Sensor Fusion with Python for Machine Learning and Data Modelling
Halonen, Taito (2022)
2022
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2022112423939
https://urn.fi/URN:NBN:fi:amk-2022112423939
Tiivistelmä
Wearable sensors are becoming more common and the market for health monitoring type of consumer tech devices is growing fast. The aim of this project was to discover if a proof-of-concept type of machine learning data model can be created to recognize and classify human knee position and movements using data from sensors and machine learning practices like Support Vector Machine (SVM) algorithm. A functional model created through this study could serve as a foundation for future development of smart therapeutic leg support device. Additional use case could be to use the resulting model for subject gait analysis.
The sensor devices used in the project were wireless advanced heart sensors with accelerometer and gyroscope. Instead of monitoring heart rate, the devices were repurposed as surface electromyography (EMG) monitors which also provide accelerometer data. The devices are relatively inexpensive when compared to previously available medical professional sensor equipment. The project team generated and collected the data material based on agreed forms of subject movements.
The data processing and modelling was done using Python and open-sourced libraries as the primary development platform. Both types of raw data (ACC and EMG) provided by sensors were pre-processed and combined in effort to achieve model recognition result via sensor fusion.
The result was a collection of signal processing and analysis functions and a multiclass SVM model implementation for classification.
Conclusion was that the original data signal quality needs to be high before additional processing and feature extraction. The data processing methods proved to require considerable amount of manual work for adjustments for new movements. Based on the implementations the increased model recognition precision of sensor fusion proved challenging due to data synchronization.
The sensor devices used in the project were wireless advanced heart sensors with accelerometer and gyroscope. Instead of monitoring heart rate, the devices were repurposed as surface electromyography (EMG) monitors which also provide accelerometer data. The devices are relatively inexpensive when compared to previously available medical professional sensor equipment. The project team generated and collected the data material based on agreed forms of subject movements.
The data processing and modelling was done using Python and open-sourced libraries as the primary development platform. Both types of raw data (ACC and EMG) provided by sensors were pre-processed and combined in effort to achieve model recognition result via sensor fusion.
The result was a collection of signal processing and analysis functions and a multiclass SVM model implementation for classification.
Conclusion was that the original data signal quality needs to be high before additional processing and feature extraction. The data processing methods proved to require considerable amount of manual work for adjustments for new movements. Based on the implementations the increased model recognition precision of sensor fusion proved challenging due to data synchronization.