Deep Learning-based Hand Pose Estimation from 2D Image

Human hands, the most significant part of a human body, are used as a healthy and secure medium in modern technology for the development of human-computer interaction. There are various technical applications like virtual reality, touch-free writing, aerial handwriting, sign language which are performed based on the user's hand gestures. The ability to recognize hand shape and motion can be the basis for understanding the hand gesture control. In the above context, we propose a deep learning technique for hand gesture estimation that identifies hand pose from input images that reveal depth information. This system uses the convolutional neural network (CNN) to detect 3D hand gestures with the help of palm joints and fingertips. We analyzed the positions of 2D joints and fingertips and determined the depth information. From the experimental results, the proposed system is able to present an accurate estimate of the depth information from 2D hand images.