基于BRISK特征的动态背景下运动目标检测.docx

2018年第37卷第2期传感器与微系统(TransducerandMicrosystemTechnologies)127

DOI:10.13873/J.1000—9787(2018)02—0127—04

基于BRISK特征的动态背景下运动目标检测*

韩乐乐，王思明，李伟杰

（兰州交通大学自动化与电气工程学院，甘肃兰州730070)

摘要：针对动态背景下运动目标检测过程中对检测算法实时性和鲁棒性的要求，提出了一种基于二进制鲁棒不变尺度特征(BRISK)的运动目标检测算法。通过改进的BRISK算法检测特征点；为了保证匹配精度和速度，采用K最近邻(KNN)算法进行特征点匹配；运用基于随机抽样一致性(RANSAC)的全局运动参数估计法获取最优全局运动参数；采用帧间差分法进行运动目标检测。实验结果表明：改进的BRISK算法减少了49.8%的特征点数目，KNN算法去除了85.9%的特征点对；在各种场景下能够准确地检测出运动目标，与以往算法相比检测效果较好。

关键词：动态背景；运动目标检测；二进制鲁棒不变尺度特征

中图分类号：TP391.4文献标识码：A文章编号：1000—9787(2018)02—0127—04

MovingobjectdetectionbasedonBRISKfeatureindynamicbackground*

HANLe-le,WANGSi-ming,LIWei-jie

(SchoolofAutomationElectricalEngineering,LanzhouJiaoTongUniversity,Lanzhou730070,China)

Abstract:Consideringrequirementsofreal-timeandrobustnessofdetectionalgorithminmovingobjectdetectionprocessunderdynamicbackground,amovingobjectdetectionalgorithmbasedonbinaryrobustinvariantscalablekeypoints(BRISK)isproposed.TheimprovedBRISKalgorithmisusedtodetectfeaturepoints.Toassurematchingprecisionandspeed,theknearestneighbor(KNN)algorithmisadoptedtomatchfeaturepoints.Globalmotionparameterestimationmethodbasedonrandomsampleconsensus(RANSAC)isusedtoobtaintheoptimalglobalmotionparameters.Interframedifferencemethodisappliedtoachievethedetectionofmovingtargets.TheexperimentalresultsindicatethattheimprovedBRISKalgorithmreducesthenumberoffeaturepointsby49.8%,andKNNalgorithmremoves85.9%featurepoints.Meanwhile,thealgorithmcanaccuratelydetectmovingobiectsinvariousscenes,comparedwithpreviousalgorithm,detectingeffectisbetter.

Keywords:dynamicbackground;movingobjectdetection;binaryrobustinvariantscalablekeypoints(BRISK)

0引言

运动目标检测作为机器视觉领域的研究热点，其目的是从图像序