Master internship: Uncertainty-Aware Video Processing for Structural Health Monitoring

Context
Inria is the French National Institute for Research in Digital Science and Technology. As a leading public research institution, Inria plays a key role in driving digital transformation in France. Its core activities include research in computer science, mathematics and artificial intelligence (AI), software development, innovation in technology-intensive fields, and DeepTech entrepreneurship. Inria ranks 16th worldwide in AI research and is Europe's leading institute for exploratory research in digital science and technology.
The internship will take place at the Inria Centre at the University of Rennes, within the
I4S research team
, under the supervision of
Michael Doehler
and
Qinghua Zhang
. It will be carried out in collaboration with Gustave Eiffel University (Nantes campus, SII laboratory), under the supervision of Vincent Baltazart.
Assignment
Can we estimate how reliable a measurement is without knowing the true value of the measured quantity? And how can we extract useful information from high dimensional noisy data for a given purpose?
Video-based motion measurement offers a promising way for Structural Health Monitoring (SHM) of bridges, buildings, and wind turbines without installing physical sensors on the structures. By processing video recordings, it is possible to extract motion signals at thousands of points, providing a remarkably detailed picture of structural vibration behavior. However, two major challenges remain: assessing the reliability of these signals and efficiently processing the resulting large datasets. This internship will address both challenges by developing methods at the intersection of signal processing, uncertainty quantification, and mathematical optimization.
Estimating measurement uncertainty without reference sensors
Video-processing algorithms can extract motion signals at thousands of image locations, effectively creating a dense network of virtual sensors. Yet, assessing the accuracy of these signals usually requires reference measurements from high-precision physical sensors, something that is often unavailable in real-world applications. You will investigate how the redundancy between neighboring virtual sensors can be exploited to estimate measurement uncertainty directly from video-derived signals, without requiring ground-truth measurements.
Developing smarter data reduction methods
Processing thousands of motion signals can be computationally expensive and may introduce unnecessary noise into subsequent analyses. You will develop dimensionality reduction techniques based on orthogonal projections that maximize the signal-to-noise ratio. The objective is to retain the most useful information while reducing data size and improving the reliability of subsequent SHM algorithms.
References
:
Luo, B. Merainani, V. Baltazart, Q. Zhang, M. Döhler.
Subpixel motion estimation for video-based target-free vibration monitoring under complex environmental conditions
.
Mechanical Systems and Signal Processing
, 2025
Luo, B. Merainani, M. Döhler, V. Baltazart, Q. Zhang.
High dimensional data reduction in modal analysis with stochastic subspace identification
.
22nd IFAC World Congress
, 2023
Main activities
Literature review on uncertainty quantification and dimensionality reduction for video-based motion measurements.
Analysis of video-derived motion signals and their statistical properties.
Development and implementation of uncertainty quantification methods.
Design and implementation of dimensionality reduction algorithms aiming at signal-to-noise ratio optimization.
Validation and performance assessment of the proposed methods using experimental data and existing identification algorithms.
Skills
We are looking for a motivated Master’s student (M2) or final-year engineering student with:
A solid background in applied mathematics, signal processing, statistics, or a related field.
An interest in developing algorithms and applying mathematical methods to real-world problems.
Programming experience in Python, MATLAB, Julia, or an equivalent language.
Good English communication skills.
Curiosity, scientific initiative, and an interest in pursuing research, potentially through a PhD.
Duration:
5-6 months (flexible start date between January and April 2027)
Applications must include the following documents:
A detailed CV
A transcript of records for the Master’s degree
A cover letter explaining the applicant’s motivation and detailing relevant elective courses, specializations and projects related to the internship topic