A sliding window with a fixed size is not an effective approach for activity recognition system. Misclassifications could still happen especially for transitional activities. This is due to the fact that the length of transitional activity signals varies depending on the time to complete the activity [1], [2]. To overcome the problem, the window size is dynamically adapted during classification, based on certain characteristics in the signal, to better capture signals of different activities. As a result, a more effective window size can be selected for segmentation to achieve more accurate classification. In adaptive sliding window, the algorithm has an initial size of window used for segmentation which can be expanded dynamically to accommodate more samples if the signal is deemed longer than the current window size. The scenario is shown in Figure 1, in which windows W1’ and W3’ are the actual segmentation window expanded from W1 and W3 respectively since signals A1 and A3 are longer than initial window size. In this way, a more effective segmentation for classification can be achieved. Interested readers are referred to [3] for the details of the algorithm.
Figure 1: Activity classification with adaptive sliding window.
[1] B. Fida, I. Bernabucci, D. Bibbo, S. Conforto, and M. Schmid, “Varying behavior of different window sizes on the classification of static and dynamic physical activities from a single accelerometer,” Med. Eng. Phys., vol. 37, no. 7, pp. 705–711, Jul. 2015.
[2] O. Banos, J.-M. Galvez, M. Damas, H. Pomares, and I. Rojas, “Window Size Impact in Human Activity Recognition,” Sensors, vol. 14, no. 4, pp. 6474–6499, Apr. 2014.
[3] M. H. M. Noor, Z. Salcic, and K. I.-K. Wang, “Adaptive sliding window segmentation for physical activity recognition using a single tri-axial accelerometer,” Pervasive and Mobile Computing, Sep. 2016.