Automatic Detection and 3D Modeling of City Furniture Objects using LiDAR and Imagery Mobile Mapping Data
City furniture objects hold valuable information about urban traffic and city dynamics, making their integration into 3D city models essential for enhancing these models. This study implements two methodologies for detecting, classifying, and positioning City Furniture objects, as well as one approa...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2024-12-01
|
| Series: | The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-archives.copernicus.org/articles/XLVIII-2-W8-2024/125/2024/isprs-archives-XLVIII-2-W8-2024-125-2024.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | City furniture objects hold valuable information about urban traffic and city dynamics, making their integration into 3D city models essential for enhancing these models. This study implements two methodologies for detecting, classifying, and positioning City Furniture objects, as well as one approach for their automatic 3D modeling. The first approach uses Mobile Mapping System (MMS) imagery with YOLO for object detection and classification, coupled with the Line of Bearing (LoB) method for extracting XYZ coordinates and height. A spatial operation was conducted to determine object orientation. The second approach employs camera- LiDAR fusion, integrating KPConv for semantic segmentation and connected components for instance segmentation. Classification is performed using two complementary approaches: using Fast Global Registration (FGR) on point clouds, for lamppost types, and image-based, projecting point cloud instances to classify traffic lights and signs. This fusion approach leverages image-based classification models and point cloud accuracy, achieving an RMSE of 0.32 against ground truth data. The point cloud approach shows promise but requires refinement to improve noise sensitivity in FGR. This study presents a comprehensive workflow from detection to Level of Detail 4 (LOD4) modeling, combining KPConv and multi-source data to enhance feature detection and classification for city furniture. |
|---|---|
| ISSN: | 1682-1750 2194-9034 |