NICOP - Detecting and Extracting Surfaces from Unorganized Point Clouds
Abstract
Reconstructing virtual replicas of real environments holds an enormous potential for automaticdocumentation of complex installations. For instance, consider the large number of ships,power plants, factories, etc., designed and built based on blueprints and still in operation. Formany of these, the blueprint records do not match the actual installations. The lack of up???todateCAD models makes the design of replacing engines, turbines, and general parts a delicateoperation, as they should be manufactured elsewhere and only brought in when ready to beinstalled. Thus, the ability to obtain CAD models for existing structures has great practical andeconomic value. Unfortunately, creating models of real environments is a complex task forwhich the use of traditional 3???D modeling techniques is often inappropriate. For this task, laserrangefinders are frequently used to sample the scene from several viewpoints, with theresulting range images integrated into a final model. In practice, due to surface reflectanceproperties, occlusions, and accessibility limitations, certain areas of the scenes are usually notsampled, leading to holes and introducing undesirable artifacts in the resulting models.An ideal reverse engineering system for complex environments should then be able tohandle large, noisy, unorganized point clouds representing incompletely sampled surfaces, andshould output a CAD model for that environment. Since the datasets are expected to containfrom millions to billions of samples, computational efficiency and scalability are importantrequirements. While obtaining such an ideal system is a very ambitious goal, this projectintends to perform fundamental initial steps in this direction by systematically addressing theproblem. The idea is to break the problem into sub???problems and conquer them separately. Theoperations should be performed in such an order that the completion of each step simplifiesthe following ones. Thus, for instance, since man???made environments often contain many largeplanar surfaces representing floors, ceilings, walls, etc., the procedure will start by detectingplanar surfaces. Once detected, they will be added to the CAD model and their correspondingsamples removed from the point cloud. The resulting reduced point cloud will then be analyzedfor other structures commonly found in ships, power plants, industrial plants, etc., such ascylindrical (i.e., pipes) and spherical surfaces. Once these have been detected, added to theCAD model and removed from the point cloud, the system can search for more specializedstructures, such as propellers, valves, elbows, tees, etc., depending on the characteristics of thespecific environment. Such an approach should provide a framework for systems capable ofautomatic or semi???automatic recovery of CAD models for existing installations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 04, 2018
- Source ID
- N629091812131
Entities
People
- Manuel Menezes De Oliveira Neto
Organizations
- Fundação de Apoio à Universidade do Rio Grande
- Office of Naval Research
- United States Navy