Girardeau-Montaut, Daniel (2006) Change Detection on 3D Geometric Data. PhD thesis Traitement du Signal et des Images, TSI/TII, ENST.

Full text available as: - manuscrit_19052006_electronic.pdf ( 24308 Kb )

Licence: Copyright Daniel Girardeau-Montaut (ENST TSI/TII & EDF R&D)

Abstract

This work deals with the analysis of 3D point cloud data acquired by ground-based laser sensors in order to detect 3D geometrical changes in complicated environments (such as power plants or oil platforms). Ground-based laser sensors are compact and portable devices that are able to acquire rapidly and accurately a large amount of 3D points on the surface of objects. These point clouds are very dense and detailed, but also huge and unstructured. Therefore, we have developed fast and ad-hoc algorithms to deal with such datasets.
This manuscript presents the study and effective implementation of a complete change detection framework based on 3D point clouds and 3D mesh models. We mainly propose a method to compute distances directly between two 3D point clouds in a robust and accurate way, and also two segmentation algorithms. The first segmentation algorithm is a "region based" approach that relies on a local statistical test. It classifies points of a cloud in two categories, depending on the distribution of distances computed for each point and its nearest neighbours. The second algorithm is an "edge based" approach that relies on a propagation scheme constrained by the gradient norms of the computed distances. We also propose a specific coding scheme of an "octree" structure. Such a structure can be computed very efficiently and in such a way that most of the point based treatments can be optimized and the global process is therefore greatly accelerated.
Eventually, we present several industrial applications of our framework. We also investigate the potential advantages that it could bring to a crisis management process, as it is fast and accurate.

Statistiques de consultation

Repository Staff Only: edit this item