Jobs & Karriere

When combining the results of airborne laser scanning (LiDAR) and the results of airborne 3D reconstruction from optical sensors, one major problem is the alignment of the two point clouds in 3D space, called point cloud registration. The main idea is usually to identify corresponding points in both data sets and estimate a rigid body transformation that minimizes the distance between these corresponding points.

Finding the corresponding points is not a trivial task, as pure nearest neighbor search in Euclidean space only works for point clouds which are already very well aligned and tends to fail completely, if the offset in both rotation and translation is large.

In this work, Point Feature Histograms (PFH), describing the local neighborhood of each point, should be applied to find corresponding points in bad aligned, noisy and partially overlapping point clouds and to use them for the estimation of the point cloud alignment in an iterative way. The iterative estimation normally is done by using for example the Iterative Closest Point algorithm, with the distance metric exchanged to the one evolved by the Point Feature Histograms.

A special case are already triangulated point clouds, where additional information about the local neighborhood of each point can be derived from the mesh and the combination with the abovementioned PFH should be evaluated.

Tasks:

• Getting an overview of the State-of-the-Art algorithms
• Implementing Point Feature Histograms into the existing framework
• Evaluating the accuracy of the algorithms (and their modifications) on ground truth data