Dynamic sliding window method for physical activity recognition using a single tri-axial accelerometer

Previous studies on physical activity recognition have utilized various fixed window sizes for signal segmentation selected based on past experiments and hardware limitations. Specifically, there is no optimum fixed window size because it is subject to the characteristics of the activity signals. This paper presents a novel approach of activity signal segmentation for enhanced physical activity recognition. Central to the approach is that the window size could be dynamically adjusted by using signal information to determine the most effective segmentation. The approach recognizes not only well defined static and dynamic activities, but also transitional activities. The presented approach has been implemented, evaluated and compared with an existing approach and the fixed sliding window approach in a number of experiments. Results have shown that dynamic window segmentation achieved better overall accuracy of 96% in all activities considered in the experiments compared to the existing approach.