"This image was created with the assistance of DALL·E 2."
Gait Stride Length Estimation using Embedded Machine Learning.
Spatiotemporal gait parameters, e.g. the gait stride length, are the measures obtained from gait analysis. Gait analysis is vital in clinical applications like physical medicine and rehabilitation. Today, there are different systems available for gait analysis. However, these systems require an expensive laboratory setup and cannot be used in private practice, for example, by a physiotherapist. Also, monitoring individuals for a long time and during their daily activities is impossible. One of the solutions is an affordable battery-powered wearable device consisting of a microcontroller (MCU) and an Inertial Measurement Unit (IMU) sensor.
Data preprocessing and machine learning pipeline.
The project uses Edge Impulse Studio and open-source software frameworks such as TensorFlow and Keras. We use the MLOps (Machine Learning Operations) framework Weights and Biases for the experiment, metrics and version tracking.
Starting from the TRIPOD dataset and embedded machine learning, a regression model is developed for estimating the gait stride length. This model can be deployed on a wearables consisting of a MCU and an IMU-sensor. Besides, additional IMU data is collected in a gait laboratory - a written request to a Medical Ethics Committee is submitted and approved - to investigate the model's ability to adapt appropriately to unseen data. Finally, strategies are formulated to improve the machine learning model design and workflow.