Neural network-based user-independent physical activity recognition for mobile devices

Bojan Kolosnjaji; Claudia Eckert

doi:10.1007/978-3-319-24834-9_44

Neural network-based user-independent physical activity recognition for mobile devices

Bojan Kolosnjaji
, Claudia Eckert

Informatics 20 - Chair of IT Security

Technical University of Munich

Research output: Contribution to journal › Conference article › peer-review

46 Scopus citations

Abstract

Activity recognition using sensors of mobile devices is a topic of interest of many research efforts. It has been established that user-specific training gives good accuracy in accelerometer-based activity recognition. In this paper we test a different approach: offline userindependent activity recognition based on pretrained neural networks with Dropout. Apart from satisfactory recognition accuracy that we prove in our tests, we foresee possible advantages in removing the need for users to provide labeled data and also in the security of the system. These advantages can be the reason for applying this approach in practice, not only in mobile phones but also in other embedded devices.

Original language	English
Pages (from-to)	378-386
Number of pages	9
Journal	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	9375 LNCS
DOIs	https://doi.org/10.1007/978-3-319-24834-9_44
State	Published - 2015
Event	16th International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2015 - Wroclaw, Poland Duration: 14 Oct 2015 → 16 Oct 2015

Keywords

Activity recognition
Deep learning
Machine learning
Mobile sensors
Neural networks

Access to Document

10.1007/978-3-319-24834-9_44

Cite this

@article{678e9e5197e3427c846018fdde2ce5bf,

title = "Neural network-based user-independent physical activity recognition for mobile devices",

abstract = "Activity recognition using sensors of mobile devices is a topic of interest of many research efforts. It has been established that user-specific training gives good accuracy in accelerometer-based activity recognition. In this paper we test a different approach: offline userindependent activity recognition based on pretrained neural networks with Dropout. Apart from satisfactory recognition accuracy that we prove in our tests, we foresee possible advantages in removing the need for users to provide labeled data and also in the security of the system. These advantages can be the reason for applying this approach in practice, not only in mobile phones but also in other embedded devices.",

keywords = "Activity recognition, Deep learning, Machine learning, Mobile sensors, Neural networks",

author = "Bojan Kolosnjaji and Claudia Eckert",

note = "Publisher Copyright: {\textcopyright} Springer International Publishing Switzerland 2015.; 16th International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2015 ; Conference date: 14-10-2015 Through 16-10-2015",

year = "2015",

doi = "10.1007/978-3-319-24834-9\_44",

language = "English",

volume = "9375 LNCS",

pages = "378--386",

journal = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

issn = "0302-9743",

publisher = "Springer Science and Business Media Deutschland GmbH",

}

Neural network-based user-independent physical activity recognition for mobile devices. / Kolosnjaji, Bojan; Eckert, Claudia.
In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 9375 LNCS, 2015, p. 378-386.

Research output: Contribution to journal › Conference article › peer-review

TY - JOUR

T1 - Neural network-based user-independent physical activity recognition for mobile devices

AU - Kolosnjaji, Bojan

AU - Eckert, Claudia

N1 - Publisher Copyright: © Springer International Publishing Switzerland 2015.

PY - 2015

Y1 - 2015

N2 - Activity recognition using sensors of mobile devices is a topic of interest of many research efforts. It has been established that user-specific training gives good accuracy in accelerometer-based activity recognition. In this paper we test a different approach: offline userindependent activity recognition based on pretrained neural networks with Dropout. Apart from satisfactory recognition accuracy that we prove in our tests, we foresee possible advantages in removing the need for users to provide labeled data and also in the security of the system. These advantages can be the reason for applying this approach in practice, not only in mobile phones but also in other embedded devices.

AB - Activity recognition using sensors of mobile devices is a topic of interest of many research efforts. It has been established that user-specific training gives good accuracy in accelerometer-based activity recognition. In this paper we test a different approach: offline userindependent activity recognition based on pretrained neural networks with Dropout. Apart from satisfactory recognition accuracy that we prove in our tests, we foresee possible advantages in removing the need for users to provide labeled data and also in the security of the system. These advantages can be the reason for applying this approach in practice, not only in mobile phones but also in other embedded devices.

KW - Activity recognition

KW - Deep learning

KW - Machine learning

KW - Mobile sensors

KW - Neural networks

UR - https://www.scopus.com/pages/publications/84983552237

U2 - 10.1007/978-3-319-24834-9_44

DO - 10.1007/978-3-319-24834-9_44

M3 - Conference article

AN - SCOPUS:84983552237

SN - 0302-9743

VL - 9375 LNCS

SP - 378

EP - 386

JO - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

JF - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

T2 - 16th International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2015

Y2 - 14 October 2015 through 16 October 2015

ER -

Neural network-based user-independent physical activity recognition for mobile devices

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this