Recognizing Driver Activities Using Deep Learning Approaches Based on Smartphone Sensors

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.