Sign Language Recognition Using Modified Convolutional Neural Network Model

Sign Language is an interesting topic and similar to Action Recognition. Especially along with the great development of Deep Learning. Video-based Sign Language Recognition is our concern because we want to recognize a sign not only by the shape but also by the action the signer does. The problem is sign language is very complex and vary. The variation of sign language is making the system harder to recognize all the words accurately. Many researchers have been researching Sign Language Recognition for a long time. So many methods had been used to find out which one is the best method. Because of similarity between Sign Language Recognition and Action Recognition, we are trying to implement one of the top-tier models in Action Recognition which is i3d inception this model is also a new Action Recognition model with very high accuracy. So we can know is it possible to adopt Action Recognition behavior into Sign Language Recognition. The goal of this paper is to implement the i3d inception model to Sign Language Recognition with transfer learning method. From the test we've been done, we got 100% accuracy on training with 10 words and 10 signers with 100 classes but the validation accuracy is pretty low. This model is too overfit.