SIFT特征提取算法的FPGA实时实现  

Real-time Implementation of Sift Feature Extraction Algorithms in FPGA

作　　者：陆恺立 耿俊梅[2] 顾国华[1] 钱惟贤[1] 徐富元[1] 

机构地区：[1]南京理工大学电光学院,江苏南京210094 [2]济源职业技术学院,河南济源454650

出　　处：《电视技术》2014年第13期54-58,61,共6页Video Engineering

基　　金：国家自然科学基金项目(61271332);江苏省"六大人才高峰"支持计划项目(2010-DZXX-022)

摘　　要：为了在运动目标检测与跟踪系统中能够实时地提取目标的特征点,从而完成图像实时匹配操作,提出了在FPGA实现SIFT特征提取算法。该算法采用SRAM复用技术简化程序,合理设计FPGA各模块结构。此外,该算法采用定点小数来保证算法的精度要求。整个算法在Virtex-5硬件平台上实现,采用verilog语言进行程序的编写和调试。结果分析表明,优化后的SIFT算法能够稳定地在FPGA上实现,同时算法的复杂度得到了降低并达到了精度要求,且具有良好的实时性。In order to extract real-time feature point of target in the system of moving target detection and tracking system,thus completing the real- time image matching operation, the feature extraction algorithm is proposed to achieve SIFT in FPGA. The SRAM reuse technology is used to simplify the procedures,and the module structure of FPGA is designed reasonably. Besides the fixed-point decimal is used to ensure the accuracy of algorithm. The whole algorithm is realized in the Virtex-5 hardware platform and the verilog language is used for the writing and debugging of programs. The result shows that the optimized SIFT algorithm can be realized in FPGA stably,but also the complexity of algorithm is reduced and accuracy is achieved,which imshows the algorithm has a good real time feature.

关 键 词：FPGA 特征提取算法 SRAM VIRTEX-5 实时性 

分 类 号：TN391[电子电信—物理电子学]