一种基于SVM的无人机影像中单个建筑物的角点检测方法

针对目前无人机影像中单个建筑物角点的检测现状,提出了一种基于支持向量机(SVM)的无人机影像中建筑物的角点检测方法。首先对4个波段的无人机影像进行多尺度分割,计算影像的NDVI,通过植被与非植被区域的波谱差异剔除植被的影响。其次,用面向对象分类法将"建筑物块"从影像中提取出来,对"建筑物块"区域用Harris算子进行边缘检测,形成建筑物边缘点集数据。随后通过设计高斯径向基将边缘样本点映射到高维特征空间,构建特征向量,采用边缘点集训练SVM分类模型,最终通过SVM分类模型从粗提取的边缘点集中检测出正确的建筑物角点,实现了单个建筑物的角点提取。

UAV Image Detecting of Single Building' s Angular Points Method Based on SVM

In view of the present situation of single building' s angular points detection in UAV images, this paper proposes a method that based on support vector machine (SVM) to detect the corner of the building.Firstly,the UAV image with four bands to complete multi-scale segmentation,calculates the NDVI of this image, to eliminate the effect of vegetation by the spectral differences between vegetation and non-vegetation areas;Secondly, using object-oriented classification to extract "building block" from the image,and the edge detection for the "building block" completed by Harris, then it comes into being edge point set of building and extracts some points as samples randomly. The edge sample points are mapped to high-dimensional feature spaces by Gauss RBF and construct the feature vector,the SVM classification model is trained by edge point set. Finally, the correct building corner is detected by the SVM classification model from the rough edge points, and the corner of the single building is extracted.