Hand Written Digits Recognition based on Concatenated LSTMs

Over the last few years, contactless technology has gotten much attention because of the influence of the coronavirus (COVID-19). We focused on the hand written digits using contactless devices, Leap Motion Controller, for an information input method to accommodate the rapidly increasing demand. In this study, we verify the effectiveness of the proposed learning model that divides the subject’s hand-tracking data according to the type of fingers and learn with multiple input layer. We set two kinds of evaluation methods. The normal recognition method is that specific subject data includes training and test data. The third party recognition method is that specific subject data does include training data. The classification accuracy calculated by the proposed learning model, the normal recognition method achieves a maximum of 96.7%, and the third party recognition method achieves a maximum of 80.1%.