I have worked on the development of machine and deep learning methods for determining falls from inertial measurement sensor unit data while considering direction and severity. Time series signals from accelerometer and gyroscope sensors were used from the SisFall dataset. The problem was looked at both in a fall only scenario and an activity of daily living plus fall detection scenario with experimentation conducted for different cases. The methodologies developed used signal processing techniques such as filtering, resampling, augmentation, wavelets and time and frequency domain computations. Several machine and deep learning learning methods were used including Support Vector Machines, Decision Trees, Random Forests, Grandient Boosted Trees, Convolutional Neural Networks and Long Short Term Memory Networks used for classification. The work carried out was published in conferences and journals.
Determining Fall direction and severity using SVMs
Cross dataset non-binary fall detection using a ConvLSTM-attention network
