Goal-driven three-dimensional object inspection from limited view backprojection reconstruction
Abstract
The automatic inspection of three dimensional objects is an important part of manufacturing. This process will allow detection of defects relating to the size and shape of parts during manufacturing and assembly. Tomography techniques which are currently in use in medical applications are too expensive and computationally intense for use in manufacturing. These techniques also provide more information about the object than is typically required for inspection applications. For example, a machined part may only need to have certain exterior shape characteristics for it to be properly manufactured. The inspection problem will be approached using backprojection reconstruction techniques. This basic technique will be modified into a goal driven algorithm to inspect certain locations on the surface of the object which we shall call landmarks. These landmarks could be derived from the CAD phase of object design. The use of these landmarks will allow the inspection system to use a limited number of views in its analysis. The actual reconstruction process will make use of a modified volume projection technique which will be employed to make maximal use of the information incorporated into the views of the object. A related problem to the inspection analysis is the interpretation of the results of the inspection. All information derived during inspection analysis is three dimensional in nature. A new technique for the interpretation of the inspection results is proposed which is based on the goal driven nature of our system. The limitations of inspection accuracy as they relate to the system is also examined. Before the inspection process can occur, it is necessary for the inspection views of the object to be selected. This selection process is not obvious because it is not apparent how different views interact with each other in their supplying information to the inspection process. A goal driven algorithm is presented which removes the redundancy within a set of needed views of the object. This goal driven view selection is consistent with the goal driven nature of the inspection process. This entire process assumes that the object is capable of being inspected. An object characterization, called Weak Convexity, is presented for objects to be reconstructed using backprojection reconstruction. This characterization is sufficient but not necessary for inspection to be possible. Inspection is possible if the object has this property locally, at each landmark of interest.
Degree
Ph.D.
Advisors
Allebach, Purdue University.
Subject Area
Electrical engineering|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.