Les commentaires sont clos.

Stage Master II Institut Pascal : Multi-view tracklet matching for camera network self-calibration

9 December 2022

Catégorie : Stagiaire

Keywords: 2D and 3D vision, feature tracking, feature matching, machine learning

Internship objectives:
The aim of this project is the study of airflow in large cavities, such as atrium or conference rooms, by juxtaposition of several different 3DPTV systems. A procedure for multi-3DPTV multi-camera self-calibration will be developed in order to express all particle trajectories in a common 3D coordinate system. Self-calibration of camera networks aims to jointly estimate intrinsic and extrinsic parameters of the cameras by using natural features, i.e. the same features present during the normal use of the network, instead of known patterns designed for the calibration task. In order to reduce the degree of expertise needed for the 3DPTV measuring operator, self-calibrating technique should be developed that enables the camera network to configure itself solely by using particle tracking and matching. This task usually requires that enough features are seen simultaneously by at least two cameras, which is challenging in our context because of the difficulty to find a sufficient number of correct matches due to the density and the similarity in the appearance of the features. In order to boost the candidate matches generation, we will investigate the use of deep neural networks for both tracklet generation and matching steps building on state-of-the-art architectures such as Superglue.

This work is a part of the project TRAQ (Three-dimensional tracking of monodisperse TRAjectories by Quantitative measurements) founded by ANR. In buildings and more generally in confined ventilated spaces such as aircraft cabins, it is important to be able to predict the trajectory and velocity of the air and of airborne pollutants and contaminants. The challenge is both environmental and energetic. On the environmental front, the stake is the control of indoor air quality, a public health issue. Indoor air is indeed loaded with particles emitted by human breathing and activity and by indoor furnitures. These particles can be harmful to the occupants. For example, many chemical compounds such as formaldehyde or carbon monoxide can be fatal. Predicting the trajectory of such airborne particles makes it possible to assess the exposure risks of building occupants, helps optimizing the ventilation strategies and even the evacuation of people in the event of a fire, or the emission of a dangerous gas. The current Covid-19 crisis is a stern reminder on how important it is to predict the motion of indoor contaminants. In terms of energy, predicting the trajectory and velocity of the air allows optimizing heating or cooling systems dedicated to indoor thermal comfort. For example, in low energy buildings, it can help choosing the shape and position of the heating devices and air diffusers offering the best coverage of the living area, and therefore to reach the best comfort at the lower cost.

- Institut Pascal, is a joint interdisciplinary research and training unit of 400 people, under the twofold supervision of Université Clermont Auvergne (UCA) and CNRS. CHU University Hospital of Clermont-Ferrand is secondary supervisor of the unit. Institut Pascal is member of Clermont Auvergne INP, which includes three engineering schools ISIMA, POLYTECH Clermont and SIGMA Clermont.
- The ISPR (Image, Systèmes de Perception, Robotique) department of the Institut Pascal focuses research in Computer Vision and Robotics. The goal of ISPR is to develop theoretical, methodological and architectural concepts for the perception and control of systems in order to make them as autonomous as possible.

Contact and application:

Send a CV + Master transcripts (recommandation letters will be appreciated) to
Omar Ait Aider
Email :
Phone : 04 73 40 55 67