24 Feb
14:00 Master's Defense Fully distance
Monocular Visual Odometry Based on Epipolar Geometry
Ronnypetson Souza da Silva
Advisor / Teacher
Esther Luna Colombini
Brief summary
In Monocular Visual Odometry there are several approaches related to the optimization of pose and structure. Some allow the use of fine-grained observations (eg, points of interest) but require a large number of parameters. This limits the density of the structure reconstruction and the number of poses that can be optimized in real time. We investigated a characterization with few parameters of Visual Odometry that consists only of poses and the positions of epipoles in pairs of images. We do this by formulating an indirect type of bundle adjustment where structure parameters such as inverse point depths are replaced by epipoles from the geometry of two points of view. The proposed approach dramatically reduces the number of parameters and still allows some control over the structure optimization. To evaluate the method in detail, we performed odometry and scene reconstruction in the KITTI database.
Examination Board
Esther Luna Colombini IC / UNICAMP
Luiz Marcos Garcia Gonçalves DCA / UFRN
Helio Pedrini IC / UNICAMP
Sandra Eliza Fontes from Avila IC / UNICAMP
Reinaldo Augusto da Costa Bianchi FEI