What is it about?
We compared the effect of applying various Principal Component Analysis (PCA) based dimension reduction methods, namely, Singular Value Decomposition (SVD), Kernel PCA, and Sparse PCA, on triaxial accelerometer signals for Convolutional Neural Network (CNN) based human activity recognition (HAR) with a focus on fall detection. We found that SVD was the most effective dimension reduction method for improving CNN based fall recognition accuracy.
Featured Image
Why is it important?
Human activity recognition including fall recognition requires constant prediction, thus making a small increase in recognition performance meaningful. Applying dimension reduction on triaxial accelerometer signals achieves this goal by realizing an increase in recognition performance by reducing noise. By concatenating raw triaxial acceleration signal with SVD signal, or signal magnitude vector with SVD signal, fall recognition performance can be possibly increased.
Perspectives
We investigated the usefulness of three PCA based dimension reduction methods, namely SVD, Kernel PCA, and Sparse PCA, for CNN based fall recognition by measuring their performance on three benchmark fall recognition datasets (UniMiB SHAR, SisFall, and UMAFall). Based on our study, we recommend the application of SVD on triaxial acceleration signal and concatenation of the dimension-reduced signal with the original signal for improving 1D CNN fall recognition.
Heeryon Cho
Kookmin University
Read the Original
This page is a summary of: Applying singular value decomposition on accelerometer data for 1d convolutional neural network based fall detection, Electronics Letters, January 2019, the Institution of Engineering and Technology (the IET),
DOI: 10.1049/el.2018.6117.
You can read the full text:
Resources
Contributors
The following have contributed to this page
