A Novel Approach to Classify Human-Motion in Smart Phone using 2D-Projection Method

Yi Suk Kwon, Yeon Sik Noh, Ja Woong Yoon, Sung Bin Park, and Hyung Ro Yoon

Keywords

Smart phone, Acceleration, Orientation, Projection, Humanmotion classification

Abstract

Classification of human motion is important for measuring the intensity of physical activities. Although the accelerometer in the smart phone has been used widely for classification, phone’s various orientations lead to erroneous wrong classification of human motions during physical activity. Hence, in this paper, we present a 2D-projection method to infer users’ horizontal (forward) and vertical (upward) acceleration from the phones’ roll angle and horizontal and vertical acceleration. To validate our method, we collected acceleration and orientation data for six common physical activities, such as sitting, standing, walking, running, climbing-up stairs and climbing-down stairs. We then used the 2D-projection to transform phones’ horizontal and vertical acceleration into users’ horizontal (forward) and vertical (upward) acceleration. In the results of decision tree classification, we find that using acceleration based on users improves performance over using acceleration based on phone, while the classification of climbing-up stairs and climbing-down stairs is improved somewhat but is still insufficient. Furthermore, the number of misclassification between two activities with similar horizontal (forward) acceleration, such as walking and climbing-up or climbing-down stairs was reduced by using 2D-projection.

Important Links:



Go Back