Dissertation > Excellent graduate degree dissertation topics show

Research on Heterogenenous Accelerating for Collision Detection in Virtual Reality

Author: HuangTao
Tutor: ZhanYinWei
School: Guangdong University of Technology
Course: Applied Computer Technology
Keywords: Collision detection Heterogeneous programming Hierarchical bounding box Graphics Processor
CLC: TP391.9
Type: Master's thesis
Year: 2011
Downloads: 102
Quote: 0
Read: Download Dissertation

Abstract

GPU Computing development in recent years, good conditions for the virtual reality field algorithm improvements, provide opportunities for parallelization to solve the efficiency problem. CPU GPU heterogeneous computing environment is to provide a better method for virtual real-time problems in the real algorithm. Virtual reality technology is widely used in computer games, computer-aided instruction, computer-aided design and other fields, and there is a huge development in the last decade. As people increasingly realistic and highly interactive requirements scenes of virtual reality systems, the system calculates the amount of data also improved significantly. Wherein the collision detection computation of the objects in the scene is an important part of the entire calculation. Therefore, the real-time collision detection calculation is often the key to the efficient implementation of the entire virtual reality system. Collision detection is to determine whether there is one or a plurality of intersections the process two or more objects. Computer graphics, collision detection is an important branch of the field of virtual reality medical simulation. In many cases, the moving object is not allowed to move in the virtual scene map with the penetration of certain objects in real-time of the collision detection algorithm is the key to solving such problems. Bounding box technology and space partitioning technique are two basic methods for collision detection algorithm, they reduce the computational complexity of the collision detection process from two aspects. Collision detection, even though in recent years research has been more valuable results, but as the scene with the model complexity increasing scale, real-time and accuracy requirements of the users of human-computer interaction is also increasing, the core algorithm the need for higher efficiency. However, the level of classic segmentation techniques can not be done directly using the GPU accelerated computing technology and space bounding box, and need to be improved to accommodate the GPU hardware multithreading. This paper focuses on the virtual scene collision detection problems, analysis and research to improve the detection efficiency in complex scenes, based on in-depth understanding of the various basic collision detection algorithm, combined with the NVIDIA CUDA general-purpose computing platform, heterogeneous programming ideas from next several ways to improve the algorithm: First, ball hierarchical tree, and the triangle data collection data structure surrounded by objects in collision detection algorithm model for detailed analysis, based on the idea of ??CPU GPU heterogeneous computing problem decomposition, layered discuss the wide stage preliminary testing with a large-scale multi-object detection. Collision detection algorithm with a narrow phase collision detection algorithm in a wide stage improved the implementation of the CPU side portion GPU end to accommodate the requirements of the input data, separating out the computationally intensive Device (GPU) end portion; Then, create the intersection test for GPU processing logic, then the scene intersection test between the objects mapped to CUDA thread execution model; Finally, according to the model of the GPU memory efficiency optimization. Specifically, the collision detection in a wide stage, the CPU is responsible for Construction and traverse of the bounding spheres of the hierarchical tree, GPU is responsible for the collection of an object bounding sphere hierarchical tree with another object triangle intersection tests in parallel, excluding impossible to intersect the triangle set to give potentially disjoint sets; data as input in the narrow phase collision detection of more than one stage of the GPU, triangular collection precise parallel intersection test, and ultimately reach the test results. The first end of the CPU for multi-object scene to scene objects collection grouped and executed in parallel in order to follow-up GPU, GPU then trip the way to do object detection, and improve efficiency in the implementation of concurrent Kernel. Serial Rapid algorithms and GPU parallel OBBs CG algorithm comparison experimental results show that the the heterogeneous collision detection algorithm more outstanding performance in real-time, and with the scene model complexity enhanced processing time to maintain a gentle growth, while having a higher stability. , The present method also shows the good real-time in a multi-object scene.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Computer simulation
© 2012 www.DissertationTopic.Net Mobile