Dynamic Time Warping-Based K-Means Clustering for Accelerometer-Based Handwriting Recognition

摘要: Dynamic time warping(DTW) is widely used for accelero-meter-based gesture recognition. The basic learning strategy applied with DTW in most cases instance-based learning, where all the feature vectors extracted from labeled training patterns are stored as reference pattern matching. With brute-force number of a class increases easily to big number. A smart generating small good needed. We propose use DTW-based K-Means clustering algorithm purpose. Initial performed by and then we apply over per so that each represented 5 ~ 10 which corresponds cluster centroid. Experiments were on 5200 sample 26 English uppercase alphabets collected 40 personals using handheld device having 3-d accelerometer inside. Results showed reducing more than 90% decreased recognition rate only 5%, while obtaining 10-times faster classification speed.