复杂背景下多目标提取的高灵敏度方法

A high sensitivity method for detection of small targets from the complicated background

对电视经纬仪拍摄到的飞行小目标图像进行了归一化高斯模板相关。传统的模板相关方法不能将真假目标集合充分分离,导致真目标的漏检或引入虚假目标。基于模板相关的高斯函数拟合方法(Gaussian Function Fitting Method, GFFM),对模板相关所得到的目标集(含有真目标和假目标)中的每一个元素进行高斯函数拟合,并引入了一个更为灵敏的检验量—高斯函数拟合误差,可以将真假目标集合明显区分开,减小阈值确定的难度。实验表明:当相关系数阈值rth=0.8时,传统模板相关方法漏检率20%,虚警率40%;而GFFM方法则检出了所有真目标,且无虚假目标。

In order to detect the small flying targets acquired TV theodolite with complicated background, the template correlation is performed with the normalized Gaussian template. The real targets cannot be entirely separated from the real and false targets set with the traditional template correlation method. This leads to omission of real targets or the confusion of the false ones. A Gaussian function fitting method based on template correlation is proposed. With this method, Gaussian function fitting for each element in targets set (including real targets and false targets) obtained by template correlation can be performed and a more sensitive testing quantityGaussian function fitting error has been introduced. This can distinguish the real targets set from the false targets set and this reduces the difficulty for determining threshold. The experiments show that when correlation coefficient threshold rth=0.8, the undetected probability of the traditional template correlation method is 20% and false ratio is 40%. With Gaussian function fitting method, all the real targets is detected without false