A Temporal Deep Convolutional Sequence Analysis to Track Human Activity


Raja Sekaran, Sarmela and Pang, Ying Han and Ooi, Shih Yin (2022) A Temporal Deep Convolutional Sequence Analysis to Track Human Activity. In: Postgraduate Colloquium December 2022, 1-15 December 2022, Multimedia University, Malaysia. (Unpublished)

[img] Text
22-SARMELA A P RAJA SEKARAN.pdf - Submitted Version
Restricted to Registered users only

Download (588kB)


In recent years, sensor-based human activity recognition (HAR) has gained popularity among researchers due to its potential applications (i.e., physical activity monitoring, geriatric patient monitoring, gesture and posture analysis, etc.). Generally, HAR solutions can be categorized into handcrafted feature-based (HCF) and deep learning (DL) algorithms. HCF manually extracts features using feature engineering techniques and classifies them using traditional machine learning classifiers. On the contrary, DL can extract salient underlying features without human supervision. The most common DL models in HAR are Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). In this work, we propose a temporal deep convolutional sequence model, known as Multiscale Temporal Convolutional Network (MSTCN), to classify human activities effectively. The proposed model was evaluated using a user-independent protocol where the training and testing samples do not contain samples from the same user. As a result, MSTCN achieved accuracy scores of 97.46% and 95.2% on UCI HAR and WISDM V1.

Item Type: Conference or Workshop Item (Poster)
Uncontrolled Keywords: Machine Learning
Subjects: Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions: Faculty of Information Science and Technology (FIST)
Depositing User: Ms Suzilawati Abu Samah
Date Deposited: 21 Dec 2022 02:49
Last Modified: 21 Dec 2022 02:49
URII: http://shdl.mmu.edu.my/id/eprint/10960


Downloads per month over past year

View ItemEdit (login required)