Temporal sign language analysis based on DTW and incremental model

Sign language data can be expressed as the positional changes of hands over time. Although increasing the number of hand movement sensors increases the recognition rate, the data scales become larger. In addition, each sign language data has a different duration. When large data are generated continuously, lower memory usage and a standardized form of data are necessary to be applied immediately in the recognition system. In this paper, we propose an incremental model that can discover major patterns and summarize entire multivariable time series data. First, Dynamic Time Warping (DTW) is applied to preprocess the data to align the length of the time series data with the same size. The contribution of the proposed incremental model is hidden variable detection and correlations of the data with temporal features. Our experiment exhibits much lower reconstruction error rates compared to Principal Component Analysis (PCA). Low reconstruction error rate can minimize the loss of information arising from feature extraction or the dimensional reduction of data. Then, it can be applied in the field of computer vision to compress an image, for image restoration and intelligent recognition systems.