Research And Implementation Of Key Technologies In3D Virtual Clothing Animation

Based on the cloth simulation technology,3D virtual clothing animation has a great prospect in urban planning, architectural design, aerospace, film entertainment, education and other areas, and has been researched over30years. Particularly, when it is applied to the online dressing systems, customers can experience real dynamic clothing dressing effect before their computers to determine things like whether the clothing size is fit or not. This will revolutionize the current online clothing sales model. In recent years, researchers have gradually adopted the physically-based models for simulation, which can get more realistic effects. However, because its computation is too complex to meet real-time requirements on most current computers, the physically-based clothing animation still has not been widely used in practice.This paper first describes the background and history of3D virtual clothing animation system, and studies seven key technologies such as cloth modeling, numerical integration, collision detection, self-collision detection, dressing method, dynamics simulation architecture, GPU parallelization, and so on, then proposes eight new improvements and solutions to accelerate the system performance, which are listed in the followings.(1) In order to solve the over-stretching problem in Provot’s mass-spring cloth model more effectively, we propose an adaptive method to update cloth vertices, which effectively enhanced the performance of clothing animation.(2) By using the Verlet method to forecast the initial value in trapezoidal integration method, we propose a more efficient numerical algorithm, with less error and faster convergence.(3) According to the characteristics that collision elements are usually more concentrated in certain areas, we propose the concept of Local Collision Impact Region (LCIR), and effectively accelerate the performance of collision detection.(4) Based on the bounding volume hierarchy (BVH), we propose a coherent collision detection algorithm to avoid the redundant detection in the middle and upper level of BVH, which can further accelerate collision detection process.(5) To enhance the performance of self-collision detection, we propose the second subspatial subdivision method to reduce the potential collision pairs without losing any detection accuracy.(6) In order to improve the applicability of clothing animation, we propose a more user-friend automatic dressing method, which can be easily applied to network clothing sales platform.(7) Based on the object-oriented thinking, we implement the clothing animation system with better scalability, reusability, which can generate over10key frames per second, and can be easily used for simulating more objects. We use the C++language for programming and OpenGL graphics library for rendering, not using any other third-party kits. By doing this, we can easily meet the requirement of multiple platforms such as Windows, Linux. All the valid code over15,000lines is independently programmed by me, excluding notes and blank lines.(8) Based on the CUDA3.1, we propose the CPU-GPU hybrid architecture for simulating cloth in parallel, which can generate over100key frames per second, thus further accelerate the performance of clothing animation.With the improvement of above work, the performance of our clothing animation has been greatly enhanced than other open simulation technology, and can be easily used in more areas. Another advantage of this work is our solutions are more practical for network virtual dressing room. We believe that the3D virtual dressing room will be used in network clothing e-commerce platform in the near future.