Abstract
The large amount of irregular and unstructured data collected by the laser scanner, often refered to as “cloud data”, presents a serious challenge in reverse engineering applications. Data reduction is essential for simplifying the process of creating surface models in order to remove redundancy, save space, and speed up the display and recognition tasks. In this paper, an algorithm which effectively reduces the data collected by laser scanners is presented. Data points are first organized into a large regular grid of data points which is then interpolated using the linear polynominal dual Kriging technique to form the initial surface. Next, the algorithm generates a subset of this grid (within the specified deviation) by identifying the knots that play important role in the Kriging interpolation. Finally, additional reduction is done by merging the rectangles of the grid where the tangent variation between two rectangles is insignificant. Since the final surface is usually represented by a set of connected rectangles, the surface segmentation and fitting procedures can be accomplished in a short period of time. Examples are used to show the efficiency and validity of the proposed algorithms.
