双路视频融合子系统的研究与设计

发布时间：2018-05-14 04:38

本文选题：可重构处理器 + 视频融合　；参考：《天津大学》2016年硕士论文

【摘要】：随着多媒体技术的快速发展,高清甚至超高清视频技术得到普及,视觉信息处理系统中的数据吞吐量以及设计复杂度也在急剧上升,使用传统的专用集成电路以及通用处理器很难满足高性能、高灵活性的需求。然而,可重构处理器因其较高的能效比、丰富的运算资源、灵活的互连形式,能够有效的解决上述问题,并且成为视觉信息处理领域中的新的研究热点。本文介绍了双路视频融合子系统的整体架构和实现方式:首先将多源摄像头输入的视频图像输入给可重构处理器,可重构处理器根据Harris等经典图像拼接算法计算出仿射矩阵,并将仿射矩阵输出给融合子系统模块。融合子系统利用DDA(Digital Differential Analyzer)数值微分分析的方式将两幅视频图像拼接成一幅宽视野的图像,并采用片内乒乓读写的方式缓存到DDR2中。然后,将拼接缓存后的视频图像进行降噪处理。本文提出了一种基于半自适应阈值的边缘保护各向异性滤波算法,该算法相比传统的各向异性滤波算法,峰值信噪比PSNR提高了35%,结构相似比SSIM提高了10%,因此不仅提高了滤波降噪性能,还有效的保护了边缘细节信息。同时,整个系统可以通过人机交互控制界面OSD进行系统配置以及功能选择。OSD(On Screen Display)模块由Microblaze软核作为主控核,产生控制信号并驱动点阵显示模块产生相应的菜单画面。该系统的开发平台为Xilinx FPGA,型号为Virtex-5 XUPV5 LX110T。此外,由于整个系统的数据量大,需要仿真的功能多,该系统搭建了基于VMM(Verification Methodology Manual)方法学的验证平台,该验证平台使用SystemVerilog与C语言进行编写,仿真环境为Sysnopsys公司的VCS 2011。验证环境层次化,降低了验证平台的设计复杂性。同时,还使用了随机约束向量作为驱动向量,尽可能使功能覆盖率达到100%。相比于传统的动态仿真,该验证方式仿真速度快,覆盖率高,且便于维护与移植。
[Abstract]:With the rapid development of multimedia technology, high-definition and even ultra-high-definition video technology has been popularized, and the data throughput and design complexity of visual information processing system are also rising rapidly. It is difficult to meet the requirement of high performance and high flexibility with traditional ASIC and general purpose processor. However, reconfigurable processors can solve these problems effectively because of their high energy efficiency ratio, rich computing resources and flexible interconnection forms, and become a new research hotspot in the field of visual information processing. This paper introduces the architecture and implementation of the dual-channel video fusion subsystem. Firstly, the video image input from the multi-source camera is input to the reconfigurable processor. The reconfigurable processor calculates the affine matrix according to the classical image mosaic algorithm such as Harris. The affine matrix is outputted to the fusion subsystem module. The fusion subsystem uses the method of DDA(Digital Differential Analyzer numerical differential analysis to splice two video images into a wide-field image and cache it into DDR2 by in-chip ping-pong reading and writing. Then, the concatenated video image is de-noised. In this paper, an edge protection anisotropic filtering algorithm based on semi-adaptive threshold is proposed, which is compared with the traditional anisotropic filtering algorithm. The peak signal-to-noise ratio (PSNR) is increased by 35%, and the structure similarity is improved by 10% than that of SSIM, so it not only improves the performance of filtering and noise reduction, but also effectively protects the edge detail information. At the same time, the whole system can be configured by man-machine interactive control interface (OSD) and function selection. The Microblaze soft core is used as the main control core to generate the control signal and drive the dot matrix display module to generate the corresponding menu screen. The development platform of the system is Xilinx FPGA, and the model is Virtex-5 XUPV5 LX110T. In addition, due to the large amount of data of the whole system and the need for more functions of simulation, the system has built a verification platform based on VMM(Verification Methodology Manual methodology. The verification platform is written in SystemVerilog and C language, and the simulation environment is VCS 2011 of Sysnopsys Company. The verification environment is hierarchical and the design complexity of the verification platform is reduced. At the same time, the random constraint vector is used as the driving vector to make the function coverage as high as 100. Compared with the traditional dynamic simulation, this verification method has the advantages of high speed, high coverage and easy maintenance and transplantation.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】