Hand Gesture Recognition using Auto Encoder with Bi-direction Long Short Term Memory
International journal of intelligent engineering and systems · 2021
Hand Gestures provide a means for a series of interactive communications with human beings who are deaf and dumb.Deep learning based models have been revolutionized for achieving human level performance through computer vision that automates, classifies and detects presence of objects in an image.Various challenges were faced through this process such as variation in the hand size, color, illumination, skin tone, view point, complex natural backgrounds etc, because of the various conditions present on the images.The present research overcomes the problem of high dimensional data and improved the classification accuracy in gesture pattern recognition for the incomplete data.Then, multilevel segmentation method and morphological model are used to segment the gesture regions and furthermore, Auto Encoder (AE) algorithm is applied to reduce the dimension of the data.Additionally, Bi-Long Short Term Memory (LSTM) classifier is applied to classify the alphabets, numbers, and video symbols.In the experimental phase, the improved AE-BiLSTM model achieved effective performance in recognition of hand gesture in terms of accuracy.Compared to the existing CNN and PCA based uni-modal feature-level fusion model, the proposed AE-BiLSTM model showed a maximum value of 99.85 % in terms of accuracy for ISL dataset and 99.75% for NUS dataset.