Abstract | ||
---|---|---|
Meaningful 3D maps have a lot to offer to the design of safe and intelligent transportation systems. To do so, street content such as cars, traffic lights and signs need to be segmented in their 3D form, and accurately localized in a map. This paper proposes a first step towards producing 3D rich urban maps in which cars are segmented from a point cloud, and rendered in their true form on a metric map. Our system is based on the integration of stereo SLAM for point cloud extraction, 3D car detection, shape completion, meshing, and optimization of camera pose based on the detected cars. We test our system on the KITTI dataset and produce very realistic maps. As a second contribution, we assess the effect of including the detected cars as objects in a semantic Simultaneous Localization and Mapping (SLAM) pipeline, and demonstrate the potential for improved localization. |
Year | DOI | Venue |
---|---|---|
2020 | 10.1109/CRV50864.2020.00014 | 2020 17th Conference on Computer and Robot Vision (CRV) |
Keywords | DocType | ISBN |
3D Object Detection,3D mapping,object-SLAM | Conference | 978-1-7281-9892-7 |
Citations | PageRank | References |
0 | 0.34 | 7 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Maya Antoun | 1 | 0 | 0.34 |
Daniel C. Asmar | 2 | 82 | 20.11 |
Rema Daher | 3 | 0 | 0.34 |