In the measurement of buildings using terrestrial laser scanners, the measurement data contains a large amount of noise that is erroneously measured due to reflections from reflective and transparent glass objects. Measurement data containing such reflection noise adversely affects the analysis that uses it. In this study, we propose an efficient method to remove these erroneous measurements. First, we estimate the glass regions that are the source of the reflection noise based on the reflection intensity of the measured point cloud. Next, we trace the path of the laser beams using the estimated glass regions and the laser scanner position to estimate the area where reflection noise can occur. Finally, we detect reflection noises based on a combination of (1) the symmetry of the noise point cloud and its source entity due to reflection by the window and (2) the geometric similarity of those point clouds, and only reflection noise is removed based on them. Experimental results showed that the reflection noise removal rate reached around 90%, confirming that only the reflection noise can be selectively removed from a measurement point cloud.

Damages of reinforced concrete buildings caused by aging or complicated environmental factors have become a worldwide problem. It is critical to accurately obtain damage information to determine the current state of the aging structure or its levels of decay. However, the inspection of multiple damages o the whole surface of a concrete structure is challenging. In this study, we present a two-stage method for damage detection of surface deflects of reinforced concrete buildings in point clouds captured by a terrestrial laser scanner using a 3D neural network and a novel cluster analysis technique. In the first stage, we divide the whole building into 3D grids, then a classification of multiple damages is performed using PointNet++ along with 3D data with color information mapped on it. In the second stage, we propose a four-step post-processing method based on cluster analysis including removal of isolated damaged clusters, dilation, and filling to develop the detection results. Experimental results show that the voxel-based recall rate reaches 0.928 and the precision rate reaches 0.753. The proposed method offers an acceptable damage detection performance on aging concrete surfaces.

We are working on a research project to evaluate the safety and structure of reinforced concrete buildings damaged by earthquakes using point cloud data acquired by a terrestrial laser scanner. We propose a framework for damage analysis as a classification problem that divides a building in point clouds into small 3D voxel grids and determines which voxels are damaged, instead of detecting the damaged parts from the point cloud of the whole building. Our framework is divided into three steps: First, the damaged building in point clouds is divided into small 3d voxel grids. Second, every voxel is fed into the deep neural network for damage classification. As a deep neural network to classify the voxel grids, we used PointNet++. Finally, the original damage map is refined by a simple cluster analysis. After post-processing, the recall reaches 0.929. That is, 92.9% of damage portions are correctly detected although the damage map still contains some FPs.