SMARTPHONE-BASED HUMAN ACTIVITY RECOGNITION USING ARTIFICIAL INTELLIGENCE METHODS AND ORIENTATION INVARIANT FEATURES

ABSTRACT

The implementation of digital health for monitoring human health and physical activities has become one of the integral aspects of maintaining a healthy lifestyle and reducing the onset of diseases. The process of developing a physical activity monitoring system using sensors embedded in smartphones has extensive applications in elderly care, ambient assisted living, sports injury detection, and rehabilitation. Although numerous studies in human activity recognition using smartphone-based accelerometer sensors and artificial intelligence methods have been developed by various studies. However, current studies are hampered by orientation invariant problems. Therefore, current feature sets are unable to capture the changing position of smartphones especially when placed in the pocket position. To resolve orientation invariance inherent in smartphone-based sensors, this study proposed computation and fusion of different feature vectors that are independent of sensor orientation and displacement. Performance results from the implementation show an increase in performance accuracies and f-measure. The results indicated that the fusion of the feature showed higher performance when compared with the individual feature. The fusion achieved an overall accuracy of 99.40%.

Keywords: Human activity recognition, accelerometer, artificial intelligence, machine learning