Propositions for projects more oriented computer science, involving mostly deep learning (but not restricted to) (2021-2022)...
Extension of GIS tool based on 360 multimodal image navigation

We develop internally a GIS project (web based), based on 3D scanning acquisitions, coupled with photographs. The aim of this project is the extension of the tool (e.g. for automatic imports, dynamic colorization).

Stereo capturing by plenoptic and 2D cameras [ASSIGNED]

Abstract:

In a Plenoptic camera, through a microlens array (MLA), the real-world is captured from multiple angles. The MLA is placed next to the image sensing sensor, and a Lenslet (LL) image. A LL image which consists of light rays from different angles, contains the 3-D information of the scene. An LL image can always be converted to multiview (MV) images (e.g. 5x5). When a plenoptic camera is stereo-paired with a normal 2D camera due to different optics, the captured contents are not identical. In this project, we want to synthesize equivalent LL and/or MV images at the location of the 2D camera, given the images captured in LL and 2D formats by the stereo set up The challenges in the project consists of understanding the plenoptic camera optics, calibration, depth estimation and view synthesis.

Keywords:

Plenoptic camera, 3D acquisition systems, Calibration, Depth estimation, View synthesis

Supervisor:

Mehrdad Teratani (mehrdad.teratani@ulb.be), Gauthier Lafruit (gauthier.lafruit@ulb.be)

Director:

Mehrdad Teratani (mehrdad.teratani@ulb.be)

Deep neural network image processing

Please register to UV for PROJH419 for more details

contact:

Olivier.debeir@ulb.be

Chest XR covid detection [ASSIGNED]
Multi-camera pose estimation using MediaPipe [ASSIGNED]
Objet : Intégration d'un système de motion capture dans le Sam.
Le Sam est un dispositif médical multisensoriel qui génère des environnements calculés en temps réel, basés sur les besoins émotionnels des patients.
Afin de générer des feedbacks en fonction des mouvements du patient, un dispositif de computer vision va être intégré à l’architecture existante du Sam.

Le système de capture doit répondre aux objectifs suivants :
- Captation sur un seul individu à la fois.
- Système capable de fonctionner avec une luminosité variable
- Le patient est susceptible de tourner à 360°. Dans la mesure du possible, le système d'aquisition doit prendre en compte les angles morts. (Triangulation avec plusieurs caméras, par exemple).
- Le degré de précision de la captation ne doit pas être 100% accurate.

Le projet tourne sur le moteur de rendu 3D Unreal Engine 5 (c++). Le framework mediapipe semble le bon candidat pour l'aquisition des données, mais des questions subsistent concernant la capture de la profondeur et des angles morts.

www.inmersiv.com - maxime@inmersiv.com

Simulation of a tensor display by taking into account aberrations caused by the actual display

Summary: Tensor displays are one kind of 3D display. It allows seeing different images from different points of view without wearing any glasses or helmet. One of the major problems concerning tensor displays is that prototypes do not reproduce exactly what is expected as in simulation. Indeed, the materials that are used to build a prototype cannot be considered to work perfectly and thus introduce noise in the reproduction of the 3D content. The goal of this project aims to simulate a model of tensor display that includes non-perfect behaviors of the materials in order to see the impact of the imperfections by simulation. The student is required to (1) do a simulator of a tensor display, (2) model some realistic imperfections that could occur, (3) an (some) interactive method(s) (GUI, command-line, etc.) to adjust the imperfection factors.

Skill required: C/C++ programming, Java/C#

Interested in: computer vision, virtual reality, and programming.

Director: Mehrdad Teratani (mehrdad.teratani@ulb.be)

Supervisor: Mehrdad Teratani, Gauthier Lafruit

Support and contact: Armand Losfeld, Eline Soetens

Updated on March 30, 2023