改进SIFT匹配的动态背景下运动目标检测算法  

Moving Target Detection Algorithm under Dynamic Background with Improved SIFT Matching

作　　者：侯宏录[1] 李媛 李光耀 HOU Honglu;LI Yuan;LI Guangyao(School of Optoelectronic Engineering,Xi'an Technological University,Xi'an 710021,China)

机构地区：[1]西安工业大学光电工程学院

出　　处：《自动化仪表》2019年第8期60-64,共5页Process Automation Instrumentation

基　　金：陕西省工业科技攻关基金资助项目(2016GY-051);陕西省教育厅重点实验室科研计划项目(15JS035)

摘　　要：无人机/车载运动目标检测与识别系统由于存在摄像机运动,使得场景中的背景与前景目标同时发生运动,增加了运动目标检测的难度。目前,动态背景下运动目标检测的主要方法是利用特征点的运动估计来补偿背景的运动。针对尺度不变特征变换(SIFT)特征匹配算法存在耗时长、误匹配率大,使得背景运动估计不准确,进而影响运动目标检测效果。提出了一种改进的SIFT特征匹配算法——先剔除容易丢失的边缘特征点,再利用最近邻点与次近邻点的欧式距离比确定匹配点,并用RANSAC删除运动目标上和误匹配点对,进一步实现背景运动的准确估计,从而检测得到运动目标的准确位置。试验结果表明:改进的特征点匹配算法在匹配准确度和运算速度上均优于传统SIFT特征匹配算法的性能。最终,通过背景补偿,有效消除了相机运动引入的动态误差,提高了运动目标检测的实时性和精度。Due to the existence of camera motion,for UAV/vehicle moving target detection and recognition system,the background in the scene and the foreground target are moving simultaneously,which increases the difficulty of detecting the moving target.At present,the main method of moving target detection in dynamic background is to use the motion estimation of feature points to compensate the background motion.For the scale-invariant feature transform(SIFT)feature matching algorithm,it takes a long time and the mismatch rate is large,which makes the background motion estimation inaccurate,and thus affects the moving target detection effect,thus an improved SIFT feature matching algorithm is proposed.Firstly,the edge feature points that are easily to be lost are excluded,then and the Euclidean distance ratio between the nearest neighbor and the next nearest neighbor is used to determine the matching point,and the RANSAC is used to delete the moving target and the mismatched pair.An accurate estimation of the background motion is further achieved to detect the exact position of the moving target.The experimental results show that the improved feature point matching algorithm is superior to the traditional SIFT feature matching algorithm in matching accuracy and operation speed.Finally,the background error effectively eliminates the dynamic error introduced by camera motion,and improves the real-time performance and accuracy of moving target detection.

关 键 词：尺度不变将化变换 特征点匹配算法 背景补偿 运动目标检测 动态背景 摄像机运动 背景运动估计 背景全局运动模型 

分 类 号：TH745[机械工程—光学工程]