Robust Meshes from Multiple Range Maps: Slide 2 of 14.


Click slide for next, or goto previous, first, last slides or back to thumbnail layout.

Click slide for next, or goto previous, or back to thumbnail layout.

Surface reconstruction using a range scanner is often done in the following four steps.

First, one needs to scan the object and obtain data from all over its surface. Depending on the data one may have to preprocess it by filtering and by removing background.

The data sets are first expressed in sensor-based coordinates. Typically, the sensor moves with respect to the object between the scans, so the data sets do not align well. We need to register the data sets, that is rotate and translate them so that the data points do align. Registration can then be thought of as transforming the data into an object-centered coordinate system.

The separate views or data sets are then merged or integrated into some coherent representation. This representation serves as a convenient framework for optimizing the surface to data accurately. We used to use methods developed by Hugues Hoppe for his dissertation to get an initial mesh (step 3) and then optimize it (step 4). His methods were developed to work when the input is unorganized point clouds. Such methods are very general, but may fail to work with noisy data, or if the surface is not sampled uniformly.