A Local Enhancement Method for Large-Scale Building Facade Depth Images using Densely Matched Point Clouds

Abstract

In recent years, laser scanning systems have been widely used to acquire multi-level three-dimensional spatial objects in real time. The laser scanning system is used to acquire the three-dimensional point cloud data of urban scenes. Due to the large-scale characteristics of urban scenes, and the problems of scanning occlusion, scanning path, and limited scanning laser range, the laser scanning system cannot scan every object in the scene comprehensively, multidirectionally and finely, so the corresponding three-dimensional point cloud data collected by many objects are incomplete, and the data images are relatively sparse and unevenly distributed. The existing point cloud denoising and enhancement algorithms, such as AMLS, RMLS, LOP, and WLOP, all use local information to enhance the missing or sparse parts of the point cloud. This point cloud enhancement method is only limited to a small range and cannot do anything for the larger missing area of the point cloud. Even if it is done reluctantly, the effect is not satisfactory. There are a lot of repetitive and similar features in urban buildings, such as the repetitive areas of floors and balconies in buildings. These repetitive areas are distributed in different positions of point clouds, so the repetitive information has non local characteristics. Based on the nonlocal characteristics of building point cloud data and the repetitive structure of buildings, this article proposes a nonlocal point cloud data enhancement algorithm, which organizes the point cloud data in the repeated area into a set of basic geometric elements (planes). The structures are registered in a unified coordinate system, and the point cloud is enhanced and denoised through two denoising processes, “out-of-plane” and “in-plane.”