ML based Quantitative Movement Analysis on a Portable IoT Camera (1-2S/B)
Muscles and tendons within the muscle-tendon unit have different properties and respond differently to external loads and training stimuli. Imbalances in muscle strength or tendon stiffness, for example, can affect movement efficiency and lead to injuries. In clinical biomechanics, understanding changes in muscle and tendon properties is vital for creating effective treatment plans.
However, studying muscle and tendon mechanics during movement often requires multiple sensors such as 3D motion capture, ultrasound, electromyography, and force measurements. Obtaining this data can be challenging, particularly during fast movements like running, due to high acceleration and perspiration (falling off sensors), or when working with specific patient groups, such as children with cerebral palsy.
The goal of this work is to adapt a trained machine learning model that can estimate body positions and calculate changes in the muscle-tendon unit lengths to work on a portable, smart IoT camera. This could impact the way biomechanical measurements are performed today, as the proposed system should be easy and quick to use (e.g., time is key for testing patients in clinics) and inexpensive compared to traditional necessary hardware.
- Students will be co-supervised by the Center for Project Based Learning.
- Looking for 1-2 Semester/Master students
- Contact: Christoph Leitner, Marco Giordano (PBL)
- Machine Learning on Microcontrollers
- 10% Literature research
- 60% tinyML
- 30% Interface engineering
Detailed Task Description
This research aims to transfer an existing machine learning model and an accompanying post-processing algorithm to an ML accelerator, e.g., a Google Coral device. The accelerator will be paired with a portable IoT camera, and the algorithm's predictions will be sent via, e.g., an USB cable to a sub-computer. The model will be tested under lab conditions using data from walking or running subjects.
- Research of available ML accelerators and cameras
- Familiarization with ML accelerator
- Model quantization from tensorflow to tensorflow lite
- Deployment of the model on the accelerator and testing
- Build interfaces between IoT camera (2d video capture), the accelerator (prediction) and the sub-computer (display)
- Testing of the implemented pipeline on real data