Dynamic Time Warping-Based K-Means Clustering for Accelerometer-Based Handwriting Recognition - Developing Concepts in Applied Intelligence - page 26

Then non-selected patterns are warped into the selected pattern. Then a simple
averaging is used to derive the center pattern [2]. Their approach is similar to
ours in that DTW is utilized for calculating cluster centers. But they did not
clearly show the details of their method. We present details of our method and
algorithms for deriving cluster centers using DTW.
5Con lu on
Clustering temporal patterns is not easy because temporal patterns are non-
uniform in length and discriminating features are not temporally aligned. We
proposed an algorithm for calculating centers of temporal patterns based on
DTW, which enables us to implement K-Means clustering for temporal patterns.
We validated the performance of the algorithm by utilizing it for optimizing the
number of reference patterns of a 1-NN classifier that learns by instance-based
learning. We observed, in our experiment, that reducing the number of reference
patterns down to 5% using the proposed clustering algorithm did deteriorate the
recognition performance of the classifier only marginally. On the other hand, af-
ter reducing the number of reference patterns, the classifier performed more than
10 times faster, which is a critical benefit to utilize such classifiers in resource-
bounded platforms.
Acknowledgment
The IT R&D program of MKE/IITA supported this work. [KI001836, Develop-
ment of Mediated Interface Technology for HRI].
References
1. Chen, G., Wei, Q., Zhang, H.: Discovering similar time-series patterns with fuzzy
clustering and DTW methods. In: The Proceedings of IFSA World Congress and
20th NAFIPS International Conference, pp. 2160-2164 (2001)
2. Oates, T., Firoiu, L., Cohen, P.R.: Clustering time series with hidden Markov models
and dynamic time warping. In: Proceedings of the IJCAI 1999 Workshop on Neural,
Symbolic, and Reinforcement Learning Methods for Sequence Learning (1999)
3. Leong, T.S., Lai, J., Panza, J., Pong, P., Hong, J.: Wii Want to Write: An Ac-
celerometer Based Gesture Recognition System (2009)
4. Liu, J., Wang, Z., Zhon, L., Wickramasuriya, J., Vasudevan, V.: uWave:
Accelerometer-based Personalized Gesture Recognition, TR0630-08, Rice Univer-
sity and Motorola Labs (June 2008)
5. Kela, J., Korpipaa, P., Mntyjarvi, J., Kallio, S., Savino, G., Jozzo, L., Marca, D.:
Accelerometer-based gesture control for a design environment. Personal and Ubiq-
uitous Computing 10(5), 285-299 (2006)
6. BongWhan, C., Jun-Ki, M., Sung-Bae, C.: Online gesture recognition for user in-
terface on accelerometer built-in mobile phones. In: Wong, K.W., Mendis, B.S.U.,
Bouzerdoum, A. (eds.) ICONIP 2010. LNCS, vol. 6444, pp. 650-657. Springer,
Heidelberg (2010)