Recognizing Driver Activities Using Deep Learning Approaches Based on Smartphone Sensors
Issued Date
2022-01-01
Resource Type
ISSN
03029743
eISSN
16113349
Scopus ID
2-s2.0-85142749434
Journal Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume
13651 LNAI
Start Page
146
End Page
155
Rights Holder(s)
SCOPUS
Bibliographic Citation
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol.13651 LNAI (2022) , 146-155
Suggested Citation
Mekruksavanich S., Jantawong P., Hnoohom N., Jitpattanakul A. Recognizing Driver Activities Using Deep Learning Approaches Based on Smartphone Sensors. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol.13651 LNAI (2022) , 146-155. 155. doi:10.1007/978-3-031-20992-5_13 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/85117
Title
Recognizing Driver Activities Using Deep Learning Approaches Based on Smartphone Sensors
Author's Affiliation
Other Contributor(s)
Abstract
Human motion detection based on smartphone sensors has gained popularity for identifying everyday activities and enhancing situational awareness in pervasive and ubiquitous computing research. Modern machine learning and deep learning classifiers have been demonstrated on benchmark datasets to interpret people’s behaviors, including driving activities. While driving, driver behavior recognition may assist in activating accident detection. In this paper, we investigate driving behavior detection using deep learning techniques and smartphone sensors. We proposed the DriveNeXt classifier, which employs convolutional layers to extract spatial information and multi-branch aggregation transformation. This research evaluated the proposed model using a publicly available benchmark dataset that captures four activities: a driver entering/exiting and sitting/standing out of a vehicle. Classifier performance was evaluated using two common HAR indicators (accuracy and F1-score). The recommended DriveNeXt outperforms previous baseline deep learning models with the most fantastic accuracy of 96.95% and the highest F1-score of 96.82%, as shown by many investigations.