三维重建中线段匹配方法的研究

Line Matching in Three-Dimensional Reconstruction

针对传统基于特征点匹配的重建模型容易存在空洞现象的问题,提出一种基于直线提取算法的线段匹配方法。首先使用无人机设备获取校园建筑图片集,保证拍摄过程中相邻图片存在重叠部分,通过筛选分数最高的相似性评分得到相邻图片集合并只在相邻图片间进行匹配。接着利用直线提取算法提取图片中直线段,然后根据极线约束关系进行二维线段匹配最终生成三维线段模型。二维线段匹配阈值τ、相邻图片个数M、在相邻图像中与特定二维线段相匹配的线段个数K是其中3个重要参数,对4组不同的图片集进行参数评估,实验结果表明当二维线段阈值τ低于0.5,相邻图片个数M为10,在相邻图像中与特定二维线段相匹配的线段个数K为10时三维重建速率与模型精度可接受度最高。

A line matching method based on Line Segment Detector（LSD）algorithm was proposed to resolve holes possibly in traditional three-dimensiona（l3D）reconstruction by characteristic point matching.Firstly,we used Unmanned Aerial Vehicle to take pictures of campus buildings and ensured that there were overlaps in neighboring pictures.Secondly,we selected the highest similarity scores to obtain the neighboring picture sets and matched only between neighboring pictures.Then,LSD algorithm was used to extract the line segments in the pictures.Finally,the 3D segment model was reconstructed by using two-dimensional segment matching based on epipolar constraint.Three important parameters,fixed thresholdτ,number of the nearest neighbors M and matched specific segment in neighbors K were used in the model.A parameter evaluation was made in four different picture sets.The experiment results showed that the reconstructing rate and model accuracy were highest whenτwas less than 0.5,M was 10 and K was 10.