Identification and classification of objects in 3D point clouds based on a semantic concept
Jean-Jacques Ponciano, Frank Boochs, Alain Trémeau · HAL (Le Centre pour la Communication Scientifique Directe) · 2019
Our real world is increasingly subject to digitization processes producing huge unstructured3D data sets. In order to extract objects contained in these data sets subsequent analysissteps are necessary. The most efficient but also the most expensive and time-consuminganalysis is based on manual editing, which allows integrating human knowledge andintelligence. Computer based methods are less effective, as they mainly use implicitknowledge allowing to parametrize algorithms, which are part of a defined processingchain. We want to overcome these limitations through a more general and more flexibleintegration of any kind of useful knowledge into the processing. This paper presents anapproach fully driven by semantic technologies and uses expert knowledge. This expertknowledge is modeled into an ontology and describes objects, data, and algorithms. Thisontology guides an iterative reasoning process using the semantic technologies (e.g. OWL2,SPARQL, an engine reasoner) to provide a prime flexibility. This process firstly identifiesrelevant algorithms to parametrize and combine them. Secondly, it interprets the resultprovided by algorithms to enrich the knowledge base. Thirdly, a semantic reasoning usesadded information to classify objects and determine, if needed, other relevant algorithms.This process is thus, dynamically and iteratively adapted to the results of executedalgorithms. Its efficiency is presented through the result comparison of its results accordingto other possible approaches.