Research And Implementation On Collision Handling For Physically-based Cloth Simulation

Phy-sically-based cloth simulation.,which studies how to use computer to animate the movement of cloth in the real world,is a popular research topic in computer graphics area in recent decades.With the development of computer graphics technology and computer hardware,cloth simulation has been widely applied in various fields such as computer games,movies and electronic commerce.In the premise of guaranteeing the quality of simulation.how to improve the efficiency of simulation system has always been one of the most important research tasks in physics-based cloth simulation.Collision handling,as an important step of physically-based cloth simulation.plays a key role for the simulation system to animate the real world.which is also the time bottleneck of the whole system.Therefore.efficient collision handling algorithms will effectively improve the performance of cloth simulation,which has great values.In addition,the development of parallel hardware such as GPU also brings new research directions for exploring more efficient parallel algorithms.This dissertation focuses on how to improve the efficiency of collision handling algorithms in cloth simulation.We introduce many novel efficient methods for collision detection and response,Moreover,a GPU-based cloth simulation pipeline.I-Cloth,is proposed.The main contributions of this thesis are summarized as follows:· We present a self-collision culling met.hods via enhanced dual-cone test.Self-collision culling can eliminate many triangle pairs without collision at the beginning of collision detection.which can reduce the computing burden of the following exact collision de-tection.The most common culling method is normal cone test.However,normal cone test is time-consuming,as it performs contour test via projecting the contour edges on planes and checking for whether the project.ed contour has self-intersections.In order to accelerate the contour test,we propose a new contour test scheme using binor-mal cone and Dual-Cone Theorem to replace many edge-edge intersection tests.This scheme can improve the performance of original cont.our t.est without collision omission.Through combining the scheme into normal cone t.est,we introduce sufficient criteria for determining whether a mesh exhibits self-collisions.Moreover,we combine our culling criteria with bounding volume hierarchies and present a hierarchical traversal scheme for b ounding volume test trees.which can significanty reduce the number of redundant tests performed.Experiments show t.hat our dual-cone culling algorithm can accurately detect all self-collisions and has considerable speedup compared with other self-collision detection methods.· We present another self-collision culling method via unpro.jected norma.l cone test,which is an improvement of the contour test scheme introduced in last paragraph.Although the previously int roduced method has improved contour test,it still needs to project contour edges on 2D planes and do many edge-edge intersection tests.For this problem,we do further improvements.We first design new contour test conditions by checking whether the polygon,which is formed by the contour edges projected on a 2D plane.is a star-shaped polygon.According to the formulas used in the test conditions,we can directly use unprojected 3D contour vertex to perform our star-shape poly gon test.Based on the unprojected cont,our test,we construct our novel unprojected normal cone test and combine it with bounding volume hierarchies to form a complete self-collision culling algorithm.Compared with other self-collision culling methods,this method does not involve any projection computations and edge-edge intersection tests,which can improve the efficiency of self-collision culling.Moreover.we additionally propose continuous unprojected normal cone test to quickly perform exact collision detection.Experiments show that our method has obvious performance advantage for discrete and continuous collision detectionwhich can accelerate the entire collision detection system.· We present a non-linear GPU-based impact zone solver to resolve the penetrations and a GPU-based cloth simulation system.The underlying collision response algorithm groups all the intersecting element pairs into impact zones and computes non-linear optimization problem for each impact zone.In order to quickly solve this optimization problem on GPU,we propose an algorithm based on non-linear gradient descent solver.And we use Jacobi preconditioning and Chebyshev acceleration method to improve the convergence rate of gradient descent.Compared with previous methods,our solver demonstrates higher stability and is more suitable for GPU parallelism.With this GPU-based solver,all the impact zones can be processed in parallel and the position of vertex in each impact zone can be concurrently updated.Additionally,we construct a GPU-based cloth simulation system with a comb ination of our GPU-based non-linear impact zone solver and other GPU-based methods.To some extent,our system can achieve real-time interaction in many simulation scenarios.Experiments show that our non-linear impact zone solver demonstrates higher stability and faster convergence and the GPU-based cloth simulation system has better performance over other systems.