基于改进ORB特征匹配的运动小目标检测  

Detecting Small Moving Target Based on the Improved ORB Feature Matching

作　　者：刘威[1] 赵文杰[1] 李成[1] 徐忠林[1] 田铠侨 

机构地区：[1]空军航空大学航空航天情报系,长春130022 [2]北京遥感信息研究所,北京100000

出　　处：《光电工程》2015年第10期13-20,共8页Opto-Electronic Engineering

基　　金：国家自然科学基金资助项目(61301233);全军军事学研究生课题(2013JY512)

摘　　要：为了在航拍视频中准确实时地提取出运动小目标,提出了一种改进后的ORB特征匹配和差分相乘算法融合的检测方法。首先,针对原始ORB特征匹配算法出现大量误匹配对的问题,采用基于K最近邻的特征点描述后,对前后两帧特征点进行双向匹配,再通过顺序抽样一致性算法进一步提纯,利用提纯后的匹配点对解算背景运动模型,精确补偿背景运动量,最后利用连续四帧图像差分相乘的方法并经过形态学处理准确分割出航拍视频中的运动小目标。实验结果表明,经过本文算法提纯后匹配对准确度提升到99.9%,平均耗时0.46 s,处理速度约是SURF特征匹配算法的5倍,SIFT特征匹配算法的25倍,能够满足航拍视频实时处理的需求并具有较强的抗噪能力。In order to extract the small moving target accurately in real time in the aerial video, we propose a fusion detection method for an improved ORB feature matching and differential multiplication algorithm. First of all, as the original ORB appears to be a large number of false matching problems, we describe feature points based on K nearest neighbor. After the description of the feature points in two consecutive frames by two-way matching, we further refine consistency by sequential sampling algorithm. Then, the purified matching points are used to calculate the background motion model, compensating background activity. Finally, the four consecutive frames difference multiplication and morphology processing are used to accurately segment the moving small target in the aerial video. Experimental results show that the match after purification method of accuracy up to 99.9%, the average time of which is 0.46 s, and processing speed is about 5 times of SURF feature matching algorithm, 25 times of the SIFT feature matching algorithm, so it can meet the requirements of aerial video real-time processing and has stronger ability to resist noise.

关 键 词：ORB特征匹配 差分相乘 运动目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]