Scan-to-BIM is a process of creating Building Information Model of existing buildings and sights. The semantic information must be extracted from the scan, which is gen-erally represented as a 3D point cloud. The presented work evaluates the data preparation for point cloud segmentation by deep learning. In the preparation process, a point cloud is segmented and classified in order to create different data sets that serve as input for a neural network. The neural network KPConv is then used to evaluate the performance on these data sets. The goal of this thesis is to evaluate the effects of different segmentation and classi-fication approaches on the performance of the neural network. Varying the data set and the parameters of the neural network, multiple experiments are conducted to compare the results and thus draw conclusions on how to best prepare a point cloud for the training of a neural network for point cloud segmentation.     «

Scan-to-BIM is a process of creating Building Information Model of existing buildings and sights. The semantic information must be extracted from the scan, which is gen-erally represented as a 3D point cloud. The presented work evaluates the data preparation for point cloud segmentation by deep learning. In the preparation process, a point cloud is segmented and classified in order to create different data sets that serve as input for a neural network. The neural network KPConv is then used to e...     »