Date of Award
Master of Science
Holoimage is a technique that is capable of compressing 3D geometry scans into 2D images. The main goal of this thesis was to develop a way to compress 3D geometry coming from a structured light scanner into a manageable format. Structured light scanners have made the acquisition and display of 3D data simple. Recently an area of 3D scanning has emerged called realtime 3D scanning, which allows for the capture, reconstruction, and display to be realized in 3D. With these advancements comes the challenge of working with the large volume of data that comes from such a scanner. In an uncompressed format, these realtime 3D scanners can have data rates surpassing 400 MB/s. For realtime applications, this amount is unmanageable, thus a method to compress the data must be found.
Holoimage is a technique that was developed to compress data from such a scanner, converting the raw 3D geometry into 2D images which can then be compressed using 2D image compression techniques. This conversion from 3D to 2D is achieved through a virtual structured light scanner, which is similar to an actual structured light scanner with some key differences. In the virtual system, traditional sources of error such as lighting, camera properties, and system calibration can be controlled to provide an ideal scanning system. Thus, unlike traditional structured light systems, Holoimage does not suffer from disadvantages such as the inability to measure discontinuous surfaces or surfaces with large step height variations. Also, the Holoimage technique is constructed in such a way that all of the steps are pixel operations, thus it can be run on parallel hardware such as a graphics processing unit (GPU).
To further increase the compression of 3D geometry in the Holoimage format, 2D image compression such as portable network graphics (PNG) or joint photographic experts group (JPEG) can be applied. Since the JPEG format is a form of lossy 2D compression, this form of compression introduces error into the reconstructed 3D geometry. Investigations of this error are performed with three different experiments, drawing conclusions from each to construct a structured light pattern that is more resilient to the effects of this lossy compression. In the end, results are shown which allow for compression ratios of over (72 : 1) with root mean squared error of less than 0.1%. If further lossy compression is applied, compression ratios of over (370 : 1) can be achieved with root mean squared error of less than 4.0%. In all this thesis documents previous work in the area of 3D geometry compression, the principle of the Holoimage technique, methods to implement the technique on parallel hardware, experiments on properties of the resulting images, and avenues for future work on the technique.
Nikolaus L Karpinsky
April 30, 2012
Karpinsky, Nikolaus L., "3D geometry compression with Holoimage" (2011). Graduate Theses and Dissertations. 12090.