Author: Keerthana Arunagiri Deepa
Arunagiri Deepa, Keerthana, 2024 Determination of ideal sensor placement for activity recognition using Inertial Measurement Units: A pilot study with a single participant, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
The effectiveness of activity recognition systems is highly dependent on the placement of sensors used. This project determines the optimal sensor placement for activity recognition using the Axivity AX6 Inertial Measurement Unit (IMU). The study evaluates how activity recognition accuracy varies with different sensor placements and the number of sensors used. The research aims to identify the most effective sensor placement for activity detection while performing the following four activities: sitting, reaching and grabbing, walking, and brisk walking. Brisk walking is distinguished from walking by its faster pace and higher intensity. Accelerometer data was collected from 10 different sensor positions (right wrist, left wrist, right knee, left knee, right ankle, left ankle, neck, chest, waist, and low back) using the Axivity AX6 IMU sensor, including the neck position that is novel to the research. Combinations of sensor placements were investigated in this project. The Axivity AX6 sensors were configured by connecting to the Axivity’s OMGUI software, which acts as an interface between the sensor and the computer system for data collection and retrieval. The data was collected from a healthy adult who performed the four activities five times each, over the course of five consecutive days. The collected activity data was visualized and analysed using MATLAB. Data from the four different activities was read from Excel files, concatenated, and combined into a single dataset. The data was aligned and trimmed, ensuring that the length of data from different excel files matched before concatenation. Feature extraction was performed using both time-domain (mean, standard deviation, maximum, minimum, root mean square, skewness, and kurtosis) and frequency-domain (FFT, energy, correlation) features. A multiclass Support Vector Machine (SVM) using Error-Correcting Output Codes (ECOC) was employed to classify the four different activities. The classification accuracy, which indicates the ability to distinguish between the activities, was used to identify the most effective sensor placement from the various positions considered. The model's performance was evaluated using a confusion matrix, and key metrics such as accuracy, recall, and precision for each activity. Results showed that the right wrist achieved the highest classification accuracy of 96% among single sensor positions. Combination of right wrist and low back achieved the highest accuracy of 98.3% among the combined placements of 2 sensors and the classification accuracy improved overall across all positions. With combinations of three sensor positions, there was not much difference in accuracies and therefore considering the wearability comfort, less number of sensors are preferred. Precision and recall rates provided additional insights into the classifier's performance. This research contributes valuable insights to the field of activity recognition, particularly in healthcare, sports, and rehabilitation, where accurate activity monitoring is crucial. Future work could expand to include more diverse activities and replicating the study with a clinical population, to further validate the findings and enhance the applicability of the results.
Keywords: Accelerometer, IMU, Optimal sensor placement
Subject: Medical Biotechnology thesis
Thesis type: Masters
Completed: 2024
School: College of Science and Engineering
Supervisor: Dr. David Hobbs