With the constantly evolving range of applications for technology the quality and amount of data constantly increases as well. In this growing data environment, there is a constant search to provide more value to all data that is available for as little effort as possible. Our research tries to add such additional value by diving into the concept of classifying point cloud by using deep learning, specifically in the indoor environment. This is done by first doing a neural network comparison and then doing a case study. In the neural network comparison, a look is taken into which of the neural networks that are capable of working with point clouds is best suited for our experiments in the indoor scene, based on the training speed, accuracy, ease of use concerning training on external datasets and setting up the network and space efficiency. After the comparison, we chose to continue with the PointCNN network during the case study. The case study is performed on data the NS (Nederlandse Spoorwegen) provided to us and all test results we got from our experiments can be visualized using the web application we developed along with this project. The purpose of the case study is to add extra value to the indoor LiDAR point cloud the NS has captured from Amersfoort Station by using deep learning to automatically classify assets present in their data. The value is in purposes, such as asset management, where the data does not need possibly hundreds of man-hours to be labelled. This saves a lot of time and also money each time a scan is made. In the case study we found through 4 different experiments that unbalanced data makes for bad results, but when a scene is labelled correctly very good results can be found in a local scene.

Many MLS point cloud application scenarios, such as navigation and localization algorithms, require only static environments, but the original MLS data usually inevitably includes many dynamic objects such as moving vehicles, bicycles, and pedestrians. Therefore, these dynamic objects need to be removed before using MLS point clouds. This thesis designs an efficient and memory-friendly Octomap-based dynamic object detection and removal method for MLS data. Firstly, the original MLS data is split into multiple data frames based on the timestamp of each capture point. Each data frame is inserted into a separate Octomap along with its neighboring data frames. The free points in all Octomaps are extracted by setting an occupancy probability threshold. Second, the region of interest (ROI) related to the dynamic object is delineated by the MLS sensor mounting height and the local large vehicle height limit. Only the free points located within the ROI are retained. Then the free-point rate and the multi-return rate are calculated for each free point using a fixed radius spatial search to denoise and detect vegetation points. Finally, the KNN spatial search is used to remove vegetation points and extract dynamic objects from the free points. The proposed method is tested in four case sites in Delft, the Netherlands and its producer’s and user’s weighted average dynamic object detection and extraction accuracies are 88.004% and 82.624%, respectively. The weighted average overall accuracy is 99.833%. Compared with the original Octomap, the proposed method is 35.472% more efficient on average and can be further accelerated by parallel computing, with a maximum memory consumption of only 42.437% of the original Octomap. The implementation results and accuracy assessment demonstrate that the proposed method can be effectively applied to dynamic object detection and extraction tasks in MLS data sets in a compute-friendly and memory-friendly way.