混合粒度可重构视觉处理器关键技术研究

项目名称： 混合粒度可重构视觉处理器关键技术研究

项目编号： No.61274131

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 尹首一

作者单位： 清华大学

项目金额： 92万元

中文摘要： 随着计算机技术和集成电路技术的发展，基于计算机视觉技术的应用（包括人机交互、虚拟现实、目标识别等）正极大地影响着这人们的生产与生活方式。不断涌现的新型计算机视觉应用对支持视觉处理的视觉处理器提出了实时处理能力、低能耗以及算法灵活配置的全面要求。本课题着眼于综合解决视觉处理所需的实时处理、低能耗和灵活算法配置这三个问题，提出采用混合粒度的可重构计算架构来实现视觉处理器，通过引入细粒度和粗粒度混合的异构计算核，以及基于计算核构建多核可重构阵列来提高视觉处理效率，增强算法的灵活配置能力；通过构建资源感知的负载均衡机制来降低系统能耗。本课题的研究成果一方面可以促进视觉处理技术的发展，为日益涌现的计算机视觉应用提供硬件平台，满足多种应用的需求；另外一方面本课题将探索混合粒度可重构处理器的设计技术，这是对新型处理器架构的有益尝试，相关研究成果将有助于我国通用处理器的研究。

中文关键词： 可重构计算；机器视觉；SoC设计；；

英文摘要： With the development of computer technology and integrated circuit technology, the computer vision based applications, including the human-computer interaction, virtual reality, target recognition, greatly influence people's modes of production and life. The increasing emergence of new computer vision applications requires the vision processor to support real-time processing, low power and algorithm configurability. This project focuses on a comprehensive solution to the issues of real-time processing, low power and flexible algorithm configuration. We proposes a mixed-granularity reconfigurable computing architecture for vision processor, by adopting mixed-granularity heterogeneous processing element(PE) and building the PE array, to improve the efficiency of visual processing and enhance the flexible algorithm configuration capability. Besides, we propose a resource-aware load balancing mechanism to reduce energy consumption. The research of this subject can contribute to the development of visual processing, which provides the hardware platform for the increasingly emerging computer vision applications and meets the needs of a variety of applications. On the other hand, we will investigate the design of mixed-granularity reconfigurable processor, which will help the study of high-end general-purpose processo

英文关键词： Reconfigurable Computing；Computer Vision；SoC Design；；