Efficient large scale acquisition of building interiors
Abstract
The present study describes a novel system for creating digital 3D models of building interiors. The system is efficient, which allows modeling indoor environments with tens of rooms and thousands of square meters of floor space in a few days. The models obtained enable photorealistic virtual walkthroughs at interactive rates and are suitable for computer graphics applications such as virtual training, cultural heritage preservation, real-estate marketing, and video gaming. The system implements an automated modeling pipeline that runs at interactive rates. The operator sweeps the scene with a novel acquisition device that consists of a video camera and a laser system positioned on a tripod in the center of the scene. The laser casts an 11x11 matrix pattern of beams into the field of view of the video camera. The laser dots are detected in the video frame and converted to 3D points by triangulation. By construction, the laser beams project onto the video frame as disjoint line segments which makes dot detection efficient and robust. The dense color (720x480) and sparse depth (11x11) frames are registered and merged into a 3D model at the rate of 5 frames per second. The evolving model is rendered continually to provide immediate feedback to the operator. The operator aims the acquisition device to capture missing surfaces and adapts the depth sampling resolution to the local scene complexity. A model section (e.g. a room or a corridor segment) is created in minutes. The building model is assembled from sections by leveraging same-plane constraints specific to indoor architectural environments and minimal operator input. We validated the system by modeling the interior of a large building on campus. The model spans 6 floors consisting of 20 rooms and corridors. The model was acquired by a two-person team with a single acquisition device in 40 hours. To the best of our knowledge, the model is the largest reconstruction of an indoor scene to date.
Degree
Ph.D.
Advisors
Popescu, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.