Enabling Cost Efficiency in Sensor-Based Human Activity Recognition

Thumbnail Image

Files

isbn9789526410319.pdf (3.23 MB)

URL

Journal Title

Journal ISSN

Volume Title

School of Electrical Engineering | Doctoral thesis (article-based) | Defence date: 2023-01-19
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2022

Department

Tietoliikenne- ja tietoverkkotekniikan laitos
Department of Communications and Networking

Major/Subject

Mcode

Degree programme

Language

en

Pages

74 + app. 88

Series

Aalto University publication series DOCTORAL THESES, 172/2022

Abstract

Sensor-based human activity recognition (HAR) involves artificial intelligence methods to automatically infer the class of movements a user is conducting based on sensor readings. It has recently experienced a surge in interest from both the industry and academic sectors, with applications ranging from entertainment systems to healthcare technologies. The miniaturization of motion sensors and the commercial availability of low-cost and general-purpose computing devices, such as the Arduino, are the principal factors related to hardware driving the increase in interest for HAR. From the software perspective, the recent employment of deep learning in HAR has enabled unprecedented recognition performance and facilitated the feature engineering process. However, despite these recent advances, HAR still faces limitations and challenges, particularly from the software perspective. First, since data collection is an onerous task, the data-hungry characteristic of deep learning results in a challenging process of developing activity recognition algorithms. Second, the enormous demand for computational resources in deep learning constitutes a barrier to enabling ubiquitous HAR since the deployment of HAR algorithms in resource-constrained devices like wearables is severely hindered. Additionally, the creation process of HAR systems is immensely labor-intensive, comprising several iterative sessions of designing, prototyping, deploying, and evaluating both algorithm and hardware design. In this dissertation, the focus is on developing methods to enable cost-efficient HAR from the standpoints of data collection, computational resources, and the creation process of HAR systems. First, we presented a novel solution for alleviating the degradation in performance across subjects without resorting to extensive amounts of data collection sessions with several subjects. Second, we devised novel neural network architectures to improve computational resource efficiency in both inference and training scenarios. Finally, we developed a simulation-driven platform aimed at creating HAR systems in a significantly simpler and lighter way. The results of this work establish a basis for enabling cost efficiency in HAR.

Description

Supervising professor

Xiao, Yu, Prof., Aalto University, Department of Communications and Networking, Finland

Thesis advisor

Díaz-Kommonen, Lily , Prof. Aalto University, Finland

Keywords

human activity recognition, deep learning, resource efficiency, data efficiency, simulation-driven design

Other note

Parts

  • [Publication 1]: Clayton Frederick Souza Leite and Yu Xiao. 2020. Improving Resource Efficiency of Deep Activity Recognition via Redundancy Reduction. In Proceedings of the 21st International Workshop on Mobile Computing Systems and Applications (HotMobile ’20). Association for Computing Machinery, New York, NY, USA, 33–38.
    Full text in Acris/Aaltodoc: http://urn.fi/URN:NBN:fi:aalto-202004282890
    DOI: 10.1145/3376897.3377859 View at publisher
  • [Publication 2]: Clayton Frederick Souza Leite and Yu Xiao. 2020. Improving Crosssubject Activity Recognition via Adversarial Learning. IEEE Access, vol. 8, 90542–90554.
    Full text in Acris/Aaltodoc: http://urn.fi/URN:NBN:fi:aalto-202006254134
    DOI: 10.1109/ACCESS.2020.2993818 View at publisher
  • [Publication 3]: Clayton Frederick Souza Leite and Yu Xiao. 2021. Optimal sensor channel selection for resource-efficient deep activity recognition. In Proceedings of the 20th International Conference on Information Processing in Sensor Networks (co-located with CPS-IoT Week 2021) (IPSN ’21). Association for Computing Machinery, New York, NY, USA, 371–383.
    Full text in Acris/Aaltodoc: http://urn.fi/URN:NBN:fi:aalto-202106167405
    DOI: 10.1145/3412382.3458278 View at publisher
  • [Publication 4]: Clayton Frederick Souza Leite and Yu Xiao. 2022. Resource-Efficient Continual Learning for Sensor-Based Human Activity Recognition. ACM Transactions on Embedded Computing Systems. Advance online publication.
    DOI: 10.1145/3530910 View at publisher
  • [Publication 5]: Clayton Frederick Souza Leite, Petr Byvshev, Henry Mauranen and Yu Xiao. 2022. Simulation-driven Design of Smart Gloves for Gesture Recognition. Manuscript submitted for publication

Citation

Permanent link to this item

https://urn.fi/URN:ISBN:978-952-64-1031-9

Collections

[diss] Sähkötekniikan korkeakoulu / ELEC