Investigation on the Effect of Different Window Size in Segmentation for Common Sport Activity

Abstract

Signal segmentation is one of the most important processes in the activity recognition process. So far, windowing approaches is one of the commonly used segmentation technique to segment the data. The window size used to segment the data usually chosen based on the previous study and the effect of the activity recognition performance with the changes of window size is still vague and uncertain. Thus, in this study, we investigate the effect of different window size in segmentation process for common sports activity recognition. The study was conducted on ten subjects who wore a sensor from Gait Up called as Physilogic OR 4 Silver inertial sensor on their chest while performing several common sports activities such as stationary, walking, jogging, sprinting, and jumping. Three common used classifiers which are Decision Trees, k-Neighbor Nearest and Support Vector Machine were evaluated. Among the different ranges of window sizes tested, it was found that 2.5 seconds window size represents the best trade-off in recognition of common sports activity, with an obtained accuracy above 90%. From the result, it indicates that the selection of window size in segmentation process can affect the accuracy in detecting the common sports activity. The preferably employed window size in detecting the common sports activity is determined.