Comparing CNN and Human Crafted Features for Human Activity Recognition

Federico Cruciani, Anastasios Vafeiadis, CD Nugent, I Cleland, P McCullagh, Konstantinos Votis, Dimitrios Giakoumis, Dimitrios Tzovaras, Luke Chen, Raouf Hamzaoui

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Deep learning techniques such as Convolutional Neural Networks (CNNs) have shown good results in activity recognition. One of the advantages of using these methods resides in their ability to generate features automatically. This ability
greatly simplifies the task of feature extraction that usually requires domain specific knowledge, especially when using big data where data driven approaches can lead to anti-patterns. Despite the advantage of this approach, very little work has been undertaken on analyzing the quality of extracted features, and more specifically on how model architecture and parameters
affect the ability of those features to separate activity classes in the final feature space. This work focuses on identifying the optimal parameters for recognition of simple activities applying this approach on both signals from inertial and audio sensors. The paper provides the following contributions: (i) a comparison
of automatically extracted CNN features with gold standard Human Crafted Features (HCF) is given, (ii) a comprehensive analysis on how architecture and model parameters affect separation of target classes in the feature space. Results are evaluated using publicly available datasets. In particular, we achieved a 93.38% F-Score on the UCI-HAR dataset, using 1D CNNs with 3 convolutional layers and 32 kernel size, and a 90.5% F-Score on the DCASE 2017 development dataset, simplified for three classes (indoor, outdoor and vehicle), using 2D CNNs with 2 convolutional layers and a 2x2 kernel size.
LanguageEnglish
Title of host publicationThe 16th IEEE International Conference on Ubiquitous Intelligence and Computing
Publication statusAccepted/In press - 21 May 2019

Fingerprint

Neural networks
Feature extraction
Sensors
Deep learning
Big data

Cite this

Cruciani, F., Vafeiadis, A., Nugent, CD., Cleland, I., McCullagh, P., Votis, K., ... Hamzaoui, R. (Accepted/In press). Comparing CNN and Human Crafted Features for Human Activity Recognition. In The 16th IEEE International Conference on Ubiquitous Intelligence and Computing
Cruciani, Federico ; Vafeiadis, Anastasios ; Nugent, CD ; Cleland, I ; McCullagh, P ; Votis, Konstantinos ; Giakoumis, Dimitrios ; Tzovaras, Dimitrios ; Chen, Luke ; Hamzaoui, Raouf. / Comparing CNN and Human Crafted Features for Human Activity Recognition. The 16th IEEE International Conference on Ubiquitous Intelligence and Computing. 2019.
@inproceedings{1791cb5fd98e430991fbdd9240396fb0,
title = "Comparing CNN and Human Crafted Features for Human Activity Recognition",
abstract = "Deep learning techniques such as Convolutional Neural Networks (CNNs) have shown good results in activity recognition. One of the advantages of using these methods resides in their ability to generate features automatically. This abilitygreatly simplifies the task of feature extraction that usually requires domain specific knowledge, especially when using big data where data driven approaches can lead to anti-patterns. Despite the advantage of this approach, very little work has been undertaken on analyzing the quality of extracted features, and more specifically on how model architecture and parametersaffect the ability of those features to separate activity classes in the final feature space. This work focuses on identifying the optimal parameters for recognition of simple activities applying this approach on both signals from inertial and audio sensors. The paper provides the following contributions: (i) a comparisonof automatically extracted CNN features with gold standard Human Crafted Features (HCF) is given, (ii) a comprehensive analysis on how architecture and model parameters affect separation of target classes in the feature space. Results are evaluated using publicly available datasets. In particular, we achieved a 93.38{\%} F-Score on the UCI-HAR dataset, using 1D CNNs with 3 convolutional layers and 32 kernel size, and a 90.5{\%} F-Score on the DCASE 2017 development dataset, simplified for three classes (indoor, outdoor and vehicle), using 2D CNNs with 2 convolutional layers and a 2x2 kernel size.",
author = "Federico Cruciani and Anastasios Vafeiadis and CD Nugent and I Cleland and P McCullagh and Konstantinos Votis and Dimitrios Giakoumis and Dimitrios Tzovaras and Luke Chen and Raouf Hamzaoui",
year = "2019",
month = "5",
day = "21",
language = "English",
booktitle = "The 16th IEEE International Conference on Ubiquitous Intelligence and Computing",

}

Cruciani, F, Vafeiadis, A, Nugent, CD, Cleland, I, McCullagh, P, Votis, K, Giakoumis, D, Tzovaras, D, Chen, L & Hamzaoui, R 2019, Comparing CNN and Human Crafted Features for Human Activity Recognition. in The 16th IEEE International Conference on Ubiquitous Intelligence and Computing.

Comparing CNN and Human Crafted Features for Human Activity Recognition. / Cruciani, Federico; Vafeiadis, Anastasios; Nugent, CD; Cleland, I; McCullagh, P; Votis, Konstantinos; Giakoumis, Dimitrios; Tzovaras, Dimitrios; Chen, Luke; Hamzaoui, Raouf.

The 16th IEEE International Conference on Ubiquitous Intelligence and Computing. 2019.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Comparing CNN and Human Crafted Features for Human Activity Recognition

AU - Cruciani, Federico

AU - Vafeiadis, Anastasios

AU - Nugent, CD

AU - Cleland, I

AU - McCullagh, P

AU - Votis, Konstantinos

AU - Giakoumis, Dimitrios

AU - Tzovaras, Dimitrios

AU - Chen, Luke

AU - Hamzaoui, Raouf

PY - 2019/5/21

Y1 - 2019/5/21

N2 - Deep learning techniques such as Convolutional Neural Networks (CNNs) have shown good results in activity recognition. One of the advantages of using these methods resides in their ability to generate features automatically. This abilitygreatly simplifies the task of feature extraction that usually requires domain specific knowledge, especially when using big data where data driven approaches can lead to anti-patterns. Despite the advantage of this approach, very little work has been undertaken on analyzing the quality of extracted features, and more specifically on how model architecture and parametersaffect the ability of those features to separate activity classes in the final feature space. This work focuses on identifying the optimal parameters for recognition of simple activities applying this approach on both signals from inertial and audio sensors. The paper provides the following contributions: (i) a comparisonof automatically extracted CNN features with gold standard Human Crafted Features (HCF) is given, (ii) a comprehensive analysis on how architecture and model parameters affect separation of target classes in the feature space. Results are evaluated using publicly available datasets. In particular, we achieved a 93.38% F-Score on the UCI-HAR dataset, using 1D CNNs with 3 convolutional layers and 32 kernel size, and a 90.5% F-Score on the DCASE 2017 development dataset, simplified for three classes (indoor, outdoor and vehicle), using 2D CNNs with 2 convolutional layers and a 2x2 kernel size.

AB - Deep learning techniques such as Convolutional Neural Networks (CNNs) have shown good results in activity recognition. One of the advantages of using these methods resides in their ability to generate features automatically. This abilitygreatly simplifies the task of feature extraction that usually requires domain specific knowledge, especially when using big data where data driven approaches can lead to anti-patterns. Despite the advantage of this approach, very little work has been undertaken on analyzing the quality of extracted features, and more specifically on how model architecture and parametersaffect the ability of those features to separate activity classes in the final feature space. This work focuses on identifying the optimal parameters for recognition of simple activities applying this approach on both signals from inertial and audio sensors. The paper provides the following contributions: (i) a comparisonof automatically extracted CNN features with gold standard Human Crafted Features (HCF) is given, (ii) a comprehensive analysis on how architecture and model parameters affect separation of target classes in the feature space. Results are evaluated using publicly available datasets. In particular, we achieved a 93.38% F-Score on the UCI-HAR dataset, using 1D CNNs with 3 convolutional layers and 32 kernel size, and a 90.5% F-Score on the DCASE 2017 development dataset, simplified for three classes (indoor, outdoor and vehicle), using 2D CNNs with 2 convolutional layers and a 2x2 kernel size.

UR - http://www.smart-world.org/2019/

M3 - Conference contribution

BT - The 16th IEEE International Conference on Ubiquitous Intelligence and Computing

ER -

Cruciani F, Vafeiadis A, Nugent CD, Cleland I, McCullagh P, Votis K et al. Comparing CNN and Human Crafted Features for Human Activity Recognition. In The 16th IEEE International Conference on Ubiquitous Intelligence and Computing. 2019