Accurate extraction of brick shapes in masonry walls from dense terrestrial laser scanning point cloud

Abstract

Masonry structures commonly composed of a large amount of bricks or stones. Making an inventory of bricks in masonry walls is of great importance in the fields of building documentation, change detection and pattern recognition. This paper explores the problem of brick shape extraction for point cloud data with outliers as gathered by a laser scanner. For this purpose, we establish the basic brick shape model. Next, dimension reduction from 3D to 2D is done using principal component analysis (PCA). Traditionally, the bounding rectangle, as well as four vertices of the brick, are estimated by computing the maximum and minimum values of the coordinates of the brick points in 2D space. However, the existence of outliers in brick point cloud is common that may occur as random or systematic errors. If so, the brick shape extraction results are unreliable. To address this problem, we present a robust rectangle fitting algorithm for dense point cloud that uses PCA. Theoretically, the rectangle fitting model is established based on the geometry of the brick itself. We introduce the slack variable and critical rectangle to improve the anti-noise capability. The accuracy and consistency of the proposed robust rectangle fitting is evaluated by analysing single brick. Afterward, the method implemented on two patches of a masonry wall and the performance is demonstrated.