Making Sense of Geometric Data
A. C. OztireliComputer Graphics and Applications, IEEE, vol. 35, no. 4, 2015, pp. 100-106
Abstract
Most data acquired from the real world is or can be interpreted as geometric in nature. Advanced and affordable sensors, printers, displays, and the Internet make geometric data increasingly important for many disciplines. Giving structure and meaning to this data has been one of the main challenges of computer graphics as well as other fields in the last few decades. The author's PhD thesis started as an effort to turn this massive amount of data into digitally meaningful representations useful for various applications in computer graphics and beyond. In turn, his work targets the problems of reconstructing manifold surfaces and stochastic point patterns from unstructured point samples.
Overview
Most data acquired from the real world is or can be interpreted as geometric in nature. Advanced and affordable sensors, printers, displays, and the Internet make geometric data increasingly important for many disciplines. Most geometric data comes in the form of unstructured point samples. Giving structure and meaning to this data has been one of the main challenges of computer graphics as well as other fields in the last few decades.Vast amounts of geometric data are collected in many fields such as medical imaging, robotics, geography, seismology, architecture, and archeology, just to name a few. The data can hence represent many different structures. The datasets are massivea conventional depth camera with a frame rate of 30 frames per second (fps) can easily gener- ate billions of points in minutesbut the acquired data are far from perfect, with noise, outliers, and missing parts.
My PhD thesis started as an effort to turn this massive amount of data into digitally meaningful representations useful for various applications in computer graphics and beyond. We relied on the observation that the majority of geometric data in computer graphics and many other fields rep- resent object surfaces and repetitive structures. We thus targeted the problems of reconstructing manifold surfaces, which are smooth watertight surfaces bounding objects, and stochastic point patterns that are random distributions of points with certain characteristics, from unstructured point samples.
