研究目的
To develop an automatic method for detecting pole-like objects such as utility poles, street lights, traffic signs, and tree trunks from mobile LiDAR point clouds, addressing challenges like imbalanced object distribution, sensitivity to radius variations, and complex attached structures.
研究成果
The proposed skeleton-based hierarchical method is effective and robust for detecting pole-like objects, achieving high quality and outperforming existing methods, particularly for large-radius objects. Future work includes improving computational efficiency and fine classification of extracted objects.
研究不足
Computational efficiency needs improvement due to iterative smoothing; some pole-like objects may not be detected due to complex decorations, close proximity, missing data from occlusions, or errors in segmentation for objects with multiple poles.
1：Experimental Design and Method Selection:
A hierarchical method involving coarse extraction of building facades, slice-based Euclidean clustering, skeleton-based PCA shape recognition, and Voronoi-constrained vertical region growing.
2：Sample Selection and Data Sources:
Public Paris–Lille-3-D data set, consisting of urban point clouds from Lille and Paris with high density.
3：List of Experimental Equipment and Materials:
Mobile LiDAR system for data acquisition; computational tools for processing point clouds.
4：Experimental Procedures and Operational Workflow:
Steps include facade extraction, slicing and clustering, Laplacian smoothing for skeleton extraction, PCA-based recognition, and region growing for individual object production.
5：Data Analysis Methods:
Evaluation using recall, precision, and quality metrics; comparisons with voxel-based and slice-based methods.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
