基于ViBE与面向对象分类的视频卫星运动车辆检测

为了提高视频卫星对运动车辆的检测质量，在经典视觉背景提取器(ViBE)算法的基础上，结合遥感的面向对象分类技术，从提升正确检测运动目标数量和抑制虚假运动目标检测数量两个方面着手，提出了一种新的运动车辆检测方法（VOMVD）。首先通过优化ViBE模型参数，尽可能多地获取真实运动目标，但这在一定程度引入了许多的虚假目标。研究继而依据影像上地面小尺度运动目标和道路的依存关系，采用面向对象的分类方法，基于光谱、纹理、空间属性，构建了均值、标准差、卷积核内平均灰度值、卷积核内平均信息熵、面积、长度、紧密度、延伸度等8个特征，用于提取道路信息，以此掩膜ViBE提取的虚假运动目标和伪运动目标。结果表明，基于本研究提出的视频卫星运动目标检测方法较之三帧差分法、ViBE检测方法等，其精度有明显提升。在本研究中，三帧差分法、ViBE和VOMVD对运动目标的检测精度P分别为70.91%，61.49%和85.71%，召回率R分别为84.78%，98.91%和97.83%，F值分别为77.23%，75.83%和91.37%，有效提升了方法对运动目标的检测效果。

Moving vehicle detection of video satellite based on ViBE and object-oriented classification

To improve the detection quality of moving vehicles from video satellites,study was carried out from two dimensions: increasing the number of correct detection of moving objects and suppressing the number of false detection of moving objects.A new moving vehicle detection method was proposed based on the visual background extractor (ViBE) algorithm and the object-oriented classification technology in remote sensing field.By optimizing the parameters of the ViBE,real moving objects can be obtained as much as possible,but this introduced a lot of false objects to some extent.Therefore,according to the interdependent relationship between moving vehicles and road,object oriented classification method was adopted to extract road and further filter the false objects produced by ViBE.A total of eight features including the mean value and standard deviation of spectral attribute,average gray value and entropy in the convolution kernel of texture attribute,area,length,tightness and extension of spatial geometric attributes were used.The results show that, compared with the three frames difference method and ViBE detection method,the moving vehicles detection accuracy of the proposed method has been significantly improved.The detection precision P of three frames difference method,ViBE and VOMVD for moving objects are 70.91%,61.49% and 85.71% respectively,the recall R is 84.78%,98.91% and 97.83% respectively,and the F score is 77.23%,75.83% and 91.37% respectively.The effect of the proposed method on moving vehicles detection has been improved effectively.