2.2.1. Classify Changes to Permanent and Non-Permanent

The next step was to separate the changes that were part of the permanent structure from dynamic objects. This involved classifying the point clouds in each epoch to a permanent structure (e.g., walls, floors, ceilings) and a non-permanent structure (e.g., furniture, clutter and outliers).

In Figure 7c, the blue color represents the areas captured by Zeb-Revo and the red areas show the differences in the coverage where PC1 is not covered by PC2. In Figure 7d, the point clouds of epoch1 after the comparison with the epoch2 are shown and the blue points show the points in which their distance differences are less than the threshold and are not changed. The green points show the changes, because of coverage or furniture, or a permanent change, and the ceiling is removed for better visualization. We applied a method from [23] to classify the permanent structures in each epoch (see Figure 8). Four main classes were important for our change detection process. Walls, floors, and ceilings were three classes that belonged to the permanent structures. The non-permanent structures were, for example, furniture, outliers, and unknown points, which were classified as the clutter. The classification started with surface growing segmentation and generating an adjacency graph from the connected segments. By analyzing the adjacency graph, it was possible to separate permanent structures, such as walls, because of their connection to the floor and ceiling. The normal angle of the planes was important in this decision because walls in most indoor environments have an angle of more than 45 degrees with the positive direction of the *z*-axis. Figure 8c shows that the permanent structure (walls and floor) was separated from the clutter.

After the classification of points in each epoch, by comparing the changes with the semantic labels (walls, floors, and ceilings), it is possible to distinguish relevant changes for 3D Cadastre. Each point in the set of changes is a possible change for 3D Cadastre if is labeled as a wall, floor, or ceiling, otherwise it is a change only because of furniture or dynamic objects or outliers. Table 1 shows how we identified changes with labels per point, respecting the permanent structure. According to the table, points with label 1 are important for change detection in 3D Cadastre because they represent a permanent change in the building. Figure 5 represents the changes with different colors according to their label.

**Table 1.** The table shows how the point clouds are labeled regarding the changes and their role in the building structure. The points with label 1 are interesting for change detection of the 3D Cadastre.


**Figure 7.** (**a**) Point clouds from a backpack system from the first epoch. (**b**) Point clouds from a Zeb-Revo system from the second epoch. (**c**) Co-registered point clouds. (**d**) Point clouds of epoch1 after the comparison with epoch2.

**Figure 8.** (**a**) PC1 acquired by a backpack and (**b**) PC2 is acquired by Zeb-Revo, walls (orange), floor (yellow), clutter (blue). (**c**) The changes are detected in PC1 and classified to permanent structure changes (yellow) and non-permanent changes (red). The red rectangle shows the wall that is showing a permanent change.
