摘要： 为提高无人机道路检测的实时性和鲁棒性,提出一种基于改进graphcut算法的道路检测方法。利用Orchard-Boumand聚类算法聚类道路和非道路像素点,通过高斯混合模型对这2类像素点建模,构造Gibbs能量惩罚函数中的区域项函数。针对航拍图像各个区域具有不同对比度 的特点,设计Gibbs能量惩罚函数中的光滑项函数,将单一的图像全局对比度矩阵替换为局部对比度矩阵。通过Gibbs能量惩罚函数构造有权重的图,运用max-flow算法进行分割,检测出道路区域。实验结果表明,该方法在不同类型道路下都能保持较好检测性能,与现有的道路 检测方法相比,实时性好,错误率低。

关键词: 小型无人机航拍图像, 道路检测, graphcut算法, Orchard-Boumand聚类算法, 高斯混合模型, 局部对比度矩阵

Abstract: In order to improve the real-time and robustness of the Unmanned Aerial Vehicle (UAV) road detection,this paper proposes a new method based on the improved graphcut algorithm.Orchard-Boumand clustering algorithm is used to cluster the road and nonroad pixels,these pixels are modeled by Gaussian Mixture Model(GMM),so that the Gibbs energy penalty function for region term is constructed.For an aerial image,contrast in different regions changes a lot,the paper designs a new smoothness term of the Gibbs energy penalty function,the image global contrast is replaced by a local contrast matrix.A weighted graph is constructed with the Gibbs energy penalty function,and max-flow algorithm is performed to segment the road area.Experimental results show that this algorithm can keep high detection performance under different types of roads and realize road detection more accurately and improve real-time compared with the existing road detection method.

Key words: small Unmanned Aerial Vehicle(UAV) image, road detection, graphcut algorithm, Orchard-Boumand clustering algorithm, Gaussian Mixture Model(GMM), local contrast matrix

中图分类号: 

• TP391