Skip to Main content Skip to Navigation

Raffinement de la localisation d’images provenant de sites participatifs pour la mise à jour de SIG urbain

Bernard Semaan 1
1 CRENAU - Centre de recherche nantais Architectures Urbanités
AAU - Ambiances, Architectures, Urbanités
Abstract : Cities are active spots in the earth globe. They are in constant change. New building constructions, demolitions and business changes may apply on daily basis. City managers aim to keep as much as possible an updated digital model of the city. The model may consist of 2D maps but may also be a 3D reconstruction or a street imagery sequence. In order to share the geographical information and keep a 2D map updated, collaborative cartography was born. "" platform is one of the most known platforms in this field. In order to create an active collaborative database of street imagery we suggest using 2D images available on image sharing platforms like "Flickr", "Twitter", etc. Images downloaded from such platforms feature a rough localization and no orientation information. We propose a system that helps finding a better localization of the images and providing an information about the camera orientation they were shot with. The system uses both visual and semantic information existing in a single image. To do that, we present a fully automatic processing chain composed of three main layers: Data retrieval and preprocessing layer, Features extraction layer, Decision Making layer. We then present the whole system results combining both semantic and visual information processing results. We call our system Data Gathering system for image Pose Estimation (DGPE). We also present a new automatic method for simple architecture building detection we have developed and used in our system. This method is based on segments detected in the image and was called Segments Based Building Detection (SBBD). We test our method against some weather changes and occlusion problems. We finally compare our building detection results with another state-of-the-art method using several images databases.
Complete list of metadata

Cited literature [156 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Tuesday, April 16, 2019 - 2:14:17 PM
Last modification on : Tuesday, March 30, 2021 - 3:06:12 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02101016, version 1



Bernard Semaan. Raffinement de la localisation d’images provenant de sites participatifs pour la mise à jour de SIG urbain. Traitement du signal et de l'image [eess.SP]. École centrale de Nantes; Université libanaise, 2018. Français. ⟨NNT : 2018ECDN0055⟩. ⟨tel-02101016⟩



Record views


Files downloads