Robuste Flächenextraktion aus Time-of-Flight Kameradaten

Dirk Koltermann

Mastersthesis TU Kaiserslautern 5/2013.


A conceptional description of the surrounding area of a car is necessary for prospective driver assistance systems. Due to the fact that most structures in urban areas are planar, the following work will present a planar surface extraction algorithm that allows a robust real-time detection of planar segmentation that serves as the fundament for a subsequent global planar surface extraction. At first, horizontal and vertical line segments are extracted alongside a 2D-pattern and are combined to planar surface segments. The planar surface segmentation serves as initial point for a global surface extraction to increase the robust detection of planar surface structures. At the end a merging approach prevents an over segmentation of planar structures. Only the combination of these sub-steps leads to a robust detection of planar surface structures. The evaluation demonstrates the ability of the described algorithm to detect planar surfaces from 3D point data of complex scenarios in real-time. This algorithm leads to significant better results in comparison to state of the art planar surface extraction algorithms.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence