Haar-like特征在稀疏建筑物检测中的应用

Application of Haar-like Features in Sparse Building Detection

以高原无人区稀疏建筑物为研究对象,将Hough变换、Haar-like特征与AdaBoost算法相结合构造强分类器,利用高分辨率遥感影像快速精确地从无人区检测出固定稀疏建筑物。首先对影像进行边缘检测、Hough变换的直线提取与几何旋转校正,将实际可能是任何角度的建筑物旋转成水平或垂直状态,再将旋转后图像提取Haar-like特征后利用AdaBoost算法进行分类。实验证明,该算法原理简单,能有效解决仅用Haar-like特征精度不适应建筑物角度多变的问题,说明了Hough变换直线特征提取与Haar-like矩形特征提取多角度稀疏建筑物的可行性,为快速精确检测无人区的稀疏建筑物提供了新思路。

This paper takes the Akesu Xinjiang region as the study object and combines Hough transform,haar-like features and AdaBoost algorithm to form a strong detector.It detected fixed and spare buildings rapidly in the no-man’s land by using high-resolution remote sensing images.This algorithm first adopts an edge tracking and extraction.It then use line extraction by Hough transform and geometric rotation correction.This allows the building,which may actually be at any angle,to be rotated to a horizontal or vertical position.Finally,the rotated image is extracted with Haar feature and classified by AdaBoost algorithm.The experiment results prove that the algorithm is simple in principle and can effectively solve the problem that the accuracy of Haar-like features is not adaptable to buildings,and the feasibility of extracting multi-angle sparse buildings by using Hough transform and Haar-like rectangular features is illustrated.It provides new ideas for effective and accurate the spare buildings extraction in the no-man’s land.