Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Motion Classification using Machine Learning on Embedded Devices

Ols, Viktor LU and Hagelin, Elias LU (2025) EITM01 20251
Department of Electrical and Information Technology
Abstract
Monitoring of human motion classification provides several opportunities to aid the process of rehabilitation, exercise logging and much more. In this thesis, Inertial Measurement Unit (IMU) devices positioned on different body parts are used to collect accelerometer and gyroscope data. This data is then combined to predict what gym exercise is being performed by the user. Different architectures are investigated for the task, including Convolutional Neural Networks and Recurrent Neural Networks. Recurrent Neural Networks are further investigated regarding the feasibility to run inference on embedded devices with limited computing power. The ported models achieve accuracies of 87.1%, 98.5% and 98.7% when using one, two and eight devices... (More)
Monitoring of human motion classification provides several opportunities to aid the process of rehabilitation, exercise logging and much more. In this thesis, Inertial Measurement Unit (IMU) devices positioned on different body parts are used to collect accelerometer and gyroscope data. This data is then combined to predict what gym exercise is being performed by the user. Different architectures are investigated for the task, including Convolutional Neural Networks and Recurrent Neural Networks. Recurrent Neural Networks are further investigated regarding the feasibility to run inference on embedded devices with limited computing power. The ported models achieve accuracies of 87.1%, 98.5% and 98.7% when using one, two and eight devices respectively. Simultaneously, inference takes less than 1 ms and the model size is approximately 13 KB. (Less)
Please use this url to cite or link to this publication:
author
Ols, Viktor LU and Hagelin, Elias LU
supervisor
organization
course
EITM01 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Machine learning, Embedded devices, CNN, RNN
report number
LU/LTH-EIT 2025-1071
language
English
id
9201597
date added to LUP
2025-06-18 13:23:59
date last changed
2025-06-18 13:23:59
@misc{9201597,
  abstract     = {{Monitoring of human motion classification provides several opportunities to aid the process of rehabilitation, exercise logging and much more. In this thesis, Inertial Measurement Unit (IMU) devices positioned on different body parts are used to collect accelerometer and gyroscope data. This data is then combined to predict what gym exercise is being performed by the user. Different architectures are investigated for the task, including Convolutional Neural Networks and Recurrent Neural Networks. Recurrent Neural Networks are further investigated regarding the feasibility to run inference on embedded devices with limited computing power. The ported models achieve accuracies of 87.1%, 98.5% and 98.7% when using one, two and eight devices respectively. Simultaneously, inference takes less than 1 ms and the model size is approximately 13 KB.}},
  author       = {{Ols, Viktor and Hagelin, Elias}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Motion Classification using Machine Learning on Embedded Devices}},
  year         = {{2025}},
}