Automatic stem mapping using single-scan terrestrial laser scanning data and Mean Shift clustering. Issue 1 (October 2021)

• Record Type: Journal Article
• Authors: Liu, Xiangjiang; Chen, Maolin; Tan, Chunsen; Zhang, Xinyi; Yang, Wenguang
• Is Part Of: IOP conference series. Volume 865:Issue 1(2021)
• Publication Date: 2021-10
• Subjects: Earth sciences -- Periodicals; Environmental sciences -- Congresses; Environmental sciences -- Periodicals; 550.5
• DOI: 10.1088/1755-1315/865/1/012015
• Languages: English
• ISSNs: 1755-1307

Abstract

Abstract: As an effective tool for evaluation of forest structure parameters, terrestrial laser scanning (TLS) can capture forest data efficiently and automatically. In this paper, a stem mapping method of TLS data is proposed, based on Mean Shift clustering. The ground points first are extracted by cloth simulation filter (CSF) and used to generate ground model. A subsequent classification based on Random Forest (RF) identifies candidate stem points using 21-dimensional geometric features. After obtaining the candidate stem points, we utilize Mean Shift clustering to extract individual stems, instead of the usual Euclidean clustering method, and adopt adaptive density filtering to remove non-stem cluster. Finally, RANSAC cylinder fitting is used from the sliced point cloud, the intersection of the cylinder axis and the ground is taken as the stem location. The results show that Mean Shift clustering is more effective than Euclidean clustering in stem detection. Additionally, the precision and recall of stem mapping based on the proposed method are 88.83% and 93.94% respectively.