In the context of multi-camera scene image acquisition, camera calibration relies on the detection and matching of interest points between images of neighbouring views, allowing to retrieve the relative pose and properties of each camera.
We have acquired a set of uncalibrated images, using a Kinect and a Lytro camera. The captured scene includes non-Lambertian objects (i.e. with reflections and refractions) which jeopardizes conventional calibration processes.
The goal of this project/Master thesis is to identify and segment the non-Lambertian objects from the scene, using the specificities of Kinect and/or Lytro cameras, followed by a recalibration of the dataset.

Prerequisites:

C++ and/or python, opencv or image processing library, INFO-H-500

Keywords:

Computer vision, Image analysis, Segmentation.

Promotor:

Prof. Mehrdad TERATANI

Contact:

Sarah Fachada Sarah.Fernandes.Pinto.Fachada@ulb.ac.be