Search

CN-122021283-A - Clothing cloth dynamic simulation method and system based on convolutional neural network

CN122021283ACN 122021283 ACN122021283 ACN 122021283ACN-122021283-A

Abstract

The invention discloses a clothing cloth dynamic simulation method and a system based on a convolutional neural network, which relate to the technical field of computer simulation and comprise the following steps: in the dynamic simulation process of the character executing squatting or bending actions, the motion change data of a plurality of joints of the character are obtained, and action compression marks are generated according to the joint motion change speed. The method and the device ensure that the shape change of the cloth in the actions such as squatting, bending and the like maintains the time continuity through the action compression mark and the refresh rhythm control, and reasonably distributes the local change to the adjacent fold areas by combining the area maintenance constraint, so as to stabilize the shape of the cloth at the bending part, and simultaneously realize the staged rebound limitation based on the contact position sequence, so that the cloth is gradually restored to the natural fold state from inside to outside, the local aggregation and collapse are avoided, and the naturality and the visual credibility of the dynamic simulation are improved.

Inventors

  • WANG HONG
  • Wang Runbang
  • WANG RUNJI

Assignees

  • 坦博尔集团股份有限公司

Dates

Publication Date
20260512
Application Date
20260123

Claims (10)

  1. 1. The clothing cloth dynamic simulation method based on the convolutional neural network is characterized by comprising the following steps of: In the dynamic simulation process of the character executing squatting or bending actions, obtaining movement change data of a plurality of joints of the character, and generating action compression marks according to the movement change speed of the joints; based on the action compression marks, subdividing the time sections corresponding to the action compression marks, unifying cloth picture refreshing rhythms in the subdivided time sections, and introducing continuous transition treatment between adjacent pictures; Introducing area maintenance constraint to a cloth area positioned at a position adjacent to bending of a human body on the basis of unified cloth picture refreshing rhythm, and distributing the variable quantity exceeding a preset range to adjacent fold areas when the area change of the cloth area exceeds the preset range; on the basis of keeping the constraint of the area, determining the contact position sequence of the cloth and the human body, preferentially adjusting the displacement change of the cloth close to the human body according to the contact position sequence, and expanding the cloth area at the outer side; Based on the contact position sequence, the cloth area which is easy to form aggregation in the dynamic simulation process is subjected to staged rebound limiting treatment, so that the cloth form is gradually restored to a natural fold state.
  2. 2. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 1, wherein the action compression mark generation step is as follows: continuously tracking a plurality of joints participating in lower limb flexion and extension and trunk rotation, and collecting the space displacement change information of each joint in continuous frames to form a joint movement change data sequence; calculating the articulation change speed on the basis of the articulation change data sequence, and mapping the articulation change speed to a corresponding time position to form the time distribution of articulation change; extracting time zones with synchronous changes and concentrated change speeds based on the time distribution of the joint movement changes, and refining the time zones to obtain sub-time zones; and correlating the sub-time zone in which the changing speed is concentrated with the cloth covering position in the action process, and generating an action compression mark.
  3. 3. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 2, wherein the sub-time zone of joint movement change speed concentrated change is determined by analyzing the synchronicity of joint movement change speeds at adjacent time positions, and the action compression marks are generated only in the time zone where the synchronicity change continuously exists, so that the action compression marks are concentrated to correspond to the cooperative change stage of the multiple joints of the person.
  4. 4. The dynamic simulation method of clothing and cloth based on convolutional neural network according to claim 2, wherein the processing step of the time section based on the action compression mark is as follows: reading action compression marks, extracting boundary information of a corresponding time section, mapping the boundary information of the time section to a time axis of cloth dynamic simulation, and establishing an index relationship between the time section and a cloth picture sequence; counting time intervals between adjacent pictures in the time zone on the basis of the index relation, and subdividing the time zone according to the time interval distribution to form subdivided time zones which are arranged continuously; Setting a unified cloth picture refreshing rhythm in the subdivided time zone, and performing time alignment treatment on the original cloth pictures in the subdivided time zone to enable the cloth pictures to be arranged according to the refreshing rhythm; On the basis of uniform cloth picture refreshing rhythm, continuous transition treatment is introduced between adjacent cloth pictures, so that the cloth form forms a continuous and stable change process in a subdivided time section.
  5. 5. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 4, wherein when continuous transition treatment is introduced between adjacent cloth pictures, the state of the cloth form corresponding to the adjacent pictures is used as a transition boundary, the form change between the adjacent pictures is gradually expanded under the unified cloth picture refreshing rhythm, and the time consistency of the continuous transition treatment and the internal treatment of the segments is maintained at the subdivided time segment boundary, so that the cloth pictures form a continuous connected change sequence in the whole time segment.
  6. 6. The dynamic simulation method of clothing cloth based on convolutional neural network as set forth in claim 4, wherein the step of introducing area retention constraint on the basis of unified cloth picture refreshing rhythm is as follows: determining the bending adjacent positions of the human body by combining the bending angles of the joints and the curvature distribution of the surface of the human body at each refreshing time position, mapping the bending adjacent positions of the human body to corresponding cloth areas, and extracting the geometric description of the cloth areas; On the basis of geometric description of a cloth area, calculating the area variation between adjacent refreshing time positions, constructing area retention constraint by a reference area, and determining a preset range according to the joint movement variation speed; When the area variation exceeds the preset range, converting the variation exceeding the preset range into form adjustment quantity, distributing the form adjustment quantity to adjacent fold areas according to the adjacent relation of the cloth areas, and simultaneously applying corresponding reverse adjustment to the cloth areas; and keeping the area keeping constraint consistent with the unified cloth picture refreshing rhythm, so that the allocated form adjustment quantity is continuously introduced between adjacent refreshing time positions.
  7. 7. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 6, wherein the preset range is adjusted synchronously according to the joint movement change speed, and the change amount exceeding the preset range is distributed according to the shared boundary length between the cloth area and the adjacent fold area and the fold trend consistency, so that the form adjustment is transferred along the fold extension direction, and the continuous change is kept between the adjacent refreshing time positions.
  8. 8. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 6, wherein the steps of determining the order of contact positions and adjusting the cloth displacement based on the area retention constraint are as follows: Determining a contact position set of the cloth and the human body on the basis of the cloth morphological state and the human body surface morphological state at each refreshing time position, and recording the contact duration and the contact area change of the contact position to form contact description information; on the basis of the contact description information, determining the sequence of contact positions according to the contact duration time, the contact area change and the bending direction of the human body, and sequentially arranging the sequence of the contact positions from the position close to the human body to the outer side; the cloth displacement change close to the human body is preferentially adjusted according to the contact position sequence, so that the cloth displacement change close to the human body is gradually unfolded and kept continuous along the surface direction of the human body; after the cloth displacement change close to the human body is completed, the cloth area positioned on the outer side is unfolded, so that the cloth area on the outer side and the area close to the human body form a continuous space hierarchy relation.
  9. 9. The dynamic simulation method of clothing cloth based on convolutional neural network according to claim 8, wherein the step of sequentially performing the staged rebound restriction process based on the contact position is as follows: After finishing the cloth displacement change adjustment under the contact position sequence, analyzing the state of the cloth form along the contact position sequence, marking the cloth area which is easy to form aggregation, and establishing a corresponding area tracking relation; On the basis of a cloth area which is easy to cause form aggregation, dividing stages of the staged rebound limitation processing according to the contact position sequence, and setting rebound target forms and rebound limitation amplitude for each stage; Under the unified cloth picture refreshing rhythm, the cloth areas which are easy to form aggregation are subjected to rebound limiting treatment step by step according to the stage sequence, so that the rebound process is pushed to an outward unfolding area from a human body area; and the continuity of stage rebound is maintained between adjacent refreshing time positions, so that the cloth form is gradually restored to a natural fold state under the action of stage rebound limiting treatment.
  10. 10. The clothing cloth dynamic simulation system based on the convolutional neural network is used for realizing the clothing cloth dynamic simulation method based on the convolutional neural network as claimed in any one of claims 1-9, and is characterized by comprising an action compression marking module, a time sequence refreshing control module, an area holding constraint module, a contact sequence regulation and control module and a rebound limitation recovery module: The action compression marking module is used for acquiring movement change data of a plurality of joints of the person in the dynamic simulation process of the person executing squatting action or bending action and generating action compression marks according to the movement change speed of the joints; the time sequence refreshing control module is used for conducting subdivision processing on time sections corresponding to the action compression marks based on the action compression marks, unifying cloth picture refreshing rhythms in the subdivided time sections, and introducing continuous transition processing between adjacent pictures; The area keeping constraint module is used for introducing area keeping constraint to the cloth areas positioned at the positions adjacent to the bending of the human body on the basis of the unified cloth picture refreshing rhythm, and distributing the variation exceeding the preset range to the adjacent fold areas when the area variation of the cloth areas exceeds the preset range; The contact sequence regulating and controlling module is used for determining the contact position sequence of the cloth and the human body on the basis of the area keeping constraint, preferentially regulating the displacement change of the cloth close to the human body according to the contact position sequence, and expanding the cloth area positioned at the outer side; And the rebound limiting and recovering module is used for carrying out staged rebound limiting treatment on the cloth region which is easy to form aggregation in the dynamic simulation process based on the contact position sequence, so that the cloth form is gradually recovered to a natural fold state.

Description

Clothing cloth dynamic simulation method and system based on convolutional neural network Technical Field The invention relates to the technical field of computer simulation, in particular to a clothing cloth dynamic simulation method and system based on a convolutional neural network. Background The dynamic simulation of clothing cloth based on the convolutional neural network is to combine the idea of twin simulation, learn and express the dynamic appearance rules of deformation, fold, swing and the like of the clothing cloth in the process of movement, stress or gesture change by using the convolutional neural network, and construct a virtual mapping body corresponding to the real cloth state on the basis, so as to realize continuous deduction and reproduction of the cloth change along with time. According to the method, a large number of time sequence features formed by the cloth under different action, environment conditions and interaction scenes are extracted, the response relation between the physical cloth and the virtual cloth is synchronously depicted under the twin simulation framework, and the inherent association between local texture change and overall morphological evolution is captured, so that the simulation process can quickly generate dynamic expression which is highly consistent with the actual cloth behavior according to the input action or state change. In the whole, the simulation mode emphasizes the learning and reproduction of the dynamic response rule of the cloth under the support of the twin simulation mechanism, focuses on the visual continuity, the form coordination and the stable expression of the follow-up characteristics, and is suitable for application scenes with high requirements on the dynamic authenticity of the cloth, such as virtual fitting, digital clothing display, animation production and the like. The prior art has the following defects: In the prior art, when a person completes the cooperative motions of a plurality of joints such as squatting and bending, the cloth state can be changed rapidly at the same time, and the problem of local response unbalance is easy to occur in the simulation process. The method is characterized in that abnormal shrinkage occurs in certain areas of the cloth in a very short time, the shape is concentrated to local parts rapidly, and a bulk structure which is not consistent with the stress and deformation rules of real clothing is formed. The abnormal phenomenon can destroy the original continuous fold change of the cloth, so that the whole appearance of the garment presents a sudden collapse state, and is particularly obvious in the action connection stage. Because the problems are frequently generated at the bending part of the human body, the problems are more easily amplified and identified under close-up observation or close-up lens, and the naturalness and the credibility of the dynamic simulation of the clothing are seriously affected. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide a clothing cloth dynamic simulation method and a clothing cloth dynamic simulation system based on a convolutional neural network so as to solve the problems in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the clothing cloth dynamic simulation method based on the convolutional neural network comprises the following steps: in the dynamic simulation process of the character executing squatting or bending actions, motion change data of a plurality of joints of the character are obtained, and action compression marks are generated according to the joint motion change speed and are used for marking time sections in which quick shrinkage of cloth is easy to occur in the dynamic simulation process; based on the action compression marks, subdividing the time sections corresponding to the action compression marks, unifying cloth picture refreshing rhythms in the subdivided time sections, and introducing continuous transition treatment between adjacent pictures at the same time so as to reduce the mutation degree of the cloth form in the time dimension; On the basis of unified cloth picture refreshing rhythm, introducing area maintenance constraint for a cloth area positioned at a position adjacent to bending of a human body, and distributing the variation exceeding a preset range to adjacent fold areas when the area variation of the cloth area exceeds the preset range so as to inhibit abnormal shrinkage of the cloth area; on the basis of keeping the constraint of the area, determining the contact position sequence of the cloth and the human body, preferentially adjusting the displacement change of the cloth cl