The main objective of this post-doc is to keep a 3D model (textured and semantically annotated surface mesh built from mobile mapping data) up to date from more recent (but lower cost) data. Changes between the initial acquisition and the more recent one are to be detected and classified to retain the most pertinent and permanent ones. Then these pertinent changes are to be seamlessly integrated in the 3D model.
Full proposal available at:
Key Words: Change detection, Update, surface mesh, texture, semantic segmentation, Image, Lidar, Mobile Mapping
Context: Mapping and localization are two essential elements in the development of a mobility application for people or smart vehicles. Those tasks that have received intense attention in recent years are considered among the most complex perception problems. They require a large processing capacity with severe time constraints and a need for bulky storage when it is desired to cover large spaces. This service involves a large number of actors, and the information delivered must remain valid and accurate over time, requiring frequent update campaigns. The pLaTINUM project aims to develop a geo-referenced global 3D map, stored on an application cloud, and consisting of a set of 3D representations characterized by geometric, photometric and semantic information. This map must be able to both automatically update and enrich its content through information transmitted by remote agents and provide a back-up service. The dual purpose of this collaborative application is to be able to use the information contained in the repository to provide guiding information to agents who in turn can inform the cloud of differences found locally that will be processed in the cloud to update the geo-referenced global repository.
Objectives: This post-doc is integrated in the pLaTINUM pipeline. We assume that agents are able to detect parts of their environment that have changed compared to the current map [2, 3, 4, 5]. When this happens, the agent sends its own view of the change, under the form of a RGB image cropped around the detected change, with possibly depth and/or semantic information on this change. This image is also assumed localized relatively to the maps. The first objective is to classify the changes, in particular between the sustainable or not. Sustainable changes are changes identified to be sufficiently stable and long term to deserve being integrated in the global map. For instance, mobile objects or temporary installations are non sustainable, while structural changes to buildings or urban furniture are. This identification can be made based on geometry, photometry and semantics. The second objective is to integrate the sustainable changes in the global map in order to keep it up to date, possibly by stitching a piece of mesh corresponding to the new scene geometry in place of the old one . This update of the global map should integrate all the aspects of the map (geometric, photometric, semantic), and handle possible occlusions due to the fact that various agents might have acquired the same parts of the scene from different positions. Moreover, this update should be able to handle data from sensors with very variable quality, so the work will focus on qualifying the quality of the updates depending on the sensor.
Environment: The contacts and coadvisers for this post-doc are:
Bruno VALLET is full time researcher at the MATIS Team and coordinator of the work package of the pLaTINUM project on the global map. Cédric DEMONCEAUX is with the Le2i lab, ERL VIBOT CNRS 6000 and also participates to the pLaTINUM project.
The post-doc will take place in the MATIS Team from the LaSTIG Lab of IGN (Institut National de l'Information Géographique et Forestière), located at Saint-Mandé (close to Paris, metro line 1, Saint-Mandé station). It will last 18 months starting june to august 2018.
The MATIS team is located at Saint-Mandé, bordering Paris in France. It depends on the Research Unit in Geo-Information Science of the French Mapping Agency (IGN), which itself belongs to the Research and Teaching Department of IGN. The MATIS team leads research activities in the fields of mathematics and computer science applied to photogrammetry, computer vision and remote sensing dedicated to ground-based, aerial and satellite multi-sensor imagery (optical, LiDAR, radar, etc.)
Application: Interested candidates must send to the contacts, in a single pdf before june 1st 2018:
 S. D. Brandão, M. Veloso, J. P. Costeira. Joint alignment and stitching of non overlapping meshes. Proc. of IROS 2012 - IEEE/RSJ International Conference on Intelligent Robots and Systems, Workshop on Color-Depth Camera Fusion in Robotics, VilaMoura, Portugal.
 C. Jiang, D. P. Paudel, Y. Fougerolle, D. Fofi, C. Demonceaux. Static and Dynamic Objects Analysis as a 3D Vector Field. In Int. Conf. on 3DVision (3DV 2017), Octobre 2017, Qingdao, Chine .
 C. Jiang, D. P. Paudel, Y. Fougerolle, D. Fofi, C. Demonceaux. Incomplete 3D Motion Trajectory Segmentation and 2D-to-3D Label Transfer for Dynamic Scene Analysis. In IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS 2017), Septembre 2017, Vancouver, BC, Canada.
 W. Xiao, B. Vallet, M. Brédif, N. Paparoditis. Street environment change detection from mobile laser scanning point clouds. ISPRS Journal of Photogrammetry and Remote Sensing, vol. 38, pp. 38-49, 2015
 W. Xiao, B. Vallet, N. Paparoditis. Change Detection in 3D Point Clouds Acquired by a Mobile Mapping System. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume II-5/W2, pp.331-336, Antalya, Turkey, November 2013
(c) GdR 720 ISIS - CNRS - 2011-2018.