研究目的
To propose an automatic parameter estimation method for the DBSCAN algorithm to improve LiDAR point cloud data segmentation by automatically determining the critical parameter ε based on the intrinsic properties of the data.
研究成果
The proposed automatic parameter estimation method for DBSCAN effectively segments LiDAR point cloud data with higher accuracy (75%, 74%, 71% for different datasets) compared to manual parameter selection. It reduces the need for manual intervention and improves automation in data processing, demonstrating robustness across airborne and mobile LiDAR data with and without color information.
研究不足
The method requires manual setting of minPts and maxPts parameters; it may not handle very complex geometries or high noise levels effectively; the accuracy depends on the quality of the point cloud data and the reference data used for evaluation.
1：Experimental Design and Method Selection:
The study uses an improved DBSCAN method with a novel automatic parameter ε estimation based on the average of k nearest neighbors' maximum distance. It involves data normalization, spatial index building (Kd-tree), polynomial fitting, and clustering.
2：Sample Selection and Data Sources:
Airborne LiDAR data from Baltimore, Maryland, USA, downloaded from NOAA Coastal Services Centre, and mobile LiDAR data from a street area acquired by Optech Lynx V100 system.
3：List of Experimental Equipment and Materials:
Leica Airborne Laser Scanner Model ALS 50, Sanborn Aero Commander 500B aircraft, Optech Lynx V100 mobile survey system, ESRI ArcScene
4：3 for reference data collection. Experimental Procedures and Operational Workflow:
Data pre-processing (normalization, noise removal, ground point removal), parameter estimation (calculating KNN mean max distance, polynomial fitting, adding corrections, deriving first derivative to find optimal ε), clustering with DBSCAN, and accuracy evaluation using reference data.
5：Data Analysis Methods:
Accuracy evaluation based on categories (correct detection, over-segmentation, under-segmentation, missed, noise) and statistical analysis of segmentation results.
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
