From synchronized videos captured with smartphones to the three-dimensional kinematics of the lower limbs at ESA’s Vertical Treadmill Facility

written by Gioele Giuseppe Liotta  |  Master’s Thesis in Biomedical Engineering  |  September 2026

Walking seems automatic only because we are familiar with gravity. When the load changes, the gait cycle can also reorganize itself. The challenge is not only to simulate different levels of gravity, but also to reliably and noninvasively measure the resulting changes. Understanding these adaptations can contribute to the development of countermeasures for space missions and rehabilitation programs for return to Earth.

This is the central focus of my master’s thesis in Biomedical Engineering at the Polytechnic University of Turin, which I am conducting at the Austrian Space Forum (ÖWF) in Innsbruck. I am developing and validating a markerless pipeline for 3D motion capture using synchronized videos captured with smartphones. The system will enable a quantitative comparison of walking in simulated hypogravity and on Earth.

An Unusual Room for Motion Capture

ESA’s Vertical Treadmill Facility (VTF), operated by the ÖWF, is a ground-based facility in which the subject is suspended horizontally in front of a vertical treadmill. The suspension system compensates for the component of gravity parallel to the treadmill belt, while the Subject Loading System applies a controlled force perpendicular to the walking surface. By varying the load, different levels of effective gravity can be simulated.

Harnesses, back supports, cables, confined spaces, and vibrations can cause obstructions and instability in the detection of anatomical landmarks. The cameras must therefore provide complementary angles, keeping each point of interest visible in at least two synchronized views, and remain outside the structure on a stable reference point such as the ceiling.

So, I modified and 3D-printed a smartphone mount with two axes of rotation. The mount allows you to precisely position smartphones, repeat the positioning, and keep them away from the suspension system.

From Video to Biomechanical Model

OpenCap identifies two-dimensional anatomical landmarks in each synchronized view and reconstructs their 3D positions using Direct Linear Transformation (DLT). The coordinates are then mapped to a musculoskeletal model adapted to the subject’s proportions. This introduces a consistent anatomical structure—segments, joints, and constraints—and allows inverse kinematics to transform the trajectories into interpretable joint angles.

I am also developing a local interface to reprocess sessions, select the model, archive each run, check quality, and analyze kinematics and gait events.

Finding the Way Through the Data

A VTF recording may include preparation, pauses, and transitions. Directly comparing the raw signals would therefore be misleading. The analysis must isolate the usable portion of the gait, detect heel strike and toe-off, and then segment and normalize the cycles before comparing conditions.

An initial automatic detector uses the movement of the heels and forefoot relative to the pelvis. The diagnostic graphs show the selected interval and the detected events, allowing the researcher to assess their plausibility.

From a functioning pipeline to a reliable measurement

The next step is validation. The smartphone-based reconstruction must be compared with an independent system and tested after repositioning the cameras, harness, and subject. Calibration, visibility, synchronization, and processing must meet predefined acceptance criteria before interpreting differences in severity levels. The goal is to demonstrate that the system is sensitive and reliable enough to detect real variations.

What This Project Taught Me

This project has shown me that motion capture is much more than just an algorithm: the measurement chain extends from the camera mount to the final graph. A sophisticated model cannot recover a joint hidden by the harness, and a clean graph does not automatically mean a valid result. Hardware, software, biomechanical modeling, and the experimental protocol must all support one another.

That’s what makes the project so exciting: a 3D-printed support, two smartphone videos, and an open-source biomechanical model can become a research tool.