Web-based Deep Segmentation Building Structure Assisted User Interactions

This thesis introduces a web-based deep segmentation user interaction assistance on large point clouds. It allows users to view data sets with billions of points, from sources such as building structure and indoor scene, in standard web browsers, and processing 3D point clouds deep segmentation with the neural network. Since the deep learning and the neural network has been widely used in recent in research and industrial area, deep learning has empowered many tasks such as point clouds segmentation and shape recognition. One of the main advantages of point cloud interaction with deep learning in the web browser is that it allows users to share data sets and doing interaction with a neural network on any location without having powerful computational resources on the current device. Furthermore, there is no need to install third-party applications and transform huge amounts of data in advance. Our focus on large point clouds, and a variety of measuring tools analyze and validate raw point cloud data. The interaction tools allow the user to distinguish the building structure and non-building structure in one room. The result is a point cloud viewer and assisted user interactions interface, which was able to render point cloud data sets in two different ways, which are panoramic image view and perspective 3D view on a web browser.

Major Advisor: Fuxin Li
Committee: Raffaele De Amicis
Committee: Michael James Olsen
GCR: Yelda Turkan