Point Cloud Simplification and Processing for Path-Planning

Abstract

Recently the area of motion planning research has been experiencing a significant resurgence of interest based on hybrid working environments that combine point and CAD models. Companies are able to work with point clouds and perform certain operations, such as path-planning, but they lack the support for fast shortest-distance computations for point clouds with more than tens of millions of points. Therefore, there is a need for handling and pre-processing massive point clouds for fast-queries. In this thesis, algorithms have been developed that are capable of efficiently preprocessing massive point clouds for fast out-of-core queries allowing rapid computation of the exact shortest distance between a point cloud and a triangulated object. This is achieved by exploiting fast approximate distance computations between subsets of points and the triangulated object. This approach was able to compute, on average, the shortest distance in 15 fps for a point cloud having 1 billion points, given only 8 GB of RAM. The findings and implementations will have a direct impact for the many companies that want to perform path-planning through massive point clouds since the algorithms are able to produce near real-time distance computations on a standard PC.