Summary of open projects (Fall 2026)
To apply, please send an email to the contact person for the project with your CV and briefly describe your relevant experience.
Open positions for student projects for the next semester:
- AI-Powered Knee Moment Estimation: Mobile-Based Biomechanical Analysis to Predict Knee Injury Risk in Sports
Skills: Machine learning, image processing, gait kinematics.
Contact Person: Soroush Bagheri Koudakani [email protected]Details
The objective of this project is to develop a machine learning pipeline capable of estimating knee moments during change of direction (CoD) tasks using only 2D video data captured from a single mobile device (frontal view). This approach aims to provide an accessible, field-ready alternative to complex laboratory biomechanical analysis.
Background & Previous Work:
In a previous project, we successfully implemented a pipeline using MediaPipe and OpenPose to extract 2D kinematic data and joint coordinates from standard RGB videos. This project builds upon that foundation by mapping those extracted kinematics to complex kinetic variables—specifically, knee joint moments.
Data & Methodology:
- Input Data: Frontal-view videos of change of direction tasks, recorded using a single mobile device.
- Kinematic Extraction: MediaPipe/OpenPose will be utilized to track pose and joint coordinates from the mobile videos.
- Ground Truth Data: High-fidelity biomechanical data collected in a laboratory setting from 25 football players.
- Equipment used: Vicon motion capture system and force plates.
- Processing: OpenSim was used to process the laboratory data to calculate accurate joint angles and knee moments.
- Machine Learning: We will design, train, and validate machine learning models that map the video-derived 2D kinematics to the OpenSim-derived ground truth knee moments.
Key Objectives:
- Model Design: Develop a machine learning architecture suitable for translating 2D video-based kinematics into joint kinetics.
- Training & Validation: Train the model using the mobile video data against the robust 3D OpenSim ground truth dataset.
- Evaluation: Assess the model’s accuracy and reliability in predicting knee moments during dynamic change of direction tasks.
- Validating Athletic Performance: A Reliability and Validity Study of IMU and Video-Based Jump Analysis
Skills: Sensor fusion, gait analysis.
Contact Person: Soroush Bagheri Koudakani [email protected]Details
This project focuses on evaluating the reliability and validity of our previously developed algorithms for jump analysis and velocity-based training using IMU and video data.
Core Analyses:
The study will specifically investigate the following algorithms:
- Drop Jump Analysis: Evaluated using both IMU and video data.
- Squat Jump Analysis: Evaluated using both IMU and video data.
- Velocity-Based Training (VBT): Evaluated using IMU data.
Methodology & Workflow:
- Data Collection: Conduct laboratory sessions to gather comprehensive ground-truth and sensor data (IMU and video) from participants.
- Data Processing: Process the collected laboratory data through our existing, pre-developed algorithms.
- Reporting & Publication: Analyze the outcomes to determine the accuracy of the algorithms and write a scientific paper detailing the reliability and validity findings.
- Elite Performance Analytics: Large-Scale IMU-Based CMJ and DJ Assessment in Professional Football
Skills: Data preprocessing, IMU data analysis, statistical analysis.
Contact Person: Soroush Bagheri Koudakani [email protected]Details
This project involves processing, analyzing, and statistically evaluating IMU data collected during Countermovement Jump (CMJ) and Drop Jump (DJ) assessments. The dataset is extensive, covering multiple football teams and focusing on performance metrics across different player roles.
Dataset Details:
- Cohort: 16 football teams.
- Subjects: Roughly, 20 players per team, encompassing different playing roles.
- Sensor Setup: 4 IMU sensors utilized per player, placed on the: Pelvis Trunk Right Shank Left Shank.
- Movements Analyzed: Countermovement Jump (CMJ) and Drop Jump (DJ).
Project Phases:
- Data Pre-processing & Quality Assurance: Cleaning the raw IMU data. Verifying and correcting metadata and labels. Performing rigorous data quality checks to ensure reliability before analysis.
- Kinematic Analysis: Applying our previously implemented algorithms to process the IMU signals and analyze the CMJ and DJ mechanics.
- Statistical Analysis: Conducting comprehensive statistical analyses on the final extracted metrics to identify trends, compare performance across player roles, and draw meaningful biomechanical insights.
- Real-Time Image-Based Colour Assessment and Camera Calibration for a Scientific Imaging Setup
Skills: Signal processing, computer vision, mathematical modelling
Contact Person: Pierre P. Oppliger ([email protected]Details
This project focuses on developing image-based methods to assess and improve colour acquisition in a medical robotics platform.
The work aims to improve a prototype software module that detects poor illumination or colour conditions and supports reliable image processing by flagging, rejecting, or accepting images based on quality criteria. - Generalisation of Tool Segmentation in Medical Robotic Imaging
Skills: Deep learning, image processing, medical image analysis
Contact Person: Pierre P. Oppliger [email protected]Details
This project aims to extend an existing image segmentation pipeline for detecting a medical tool in a robotic experimental setup. The goal is to generalise the current system to segment multiple relevant classes, supporting more robust image understanding for medical robotics applications.
- Accuracy Assessment of RGB-Based Shape Reconstruction
Skills: Signal processing, medical robotics, experimental design, basic 3D modelling
Contact Person: Pierre P. Oppliger [email protected]Details
This project focuses on evaluating how accurately reconstructed shapes match their real counterparts in an experimental setup. The work involves designing controlled reference shapes, reconstructing them using our existing solution, and measuring the reconstruction error to identify the algorithm’s limitations and failure cases.
- Probabilistic Modelling of Arterial Geometries
Skills: Deep learning, probabilistic modelling, generative models, 3D data processing
Contact Person: Pierre P. Oppliger [email protected]Details
This project focuses on developing computational methods to model and generate realistic arterial geometries using probabilistic and deep learning approaches. The goal is to learn meaningful shape variations from vascular data and build a model that can support simulation, analysis, or data generation in medical robotics and biomedical applications.