Segmentation of unbalanced and in-homogeneous point clouds and its application to 3D scanned trees
- PDF / 4,164,071 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 119 Downloads / 195 Views
ORIGINAL ARTICLE
Segmentation of unbalanced and in-homogeneous point clouds and its application to 3D scanned trees Jules Morel1
· Alexandra Bac2
· Takashi Kanai1
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Segmentation of 3D point clouds is still an open issue in the case of unbalanced and in-homogeneous data-sets. In the application context of the modeling of botanical trees, a fundamental challenge consists in separating the leaves from the wood. Based on deep learning and a class decision process, we propose an innovative method designed to separate leaf points from wood points in terrestrial LiDAR point clouds of trees. Although simple, our approach learns trees characteristic point patterns efficiently and robustly. To train our 3D deep learning model, we constructed a 3D labeled point cloud data-set of different tree species. Experiments show that our 3D deep representation together with our geometric approach leads to significant improvement over the state-of-the-art methods in segmentation task. Keywords Point cloud segmentation · Terrestrial LiDAR · Unbalanced data-set · Deep learning
1 Introduction This paper presents a novel method, based on a deep learning approach, designed to address a major challenge in point cloud computing: segmenting leaves and wood in point clouds acquired in forests environments. The automatic processing of 3D point clouds has received increasing attention with the emergence of close-range 3D acquisition technologies, such as time-of-flight cameras and laser scanners. Those scanning processes have a wide range of applications: Architecture, urban planning, medical imaging and support for self-driving cars are some of their most noted areas of use , while assessing features of natural environments such as forests is a major challenge ahead. As the accuracy of such devices allows to produce extremely faithful point clouds of the geometry surroundings, their high acquisition rate comes with a trade-off: Massive amount of data are produced that need to be further filtered, classified and reconstructed in order to extract any relevant geometric information.
B
Jules Morel [email protected]
1
Kanai Laboratory, Graduate school of Arts and Sciences, University of Tokyo, Tokyo, Japan
2
Laboratoire d’Informatique et des Systèmes, Aix Marseille University, Marseille, France
In the recent years, the capability of Terrestrial Laser Scanning (TLS) devices to capture detailed information about the structure of the environment surrounding the sensor has attracted increasing attention in the field of forest science. Indeed, TLS enables 3D forest geometric information to be acquired at high speed [12], with applications ranging from ecology to forestry (forest monitoring, sustainable development) and industry (harvest planning, sawmill optimization). In the field of forestry, TLS data have, for instance, been successfully used to enhance allometric theory [37]. This theory consists of a set of general relations derived from a large compilation of fore
Data Loading...
