基于EM的红外扩展目标鲁棒自动跟踪方法  

Robust and automatic tracking method of infrared extended object based on EM-like algorithm

作　　者：王睿[1] 王原野[1] 王林[1] 陈旭情[2] 祝世平[1] 

机构地区：[1]北京航空航天大学仪器科学与光电工程学院,北京100083 [2]中国航天科工集团第三研究院,北京100074

出　　处：《红外与激光工程》2008年第4期616-620,共5页Infrared and Laser Engineering

基　　金：国家自然科学基金(60675018);“985”二期“微小型系统与光电探测技术”

摘　　要：针对红外视频图像的特点,提出了一种基于期望最大化算法的红外扩展目标鲁棒自动跟踪方法。首先利用局部Top-Hat形态学滤波进行背景抑制和去噪,并通过平台直方图技术突出跟踪区域的红外目标灰度信息;然后以考虑了像素空间位置信息的高斯加权直方图建立目标的灰度特征模板;最后通过期望最大化迭代计算来估计出各密度分布的最大似然函数的参数集,并由此确定跟踪目标的位置和形状尺寸。实验表明,该方法不仅实现了跟踪窗口随目标尺寸的自适应变化,而且有效克服了红外图像信噪比低的缺点,提高了红外目标实时跟踪的稳健性。A robust and automatic tracking scheme for moving extended object in infrared video images based on expectation maximization （EM） algorithm was proposed. Firstly,local morphological Top-Hat operator was adopted to restrain the image background and eliminate the noises,and the grey level of infrared objects became more salient using the platform histogram technique. Secondly,the gray level feature template of the infrared object was built based on the Gaussian weighted histogram which included gray pixel position. Finally,the localization and shape size of infrared object in every frame were obtained with the EM algorithm,which was an iterative numerical tool for computing the parameter set of maximal likelihood function of density distribution. The experiments demonstrate that the proposed method can track the infrared object with adaptive bandwidths and enhance SNR of the infrared image. It efficiently guarantees the accuracy,stability and continuity in real time tracking infrared object.

关 键 词：红外动目标跟踪 期望最大化估计 自适应跟踪窗口 局部Top-Hat形态学滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]