Skip to Main content Skip to Navigation

Traitement automatique de time-lapses stéréoscopiques de milieux naturels : application à la sureveillance de glaciers alpins

Abstract : Over the past decade, multiview photogrammetry and object tracking techniques based on proximal imaging data have made it possible to generate, at relatively low cost, various types of data describing directly or indirectly the three-dimensional shape and dynamics of indoor, urban and natural environments.In natural environments, technological and methodological developments are still needed to prove the potential of such tools for precise, metric and automatic geomorphological applications. Most of the existing techniques are not particularly designed andthought out in order to process data of high spatial and temporal definition, marred by important photometric and atmospheric artifacts and characterized by variable visibility and a dense and semi-rigid displacement.In this thesis, we try to solve these problems by proposing an automatic processing chain of stereoscopic time-lapse images for the observation of landslides such as the flow of alpine glaciers. We first study the different possible strategies for processing stereoscopic time-lapses of natural environments according to a multi-level taxonomy of the available data : monoscopic and stereoscopic, monodate and bidate, temporal and multi-temporal. We then propose a complementary suite of robust automatic processings based on a justified choice of geometric and temporal representations and models governing the link between these different types of data.In particular, we evaluate the uncertainty via a geometric approach and maximize the coherence of the metric velocity estimates derived from the fundamental bricks of the processing chain via the optimization of a functional taking into account spatio-temporal descriptors, a temporal closure constraint of intra-timelapse displacement fields and finally the direction and stationarity of the flow of alpine glaciers.This work is motivated by the current context of global warming and its impact on the cryosphere.The continuous proximal monitoring of alpine glaciers via a dedicated observation system opens up new perspectives for the quantification of metric indicators correlated to the evolution of temperature and precipitation.We apply the proposed methods to data acquired on the Argentière glacier site in the Mont-Blanc Massif. Tracking time series of the geo-referenced metric velocity of part of this glacier are thus generated with an estimate of their uncertainties on the data set acquired in 2018.The results show that it is possible to develop robust geo-photogrammetric proximal monitoring applications in extreme and uncontrolled climatic environments, provided that the characteristics of the measurand and a-priori knowledge of the observed scene are taken into account.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, January 19, 2022 - 2:53:12 PM
Last modification on : Wednesday, March 2, 2022 - 3:41:24 AM
Long-term archiving on: : Wednesday, April 20, 2022 - 6:49:55 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03534542, version 1



Héla Hadhri. Traitement automatique de time-lapses stéréoscopiques de milieux naturels : application à la sureveillance de glaciers alpins. Traitement des images [eess.IV]. Université Savoie Mont Blanc, 2020. Français. ⟨NNT : 2020CHAMA012⟩. ⟨tel-03534542⟩



Record views


Files downloads