[目的]利用无人机进行大尺度非结构地形环境测绘时,由于无人机倾斜摄影得到的图像在测绘建模时会存在仿射变形较大、透视畸变严重、局部特征变化各异等问题,进而导致建模数据匹配困难.为解决这一问题,本文提出了一种基于特征融合的倾斜摄影测绘建模优化方法.[方法]首先融合图像的颜色信息及近似最近邻快速库(fast library for approximate nearest neighbors,FLANN)优化的加速稳健特征(speed up robust feature,SURF),然后结合优化的 SURF与最稳定颜色区域特征(maximally stable color regions,MSCR)实现仿射变形图像间快速准确的特征提取及匹配.[结果]本文的特征匹配算法在1.25 s内得到757个最佳匹配点;相对于基于尺度不变特征(scale-invariant feature transform,SIFT)、SIFT+MSCR和SURF的特征匹配算法,最佳匹配点的数量分别提高141％、29％和34％,匹配时间与SURF接近,远低于SIFT和SIFT+MSCR.测绘建模得到的点云与参考点在三个方向上的距离均方根误差在7 cm以内,平均误差在11 cm以内.[结论]本文提出的算法能够实现非结构地形环境的全局三维模型快速准确构建,同时数据匹配过程中具有更良好的匹配效果和匹配效率.
Optimization method for oblique photogrammetric modeling based on feature-fusion in unstructured terrain environment
[Objective]For years,data-matching difficulties caused by large affine deformation,severe perspective distortion,and varying local feature changes among unmanned aerial vehicle oblique photography images amid large-scale unstructured terrain environments have existed.Herein,to overcome these difficulties,we propose a feature-fusion optimization method for oblique photography modeling based on unmanned aerial vehicle oblique photography technology.[Methods]First,in the proposed algorithm,we integrate the color information of the image and the fast library for approximate nearest neighbors(FLANN)algorithm to optimize and accelerate the robust feature(SURF).Then,combining optimized SURF features with maximally stable color regions(MSCR)features,we have achieved fast and accurate feature extraction and matching among affine deformed images can be achieved.Finally,the global 3D model of unstructured terrain environments is quickly and accurately constructed.[Results]The comprehensive comparative analysis shows that the feature-matching algorithm integrates advantages of SURF point features and MSCR region feature algorithms.Notably,we obtain 757 best matching points within 1.25 s and achieve more accurate matching results.Compared to scale-invariant feature transform(SIFT),SIFT+MSCR,and SURF,the number of optimal matching points obtained by our algorithm is increased by 141％,29％,and 34％,respectively.The matching time of our algorithm is close to SURF and much lower than SIFT and SIFT+MSCR.Meanwhile,traditional algorithms endure data loss due to matching failures,resulting in a large number of point cloud voids in the unstructured terrain environment point cloud constructed by those algorithms.The point cloud collected from the global terrain point cloud of the unstructured environment constructed by the optimization algorithm herein can comprehensively reconstruct the unstructured terrain environment.The average mean error of the constructed unstructured terrain environment point cloud map in three directions of the coordinate axis is reduced to 9.82 cm with the average root-mean-square error of 6.36 cm.[Conclusions]The global unstructured environment 3D terrain model can provide reliable prior information for unmanned vehicles driven in unstructured terrain environments.This paper focuses on the practical challenges faced in the construction process of unstructured terrain environments.By using drone oblique photography technology and combining optimized SURF and MSCR algorithms,fast and accurate feature detection and image matching have been achieved.Finally,the construction of a three-dimensional global unstructured terrain environment point cloud is achieved.Experimental results show that the proposed algorithm can maintain matching robustness,timeliness,and accuracy even under conditions of large affine deformation,severe perspective distortion,and diverse local feature changes in unstructured terrain environments.The proposed algorithm functions satisfactorily for oblique photography modeling in unstructured terrain environments,and ultimately achieves the construction of centimeter level 3D global unstructured terrain environment point cloud maps.
万方数据
维普
