基于自适应SIFT算法的全局运动估计方法  

RESEARCH ON GLOBAL MOTION ESTIMATION BASED ON ADAPTIVE SIFT ALGORITHM

作　　者：张二元[1] 孙艳红[2] 

机构地区：[1]重庆电子工程职业学院物联网学院,重庆401331 [2]重庆交通大学信息科学与工程学院,重庆400074

出　　处：《计算机应用与软件》2015年第11期206-209,246,共5页Computer Applications and Software

基　　金：重庆市教委科学技术研究项目(KJ110401)

摘　　要：为了更好地实现全局运动估计快速、准确的处理,根据全局运动中视频图像序列的时间冗余特性,提出一种自适应SIFT(Scale-invariant feature transform)算法。基于最近三次模型匹配的结果,采用Lagrange抛物线插值来预测需要匹配的参考帧和当前帧图像的重叠区域。在重叠区域上提取特征点和进行特征匹配,既能够消除视频图像序列中存在的大量信息冗余,加快每帧图像的处理速度,又可以提高待匹配特征点的有效性,减少误匹配。实验结果表明,改进后的算法自适应能力强、速度快、匹配精度高,基本满足实时定位。In order to make global motion estimation be processed faster and more accurate, we presented an adaptive SIFT （ scale invari- ant feature transform） algorithm according to time redundancy features of the video image sequence in global motion. Based on the model matching results of latest three times, it uses the method of Lagrange parabolic interpolation to forecast the overlapping region of the reference frame needed to match and the current frame. Feature points extraction and feature matching on overlapping region can eliminate a great deal of information redundancy in video image sequence and accelerate the speed of processing each image frame, at the same time, they can also improve the effectiveness of the feature points to be matched and decrease mismatching. Experimental result indicated that the improved algo- rithm can mostly meet the real-time localisation due to its strong self-adaptability, super-speed and high matching precision.

关 键 词：SIFT算法 全局运动估计 自适应 抛物线插值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]