BISINDO (Bahasa Isyarat Indonesia) Sign Language Recognition Using CNN and LSTM

摘要

Sign language is one of the languages which are used to communicate with deaf people. By using it, they can communicate and understand each other. In Indonesia, there are two standards of sign language which are SIBI (Sistem Bahasa Isyarat) and BISINDO (Bahasa Isyarat Indonesia). Deep learning is a model that is used to apply to this topic. In this model, there are a lot of methods such as convolutional neural network, recurrent neural network, long-sort term memory, and each model has its characteristics. There are also some issues in deep learning by sign language recognition as the object such as data training, object position, pose, lighting, and the background of objects. This research will describe how to combine background subtraction and gaussian blur pre-processing, forwarding pre-processing background subtraction with CNN by using BISINDO, LSTM, and a combination between CNN and LSTM. In conclusion, this research shows that a combination between CNN and LSTM is the best model by explaining the accuracy and testing with sign language BISINDO as the object. The accuracy showed that for CNN 96%, LSTM 86%, and combination CNN and LSTM 96%, and the loss showed that for CNN 18%, LSTM 41%, and combination CNN and LSTM 17%.