一种基于深度图改进的分块密集匹配方法

An improved block dense matching method based on depth map

倾斜摄影建模往往需要几百张甚至上万张图像,通常需要分块密集匹配重建。然而,常用的分块密集匹配方法只能对每个块单独建模,块间的重复图像会多次参与匹配计算,导致计算冗余、耗时长。因此,本文提出了一种改进的分块密集匹配方法,避免块间重复图像的冗余计算。首先,利用图像关联算法为场景中的每张图像计算邻域图像集合,同时初始化图像的重建状态对象;然后,对整个场景进行空间立体自动聚类完成分块,并抽取子块重建时所需的匹配对图像;最后,根据图像的重建状态对象估计、过滤深度图及深度图融合生成密集点云。试验结果表明,该方法在倾斜影像三维重建时,效率显著提升。

Oblique photography modeling often requires hundreds or even tens of thousands of images and usually requires block dense matching reconstruction. However, the commonly used block dense matching method can only model each block separately and the repeated images between blocks will participate in the matching calculation multiple times, resulting in redundancy and time-consuming calculations. Therefore, an improved block dense matching method is proposed. First, the image association algorithm is used to calculate the neighborhood image set for each image in the scene and the reconstruction state object of the image is initialized at the same time. Then, the entire scene is automatically clustered in space to complete the block and the sub-blocks needed for reconstruction are extracted. Finally, according to the image reconstruction state, object estimation, filtering depth map and depth map fusion generate dense point cloud. Experimental results show that this method can significantly improve the efficiency of oblique image 3D reconstruction.