Hand Gesture Recognition for Deaf and Dumb Using CNN Technique

The communication through Sign language is a successful way of communication for speech and hearing impaired humans. A Hand gesture recognition system to aid deaf and mute is developed using convolutional neural networks to identify the static signs of ISL (Indian Sign Language). In order to have sufficient amount of dataset, a total of 3500 static sign images of 10 (Indian sign language) static signs are gathered from various impaired humans. A total of 4 layers and 16 filters were used in Convolution Neutral Network (CNN) Architecture based on Deep learning technique. Adam optimizer has been used as the optimizer to tweak the weighs of the model is useful for reducing the loss and improving the accuracy. Model is trained in total of 15 epochs. The optimizer used to train and validate process is Stochastic Gradient Descent (SGD). The proposed model gives the maximum possible training accuracy of about 99.76%.