We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A 3D Semantic Segmentation Method for Large-Scale Point Cloud on Deep Learning.
- Authors
Sihan Liu; Wenyu Zhang; Yujun Zhang; Zhijian Wang; Dongxiang Gao
- Abstract
Point cloud data’s ability to preserve precise geometric details makes point cloud semantic segmentation crucial for 3D visual perception. Within a perception system, the realtime performance of the model holds pivotal significance, particularly concerning its ability to conduct semantic segmentation on large-scale point cloud data. RanSeNet, a lightweight neural architecture based on attention mechanisms, directly operates on each individual point in the point cloud data, eliminating the need for preliminary processing steps. The experiment shows that RanSeNet achieves fast processing, high segmentation efficiency, and handles millions of points simultaneously. Compared to existing results, the proposed method achieves 88.6% Overall Accuracy (OA) and 64.27% Mean Intersection over Union (MIoU) in area 5 of the S3DIS dataset, which is a challenging large-scale semantic scene segmentation task.
- Subjects
POINT cloud; DEEP learning; VISUAL perception
- Publication
Engineering Letters, 2023, Vol 31, Issue 4, p1667
- ISSN
1816-093X
- Publication type
Article