面向虚拟着装和试衣仿真的碰撞检测与碰撞响应算法研究

This project aims at efficient collision detection and physically realistic collision response techniques for virtual clothing and try-on simulation. We plan to accelerate collision detection using algebraic filtering for complex deformable objects, to design workload partitioning and scheduling algorithms for CPU/GPU hybrid computing platform, to study penetration depth computation and its continuity using machine learning techniques and to investigate continuous collision response under multiple contacts. This research is expected to contribute to the academic and industrial fields by achieving 10-50 speedup in collision detection for deformable objects, reducing the computational complexity for penetration depth to O(1), obtaining continuous and stable penetration depth and collision response. These research and investigation will help to improve the performance of virtual clothing simulation, and to enhance the user experience of virtual try-on in e-commerce. This project will also provide insights into the problem of collision detection and response for other research and applications, such as soft robot simulation, fluid simulation and virtual surgery.

本课题针对虚拟着装和试衣仿真中的碰撞检测与碰撞响应问题展开研究。拟研究1）利用代数分析加速复杂可变形物体的碰撞检测，2）面向CPU/GPU混合计算环境的碰撞检测任务分割与调度算法，3）利用机器学习实现渗入深度的快速计算和连续渗入深度算法，4）多接触点的碰撞响应和连续碰撞响应算法。预期可将变形物体碰撞检测和响应速度至少提高10-50倍，将渗入深度的实时查询的复杂度降低到O(1), 并且能够提供连续稳定的渗入深度和碰撞响应结果。这些研究结果将有助于实现虚拟服装动态仿真和试衣的物理真实感效果，为以互联网电子商务为代表的消费级应用，尤其是在线虚拟服装销售的实时动态展示提供算法基础。本研究也能够为可变形物体的碰撞检测和响应在其它领域的应用（如可变形机器人仿真、流体仿真、虚拟外科手术等），提供有益的参考和借鉴。

本课题针对虚拟着装和试衣仿真中的碰撞检测与碰撞响应问题展开研究。研究了以下问题1）利用代数分析加速复杂可变形物体的碰撞检测，2）面向CPU/GPU混合计算环境的碰撞检测任务分割与调度算法，3）利用机器学习实现渗入深度的快速计算和连续渗入深度算法，4）多接触点的碰撞响应和连续碰撞响应算法。针对这些问题，取得了较丰富的成果，分别发表在IEEE TVCG, IEEE VR，Pacific Graphics, IEEE ICRA, IEEE CASE等顶级期刊和会议上。这些成果有助于实现虚拟服装动态仿真和试衣的物理真实感效果，为以互联网电子商务为代表的消费级应用，尤其是在线虚拟服装销售的实时动态展示提供算法基础。本研究也能够为可变形物体的碰撞检测和响应在其它领域的应用（如可变形机器人仿真、流体仿真、虚拟外科手术等），提供有益的参考和借鉴。