雪模擬與應用Kinect和虛擬人物的互動

Abstract

我們提出一個整合系統，它包含三個子系統：雪景模擬、光線追蹤以及虛擬人物。 此系統是建立在多核心平台以及具有串流處理器的繪圖處理器上，應用平行運算技術於 各個部分，發展混合型的動態工作分配系統，提高運算的效能。它可以真實地模擬出雪 與體感控制的虛擬人物互動，並且用光線追蹤將場景畫出來。 在這個系統中，針對不同部分，提出相應的計算方法去提高效能，達到即時互動的 表現。雪景模擬將運用height map 和粒子系統來快速地模擬出真實的雪。光線追蹤將透 過 OPTIX 和CUDA 來達到GPU 的平行化。在虛擬人物模擬的方面，重點放在動作擷 取，將運用Kinect 和Skeletal Subspace Deformation 來完成。此外我們會對整合後的系統 進行最佳化，如簡化資料結構等，以提高系統的效能。 這計劃預期需要兩年時間。在計劃的第一年完成一個簡單的雪景模擬和光線追蹤的 部分，以及動作擷取和對雪景模擬加入粒子系統的處理；在第二年我們將結合雪和光線 追蹤的系統，並開始虛擬人物模擬的部分。最後我們將整合三個子系統，並做進一步的 優化。

We propose a hybrid parallel system for physically-based simulation and rendering, which contains three components: Snow Simulation, Ray Tracing, and Virtual Characters. Our system will be developed based on the new computation architecture which consists of multicore CPUs and GPUs. The system will be able to realistically simulate snow and allow users to interact with it through motion control for animating virtual characters, then the entire scene will be rendered using ray tracing. Different techniques will be applied for different components of the system in order to maximize the computation and achieve real-time performance. The snow simulation will be done through a combination of height fields and particle system to simulate realistic behaviors. Ray tracing will be performed by using Optix and CUDA to achieve GPU parallelism. For animating virtual characters, we will be using the Microsoft Kinect device with skeletal subspace deformation so as to capture the motion of a user and then to animate the virtual characters. We will integrate the three components and improve the integrated system further. This project will be completed in two years. In the first year, we will complete a simple snow simulation that is capable of basic user interactions. Then we will start to work on ray tracing, as well as adding particles into the snow simulation. In the second year, we will combine the snow and ray tracing system and start working on animating virtual characters. Finally, we will integrate all the components and optimize the system.