Robust dynamic point removal in point clouds through density filtering algorithms
| dc.contributor | Aalto-yliopisto | fi |
| dc.contributor | Aalto University | en |
| dc.contributor.advisor | Ojala, Risto | |
| dc.contributor.advisor | Li, Jie | |
| dc.contributor.author | Habibiroudkenar, Pejman | |
| dc.contributor.school | Insinööritieteiden korkeakoulu | fi |
| dc.contributor.supervisor | Tammi, Kari | |
| dc.date.accessioned | 2023-08-27T17:15:51Z | |
| dc.date.available | 2023-08-27T17:15:51Z | |
| dc.date.issued | 2023-08-21 | |
| dc.description.abstract | In recent years, indoor autonomous vehicle robotics have gained significant popularity due to their efficiency and ability to enhance overall quality in factory settings. These robots rely heavily on accurate maps of their surroundings for localization. However, in highly dynamic factory environments, the mapping process often registers dynamic points, which can negatively impact localization accuracy in future uses. This study aims to address this challenge by aligning and merging maps of the same location from different time periods, resulting in denser stationary points and sparser dynamic points. To achieve this, the study proposes three filtering algorithms: the max distance sphere method, the average distance sphere method, and the convex hull method. These algorithms involve different approaches to estimate the volume occupied by a point and its k (user defined) neighbouring points. By repeating this process for all points in the point cloud, a density factor is derived, which is then used as a threshold to remove the dynamic points. To evaluate the performance of the filtering algorithms, the study compares the odometry obtained from a merged noisy map with its corresponding filtered map and a clean map constructed during a period of low activity. The study also assesses factors such as computational power requirements, reliability, and ease of implementation for each method. The findings of the study demonstrate that the proposed filtering algorithms perform nearly as well as a clean map in the odometry test, while the noisy map fails to accurately localize itself, especially after high-angle turns. Among the three proposed methods, the average distance sphere method proves to be the seemingly the most effective, primarily due to its ability to minimize the false removal of dynamic points. | en |
| dc.format.extent | 51 | |
| dc.identifier.uri | https://aaltodoc.aalto.fi/handle/123456789/122868 | |
| dc.identifier.urn | URN:NBN:fi:aalto-202308275209 | |
| dc.language.iso | en | en |
| dc.programme | Master's Programme in Mechanical Engineering (MEC) | fi |
| dc.programme.major | Mechatronics | fi |
| dc.programme.mcode | fi | |
| dc.subject.keyword | SLAM | en |
| dc.subject.keyword | 3D LiDAR | en |
| dc.subject.keyword | point cloud | en |
| dc.subject.keyword | dynamic points removal | en |
| dc.title | Robust dynamic point removal in point clouds through density filtering algorithms | en |
| dc.type | G2 Pro gradu, diplomityö | fi |
| dc.type.ontasot | Master's thesis | en |
| dc.type.ontasot | Diplomityö | fi |
| local.aalto.electroniconly | yes | |
| local.aalto.openaccess | no |
