Advanced

Hand Gesture Classification using Millimeter Wave Pulsed Radar

Dagasan, Eda LU (2020) MASM01 20192
Mathematical Statistics
Abstract
Millimeter wave pulsed radar has found many applications, among them hand gesture sensing, which this work has as purpose. This application has already shown good potential, [1], and here in this work robustness aspects are taken into account. A classification task was defined, where distinction should be made between
five different small hand gestures, recorded by 18 people in total. The classification was performed using 1D and 2D convolutional neural networks, that were compared to a k-Nearest Neighbour classifier as a benchmark. Two different approaches of feature extraction for the classifiers were attempted. On one hand, a spectrogram feature, holding time-frequency information, that showed very good performance; on the other hand... (More)
Millimeter wave pulsed radar has found many applications, among them hand gesture sensing, which this work has as purpose. This application has already shown good potential, [1], and here in this work robustness aspects are taken into account. A classification task was defined, where distinction should be made between
five different small hand gestures, recorded by 18 people in total. The classification was performed using 1D and 2D convolutional neural networks, that were compared to a k-Nearest Neighbour classifier as a benchmark. Two different approaches of feature extraction for the classifiers were attempted. On one hand, a spectrogram feature, holding time-frequency information, that showed very good performance; on the other hand an estimation of the sparse room impulse response using Elastic net and Wideband integrated dictionaries, with less satisfactory results. Special attention was given to the analysis of how well a classifier can perform on many people, even if their data is completely excluded from the training set. (Less)
Please use this url to cite or link to this publication:
author
Dagasan, Eda LU
supervisor
organization
course
MASM01 20192
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Hand Gesture Classification, Millimeter Wave Radar, Statistical Signal Processing, Time-Frequency Analysis, Room Impulse Response Estimation, Sparse Dictionary Learning, Classification, Supervised Learning, Convolutional Neural Networks, k-Nearest Neighbor
language
English
id
9002607
date added to LUP
2020-02-26 11:47:09
date last changed
2020-02-26 11:47:09
@misc{9002607,
  abstract     = {Millimeter wave pulsed radar has found many applications, among them hand gesture sensing, which this work has as purpose. This application has already shown good potential, [1], and here in this work robustness aspects are taken into account. A classification task was defined, where distinction should be made between
five different small hand gestures, recorded by 18 people in total. The classification was performed using 1D and 2D convolutional neural networks, that were compared to a k-Nearest Neighbour classifier as a benchmark. Two different approaches of feature extraction for the classifiers were attempted. On one hand, a spectrogram feature, holding time-frequency information, that showed very good performance; on the other hand an estimation of the sparse room impulse response using Elastic net and Wideband integrated dictionaries, with less satisfactory results. Special attention was given to the analysis of how well a classifier can perform on many people, even if their data is completely excluded from the training set.},
  author       = {Dagasan, Eda},
  keyword      = {Hand Gesture Classification,Millimeter Wave Radar,Statistical Signal Processing,Time-Frequency Analysis,Room Impulse Response Estimation,Sparse Dictionary Learning,Classification,Supervised Learning,Convolutional Neural Networks,k-Nearest Neighbor},
  language     = {eng},
  note         = {Student Paper},
  title        = {Hand Gesture Classification using Millimeter Wave Pulsed Radar},
  year         = {2020},
}