Tuesday 12th of October, Younes Zegaoui, Ph.D. Candidate in Montpellier University will defend his thesis named “Urban objects detection with 3D multi-sensor data registered continuously at the ground level”. His thesis defense will take place at LIRMM laboratory in Montpellier, France. The jury will be composed of François Goulette, Professor, Ecole MINES ParisTech, Sébastien Lefèvre, Professor, Bretagne Sud University, Géraldine Morin, Professor, Toulouse University, Christophe Fiorio, Professor, Montpellier University, Marc Chaumont, MCF HDR, Nîmes University, and Gérard Subsol, Resarch Officer CNRS, LIRMM.
LiDAR terrestrial mobiles acquisition devices development set on vehicles or drones allow to digitise entire cities in georeferenced three-dimensional point cloud. Data exploitation of the point cloud could permit to city managers to census and monitore urban objects whether they are motionless (street lights, bus stop), mobile (bins) or natural (trees) to be able to react in case they disappear, they are displaced, deteriorated or present a potential danger.
This approach requires to treat large point cloud with hundreds of millions of points and thousands of objects. Automatize the point cloud process is necessary to extract and classify elements as urban objects. In this thesis , we’ll explore the recent track of deep learning applied to unstructured data to localize and recognize urban objects in a 3D point cloud.
We developed a 3D neuronal architecture based on an original layer allowing to group the points and extract their characteristics simultaneously. From this architecture, we’ll present the results we obtained on urban objects detection in the LiDAR point cloud data from big cities streets.