Publications / 2019 Proceedings of the 36th ISARC, Banff, Alberta, Canada

Web-based Deep Segmentation of Indoor Point Clouds

Zehuan Chen, Erzhuo Che, Fuxin Li, Michael Olsen and Yelda Turkan
Pages 552-559 (2019 Proceedings of the 36th ISARC, Banff, Alberta, Canada, ISBN 978-952-69524-0-6)

Deep learning and neural networks have empowered many tasks including semantic segmentation and image classification. Recently, novel neural networks were proposed that could directly process three-dimensional (3D) point clouds. In this paper, we present a software tool incorporating a variety of measuring tools to analyze and validate raw point cloud data while enabling an interactive segmentation of point clouds using deep learning. One of the primary advantages of implementing deep learning for point cloud segmentation is that it enables feature extraction to be learned through neural networks based on a large amount of data. Our software tool allows users to visualize 3D point cloud data sets containing millions of points in standard web browsers and process 3D point clouds using deep segmentation with deep neural networks. The interaction tool can assist with distinguishing structural buildings elements from non-structural objects in a given dataset.

Keywords: Point cloud; Deep learning; Web-based interaction