一种改进的视频序列SIFT特征检测方法  

An Improved Method of Detecting SIFT Features in Video Frame

作　　者：陈滨[1] 赵建军[1] 马同刚 王远强 

机构地区：[1]海军航空工程学院兵器科学与技术系 [2]91899部队

出　　处：《计算机与数字工程》2017年第3期527-530,557,共5页Computer & Digital Engineering

摘　　要：SIFT算法在不变特征提取方面有着较大的优势,但因为其时间复杂度高,在实时应用上仍存在较大挑战。为了提高SIFT算法的实时性,论文对SIFT特征检测算法进行改进,根据连续图像间的全局运动矢量,动态更新邻域半径,并依据该半径获取后续视频图像序列需要进行极值检测的区域图像,从而减少后续图像极值检测时间,提高SIFT特征检测算法的效率。实验证明,改进算法在确保特征点数目的前提下,有效减少了尺度空间下的极值检测时间,提高了SIFT算法效率。Although SIFT algorithm has great advantage of extracting the invariant features,it is a challenge to use the algorithm real-time because of the time complexity.The paper improves the SIFT algorithm by limiting the area of extremes detection.The radius of the area is updating with the global motion vectors between the continuous frames of a video.With the improved algorithm,the images of the areas are obtained in which the extremes to be detected.The algorithm keeps the amount of SIFT features and is proved to be more efficient by decreasing the time cost by extremes detection.

关 键 词：SIFT算法 全局运动矢量 视频图像 尺度空间 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]