Sharif Mansouri, Sina (Luleå University of Technology), Kanellakis, Christoforos (Luleå University of Technology), Pourkamali-Anaraki, Farhad (University of Massachusetts Lowell), Nikolakopoulos, George (Luleå University of Technology)

Towards Robust and Efficient Plane Detection from 3D Point Cloud

Scheduled for presentation during the Regular Session "UAS Applications I" (ThA4), Thursday, June 17, 2021, 11:30−11:50, Naoussa

2021 International Conference on Unmanned Aircraft Systems (ICUAS), June 15-18, 2021, Athens, Greece

This information is tentative and subject to change. Compiled on April 16, 2024

Abstract

This article proposes a robust and scalable clustering method for 3D point-cloud plane segmentation with applications in Micro Aerial Vehicle (MAVs), such as Simultaneous Localization and Mapping (SLAM), collision avoidance, and object detection. Our approach builds on the sparse subspace clustering framework, which seeks a collection of subspaces that fit the data. Since subspace clustering requires solving a global sparse representation problem and forming a similarity graph, its high computational complexity is known to be a significant drawback, and performance is sensitive to a few hyperparameters. To tackle these challenges, our method has two key ingredients. We use randomized sampling to accelerate subspace clustering by solving a reduced optimization problem. We also analyze the obtained segmentation for quality assurance and performing a post-processing process to resolve two forms of model mismatch. We present numerical experiments to demonstrate the benefits and merits of our method.