Contribution to wide angle visual perception for mobile robots and self driving cars

Abstract : This thesis presents a novel projection model for fisheye cameras, which is mathematically simple and yet shows a high precision when applied to real cameras. Geometric properties of the model have been analyzed using the concept of projection surface, introduced in this work. In particular, a closed-form inverse mapping and an implicit equation for straight line projection have been found. This fact has been used to develop a method of direct stereo correspondence on raw fisheye images via rasterization of implicit curve. This correspondence algorithm allows us to apply the Semi-Global Matching algorithm to get an accurate 3D reconstruction using fisheye stereo systems. All these elements have been shown to be applicable to a direct visual localization system with two different methods of image registration: direct photometric error minimization and mutual information maximization. Intrinsic and extrinsic calibration of a mobile robot with fisheye cameras has been considered and a toolbox for such a calibration has been developed.
Liste complète des métadonnées

https://tel.archives-ouvertes.fr/tel-01983700
Contributor : Abes Star <>
Submitted on : Wednesday, January 16, 2019 - 4:28:15 PM
Last modification on : Tuesday, March 26, 2019 - 9:25:22 AM

File

B_KHOMUTENKO.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01983700, version 1

Collections

Citation

Bogdan Khomutenko. Contribution to wide angle visual perception for mobile robots and self driving cars. Signal and Image processing. École centrale de Nantes, 2018. English. ⟨NNT : 2018ECDN0010⟩. ⟨tel-01983700⟩

Share

Metrics

Record views

143

Files downloads

45