时变数据的实时体绘制加速算法优化  

Accelerated Real-Time Volume Rendering Algorithm Optimization for Time-Varying Data

作　　者：贺承浩 金西[1] 郑琳琳[1] 刘子恒[1] 王浩原 

机构地区：[1]中国科学技术大学物理系微电子学教研室,合肥230026

出　　处：《计算机辅助设计与图形学学报》2014年第2期314-319,328,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：中国科学院战略性先导专项(XDA04040202-4);中央高校基本科研业务费专项资金(WK2030020010);国家基础科学人才培养基金(J1103207)

摘　　要：现有的体绘制成像技术受到硬件及算法的限制,对高速采集的时变数据绘制需要漫长的预处理时间,难以满足实时便携的要求.为此提出一种基于存储优化的Shear-Warp数据整序算法.该算法通过预判视线方向避免了原始数据的重复存储及多次缓存;在FPGA硬件加速的基础上克服了与存储器件的交互瓶颈,实现对图像和观察角度变化的实时动态响应.实验结果表明,在自主研发的"双子星"系列开发板100MHz工作频率下,完成一帧256×256×256的三维图像体绘制计算时间小于12.34ms,帧率达到81.0帧/s,且不需要数据预处理过程,可以实现对时变数据的实时动态显示,适合于便携超声等应用领域.Due to the limitations of hardware and algorithms, it is time-consuming to preprocess high- speed time-varying data, which makes it difficult to meet real-time and portable applications. In this paper, we proposed a novel data collation method based on storage optimization for Shear-Warp algorithm. Through prejudgment of sight direction, it successfully avoided duplicate storage of original data and multiple caches. The FPGA acceleration is launched to overcome the bottleneck of memory data exchange, thus the real-time response of the shifting images and observation angles are achieved. The experimental results on the independently developed ＂Gemini＂ board, with the 256 × 256 × 256 three-dimensional ultrasound data processing in a 100 MHz working frequency, demonstrated that the system could finish volume rendering per frame within 12.34ms, i.e. a frame rate of 81.0 fps. The proposed system can achieve dynamic display for time-varying data without preprocessing, which is suitable for portable device like ultrasound scanner.

关 键 词：时变数据 体绘制Shear-Warp^FPGA加速 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]