the SMPL model, as shown in Figure 10. Human body shapes and features can be divided into different categories and sizes to achieve personalized and accurate customization [7].

Figure 10

Female manikin derived from SMPL model [7].

Among them, SMPL [8] (Skinned Multi Person Linear Model) is a naked, vertex-based three-dimensional human model, which can accurately represent different shapes and postures of the human body. The SMPL model contains N = 6,890 vertices, 13,776 patches, 24 joint points (note: 23 joint points + 1 root node), and the model parameters are determined by Φ = { T ¯ , W , S , J , P } definition [8]. The internal parameters of the model are as follows:

• T̄: R ^3N , Human average network model;

• W: R ^N×K , LBS/QBS, Hybrid Matrix;

• S: R ^3N×|β|, PCA base for human body shape;

• J: R ^3N×3K , Calculate the regression matrix for the 24 nodes;

• P: R ^3N×9K , PCA base for human posture.

Then, the corresponding formula of the model: M(β, θ) = W (T _p (β, θ), J(β), θ, W) [9]. The calculation process consists of three steps: body parameters (vertex offset) – posture parameters (vertex offset) – skinning. The number of shape PCA-based parameters is 4.28 × 10⁻⁶, and the number of pose PCA-based parameters is 23 × 9 = 207.

Figure 11 is the schematic diagram of the joint point layout in the SMPL model, which is a normal posture. The coordinates of the human joint points are calculated by the formula J ( β ) = J ( T ¯ + B s ( β ; S ) ) . Given the shape parameters and posture parameters, the human body in the figure will change its shape and posture according to the changes in the parameters. The specific results are as follows: M ( β → , θ → ; φ ) = W ( T p ( β → , θ → ; T → , S , P ) , J ( β → ; J ; T → , S ) , θ → , W ) . Through the algorithm model training of attitude parameters W, J, P in multipose dataset, shape parameter T ¯ , S The multishape dataset algorithm model is trained in many aspects, and the parameters and nodes will be optimized finally to meet the application needs most suitable for the subject research. At the same time, different human models can be formed according to different shapes and postures to meet the needs of personalized customization.

Figure 11

Layout of 23 joint points in SMPL model.

After the manikin is determined, the clothing module database can be used for a certain degree of combination design from the database, including fabric library, body library, accessories library, printing and embroidery library and parts (styles) library, as shown in Figure 12. At the same time, the content of the module database can be edited and adjusted. For the clothing modules attached to the manikin, a secondary design (combination) can be carried out according to the needs of researchers (Figure 13).

Figure 12

Interface design of apparel module database.

Figure 13

Clothing module database – part library content display.

The two contents of the human body modeling and clothing module database are separate. The clothing module database appears in the form of Web pages, and the two parts can cooperate with each other through computer input ports, JAVA architecture, links, and other tools. Specifically, after processing the mannequin, you can link to the clothing module database in an embedded way. You can click what you need to get the link, and directly edit the mannequin until you are satisfied.

3.1.2 Physical model and design of fabric

Fabric simulation is an important step in the process of 3D flexible simulation of clothing. It is also the attempt and preparation of the subsequent flexible simulation model. At present, there are three methods for fabric simulation models: physical method, geometric method, and hybrid method. The most commonly used, simplest, and most valuable method is the physical method. It takes a long time to calculate the speed and position of particles by analyzing the force of particles in the model, but the results are real. The most commonly used physical model is the spring-mass model, which is simple, effective, and easy to implement (Figure 14).

Figure 14

Type of spring-mass model: (a) tensile spring, (b) square grid, (c) shear spring, and (d) bending spring [10].

The particle represents the movement track of the unit and finally forms the movement deformation route (shape) of the fabric. There are several types of specific spring particle models: a. tensile spring, b. square grid, c. shear spring, and d. bending spring; the four spring-mass models have different characteristics, which are used in physical modeling. These spring-mass models form the internal power of the fabric mechanical system and are also the cause of minor changes in the fabric, such as fabric stretching, shaking, and wrinkles.

In the process of spring-mass model simulation, the fabric is separated into many mass points, and the mass points are connected by many small springs, and the model needs to calculate their tensile force and bending coefficients and distribute the mass points reasonably, so before calculating the physical model of the spring-mass points, it is also necessary to add the physical data information to the model, these pieces of information are the basis for the movement, force, and structural bending deformation of spring particles. The associated parameters for the physical model of fabric simulation are shown in Table 1, which is based on the physical information related to the force applied to the fabric. The whole table will be divided into three parts: internal factors, external factors, and other parameters, which are all close factors affecting the four spring-plasmoid models, and a clear understanding of them is decisive for the spring-plasmoid model simulation of fabrics.

Specifically, external factors include human factors and non-human factors. Human factors mainly refer to the effect of other forces on the fabric, including pulling, rotating, etc. [11]. The non-human factors are mainly the wind force, air resistance, and gravity in nature. The wind factor coefficient is F = w ⋅ r V ¯ ∣ r V ¯ ∣ . The internal factors include three types of spring-mass models, which are mainly related to spring elongation, tensile data, friction data, and collision data, which are related to fabric folds. Other factors are mainly model correlation factors, including the number of model particles, model structure, and model difficulty. Therefore, model correlation factors involve the collision effect detection of fabrics and the setting of bending springs. The parameters in Table 1 are applicable to ordinary fabrics. If special fabric effects, shapes, or materials are involved, the parameters in the table will change accordingly.

Therefore, to build a spring-mass model (physical model of fabric), first of all, we need to build a mechanical system. Generally, the movement process of the mass is in accordance with Newton’s second mechanical law [10], and the formula is as follows:

The cloth is mainly affected by gravity, wind, collision force, and the performance of the fabric itself. These factors affect the calculation rate of the spring particle model, which is constantly changed in the process. Then, the velocity and position of the next particle of the cloth are calculated through Newton’s second law of mechanics, where m represents the weight of the particle, f _G represents that only gravity is considered in this topic, f _in represents the total internal force of the particle, and y represents the current position of the spring particle [10].

The particle is affected by wind, gravity and particle coordinates, quantity, and position. The researchers calculate the deformation force of the particle spring by using Hooke’s law [12]. The formula is as follows:

where ∆ _χ is the expansion and contraction of the mass–spring.

The mass will consume energy in the process of moving or moving, and the spring-mass will vibrate in the process of consumption, affecting the running track, so the damping force is required to simulate and eliminate such vibration [9]. The formula is as follows:

where C _d is the damping parameter, and υ is the motion direction of the mass-spring.

Through the above analysis, formulas (1)–(3) are converted into formula (4):

The entire physical model equation is a nonlinear differential equation. At the same time, the explicit midpoint method is used to calculate the motion and spatial coordinates of the spring particle, and the calculation is repeated iteratively; that is, the physical motion of the fabric can be simulated.

For the model system established by spring mass, during the movement process, due to the characteristics of spring structure, the spring will appear excessively stretching, which will eventually affect the effect, and even lead to distortion of cloth simulation. The researchers believe that this pain point problem can be solved by improving position correction. Specifically, by changing the speed of the particles on both sides of the spring and the position coordinates of the particles on both sides of the spring, the hardness and softness of the cloth can be reflected to achieve the desired simulation effect [10].

3.1.3 Analysis of collision detection between human body and fabric

Cloth simulation technology is a very important step in virtual clothing, so the collision and processing between human body and fabric are also the most time-consuming part of the cloth simulation process. When clothing is worn on the human body, many collision phenomena will be encountered when simulating the wearing effect, such as the crossing and collision between the mannequin and cloth, and between cloth and cloth. Therefore, collision detection is a key problem in the simulation process. The simplest method of collision detection is to pass through the basic geometric elements in the mannequin and fabric.

At present, there are two commonly used collision detection methods: spatial decomposition and hierarchical bounding box. Specifically, the spatial decomposition method is to divide the virtual space centered on the human model into multiple cells with equal volume and cross-test the crossing between cells occupying the same cell or adjacent cells. Hierarchical bounding boxes are used to divide multiple parts of the human body model by using the bounding boxes with large volumes and simple modeling to form multiple bounding boxes of different sizes. When the bounding boxes cross each other, the objects between the two bounding boxes will be cross-calculated. At the same time, the bounding boxes also have some hierarchical structures, such as bounding boxes along the coordinate axis, bounding balls, and directional bounding boxes.

In this study, the author combines the advantages of the two methods. Aiming at the cross collision between the mannequin and the fabric, the author makes use of the advantages of hierarchical bounding boxes (two or more bounding boxes cross each other – wrapping objects <mannequin> cross calculation), and uses hierarchical spatial decomposition (spatial decomposition of multiple color cells) to improve the order, three-dimensional and flexibility between the mannequin and the fabric. It reduces the irrelevant elements in cross-testing and improves the speed and efficiency of collision detection.

Specifically, first, the authors take the female human body model as an example and use the hierarchical enveloping box and static ellipsoid fitting to construct the static keyframe screen. In Figure 15, the schematic diagram of a hierarchical enveloping box and static ellipsoid fitting [10], the right side is the randomly selected two keyframe screens for the set of regions corresponding to the enveloping box coordinates, in which a is the ellipsoid enveloping box schematic region, indicated in green, and the regular shape is carried out to wrap the different parts of the human body; b is the interpolation region of ellipsoid fitting, when the number of keyframes of the model is small, the connection between the keyframes is not smooth, in order to reduce the interspersed fabrics and fabrics affect the cross between the human body model and fabrics, so insert linear interpolation between the two frames; c is the human body model nodes, based on the interspersed position, ellipsoidal structure to build the node coordinates around the human body model, through the node coordinates to better simulate and cross the external characteristics of the fabrics.

Figure 15

Schematic diagram of hierarchical bounding box and static ellipsoid fitting.

The second step uses the hierarchical space decomposition method to detect the collision between the mannequin and the fabric, which is also based on the first step of the hierarchical bounding box method. Figure 16a is the schematic diagram of the hierarchical space decomposition method of color labeling [13]. In this article, the color enclosure centered on the human body is wrapped with an irregular polygonal cube. In order to accurately calculate the contact value between the convex point of the fabric and the polygonal cube, the three-dimensional space is divided into multiple small squares with the same color and volume. Each small square has unique coordinate information unique color representation. When the mannequin is in motion, the clothing cloth and polygon color cube cross collide, and the coordinates will mark the coordinate hierarchy between the color bounding body and the human body and fabric. Specifically, the whole model coordinate diagram (Figure 16a) includes three parts: human body model coordinate, clothing cloth coordinate, and polygon color cube enclosure coordinate. Specifically, three collision forms are judged through algorithms, including the coincidence of human body model coordinate and clothing cloth coordinate, the coincidence of clothing cloth coordinate and polygon color cube enclosure coordinate, and the coincidence of human body model and polygon color cube enclosure coordinate. At the same time, there are also some self collisions between cloth, including the intersection of cloth points and faces. In the whole process, first, the Sweep and Prune algorithm is used to detect the physical method of AABB bounding box, and at the same time, the separating axis theorem algorithm is used to optimize. The intersection of the polygon cube’s face, line, and point with cloth and human body model is adjusted, and determined whether they intersect according to the overlapping shadow area.

Figure 16

Visualization diagram of color annotation space decomposition (a) and cross-collision pressure perception (b).

Figure 16b is the pressure perception diagram of the intersection of the garment fabric and the mannequin, which is also a reflection of the repetition of the coordinates of the two. Using different colors to indicate the size of the cross-pressure, a/b/c/d in the figure indicate the pressure value of different parts of the human body, c indicates the largest pressure value, which is orange-red, divided into the combination of crotch and fabric, the shoulder and neck points and the combination of fabric; the yellow area indicates that the pressure value is the next lowest, and the lowest pressure value is the blue area. The expression of the pressure perception map also highlights the color space, polygonal cubic enclosure, and the cross-fusion of coordinates of the human body model and multiple parts of the fabric in Figure 16a. It is the best expression in a collision detection method. Through 3D modeling and algorithms, the researcher simulated the pressure-aware visualization maps as in Figure 17 for a mannequin wearing a garment fabric, divided into four directions: front, back, and left and right sides, which provides a clear understanding of the collision-monitoring pressure values for each part.

Figure 17

Visualization diagram of pressure point perception.

3.1.4 Clothing fabric rendering

Cloth rendering is an important part of the 3D flexible design of clothing. Traditional cloth rendering uses machines to scan the tfabric, and secondary design and processing is carried out through AI or photoshop software to make the desired style. However, because the size proportion is often inconsistent with the actual, it consumes time and costs. Before the appearance of the digital physics engine, this method was a common method for garment enterprises to build fabric databases. Before analyzing the digital rendering of fabrics, researchers first proposed several factors affecting fabric rendering, as shown in Table 2.

The material definition and direction to be considered in the fabric rendering shown in Table 2 are important factors that affect the digital scanning and rendering in the later stage. Before rendering, we should first understand and be familiar with the material definition of fabric, including fabric appearance, physical attributes, and data attributes. In terms of fabric appearance, we mainly focus on the color, texture, luster, and other contents of the fabric, such as silk luster, cotton texture, etc. The physical properties of fabrics include elasticity, bending, pressure, and other properties, such as the elasticity of knitted fabrics and the bending degree of wool fabrics. Fabric data attributes include composition, yarn count, etc. For example, when rendering, the appearance of the components is considered.

The researcher is familiar with the material definition and direction attribute of the fabric. Starting from this topic, the researcher proposes the relevant process of fabric rendering. The whole process includes four parts: digital fabric creation, texture mapping, texture redesign and color scheme, and realistic rendering. Specifically, the establishment of digital fabric is mainly to scan the fabric with high definition through digital scanning equipment to form a high definition base map. Second, the base map is processed by Substance Alchemist tool and AI algorithm to generate normal, yarn texture height and other contents, and impurities, shadows and other contents on the base map can be removed to finally generate texture maps. Then, relevant software is used to redesign the texture map and adjust the color matching of the fabric, such as Adobe IIIustrator, Adobe Dimension, etc. This step is related to the diversity and accuracy of the fabric. Finally, by matching the size, light, and camera pigment of the fabric, and then through PBR rendering, the conventional PBR rendering is mainly applied to fabrics such as cotton fabrics and wool fabrics, and the special PBR rendering is mainly applied to silk fabrics with shiny and high linear requirements. In the specific implementation process, relevant algorithms in the Unity Shader tool are applied to render. Figure 18 shows the silk fabric effect in the fabric rendering [14].

Figure 18

Fabric rendering – silk fabric effect.

3.1.5 Flexible 3D garment simulation model

The establishment of the 3D simulation model of a garment flexible body is the core content of this topic, and also an important link for garment enterprises to actively design and serve key customers. At present, there are many methods for 3D modeling of clothing. This topic adopts the point line surface trinity modeling method, which is time-saving and effective. Therefore, according to the previous chapters and the particularity of 3D modeling of clothing flexible body, the researchers propose the flow of 3D clothing simulation model of clothing flexible body suitable for this topic, which is the flow of 3D clothing simulation model of flexible body [15].

Specifically, the whole process is divided into five modules. The first module is a customized manikin. According to the manikin designed using the SMPL model method, the corresponding size of the target clothing, such as women’s 160/68 A and men’s 180/92 A, is matched. At the same time, according to the characteristics of the target clothing, the design of the manikin is customized, such as a little wider shoulder, a little more waist circumference, a little bigger chest circumference, etc. The second module is to select the data points of the clothing parts. This module exactly selects multiple data points reflecting the clothing shape according to the customized manikin and corresponding to the needs of the target clothing, as shown in Figure 19. The clothing shape is set as the women’s bodice top, and the data points of the target clothing shape are selected. The selected position points are, respectively, 10 cm below the armpit, mastoid, side waist point, belly button, and neck center. The purpose is to better control the surface modeling of the clothing form after the connection is formed and reduce the number of unnecessary surfaces. There are four surfaces in the figure, and the left and right clothing pieces are distinguished by different colors [16]. The third module is to draw the internal and external line types of clothing on the basis of selecting the data points of clothing form parts in the second part, as shown in Figure 20. It mainly uses the curve equation to calculate from point to line and constructs the structure in terms of cloth topology, wrinkle shaping, etc., to form a closed-loop internal and external line type structure of clothing. This is also the initial clothing form master. The line type parameters are obtained by calculating the XYZ axis of the curve equation and each part point connected into line type. In this module, we should consider the rationality of clothing attachment to the mannequin and the reflection of a normal state under the action of gravity. At the same time, in this step, we should also consider the characteristics of fabric, force direction, structure, pleating volume, and other factors.

Figure 19

Point selection of target clothing shape.

Figure 20

Draw the inner and outer lines of clothing.

The fourth module is to build the clothing surface model [15]. Specifically, this step is the most complex and key step in the whole model construction, mainly including fabric stress structure analysis, pressure fitting points, fabric mapping and property rendering, screen Z-axis carving, low mold foreskin, etc. The fabric stress structure and pressure fitting point are mainly carefully drawn on the clothing softness and fitting degree, so that the bending degree and hardness of the fabric can maintain a certain state of confrontation, so as not to lose the sense of support of the clothing form, as shown in Figure 21. Cloth mapping and property rendering are mainly based on the previous cloth rendering for UV mapping. According to the shape effect of the target clothing, the pre-rendered fabric map is placed in the clothing model to make it fit. The space between the human body and the cloth covered by the cloth can be replaced by the conventional shadow mapping algorithm, which can make the space between the cloth and the human body more fit and more realistic. Screen Z-axis carving cannot be clearly drawn based on the relationship between the side seam of the target clothing form, the upper edge of the clothing, and the human model. Generally, we use the XY axis direction, so we use the Z-axis view of screen coordinates for interpolation and careful drawing, as shown in Figure 22. The low mode foreskin simulates the clothing effect to a certain extent through the lighting effect and depth buffer algorithm, so that the contrast between the clothing and the front and back pixels of the human body can form a certain visual gap and enhance the stereoscopic sense.

Figure 21

Adjustment of fabric stress structure and fitting point.

Figure 22

Screen Z-axis engraving (interpenetration).

The fifth module is engine rendering, which is used to work on lighting, materials, clothing shape effects, and other aspects through the unity rendering engine. As shown in Figure 23, we have constructed the primary clothing form. Through the mapping relationship between the position point coordinates (XYZ) and the texture coordinates, as well as the calculation of the lighting model, the middle form of the clothing is formed. It can be seen that the fabric is a grid texture, which cannot meet the needs of researchers. Therefore, we render again through the material characteristics, background environment (reflective, etc.), lighting model, and depth buffer algorithm. After several rendering adjustments, the final shape of the target garment is finally constructed. The above is the specific workflow of the rendering engine. Due to the different shapes of the target clothing, there will be some differences in the workflow [17].

Figure 23

Garment model rendering – engine flow diagram.