Skip to Main content Skip to Navigation
New interface

Reconstruction 3D par stéréophotométrie pour la vision omnidirectionnelle

Abstract : This thesis focuses on the photometric stereo problem and the omnidirectional vision. The photometric stereo problem is a 3D-reconstruction technique which requires several pictures of a surface under different lighting conditions from a single point of view. The omnidirectional vision encompasses the devices and the rig of cameras that capture a large part of the environment around them in a single image. Following four decades of research in the photometric stereo literature and using the unified model for central projection cameras, we try to merge these research fields. We first focus on techniques for estimating the normals to the surface, and for integrating the depth gradients to retrieve the shape. Then, we introduce a new spherical irradiance equation that we use to solve the photometric stereo problem using two central projection cameras. The approach is validated using synthetic and real images from a perspective camera and a catadioptric imaging device. We later extends the approach to perform 3d-reconstruction by photometric stereo using twin-fisheye cameras. Finally, we study some limitations of the approch and we discuss the ways to overcome these limits
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Tuesday, October 25, 2022 - 2:19:21 PM
Last modification on : Wednesday, October 26, 2022 - 3:10:28 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03828820, version 1



Jordan Caracotte. Reconstruction 3D par stéréophotométrie pour la vision omnidirectionnelle. Vision par ordinateur et reconnaissance de formes [cs.CV]. Université de Picardie Jules Verne, 2021. Français. ⟨NNT : 2021AMIE0031⟩. ⟨tel-03828820⟩



Record views


Files downloads