Dynamic Hand Gesture Recognition for Video playback control

Human hand gesture recognition is becoming one of the most valuable technologies in the world today given the widespread applications such as sign language recognition, robot control, home automation, and video surveillance. The two major approaches to this recognition are either based on sensors (wearable sensors) or based on vision. Despite the development of deep learning, it has become a substantially difficult task to recognise human hand gestures. This paper focuses on dynamic hand gesture recognition for controlling a video player. There are 5 five gestures - swipe left (to go back), swipe right (to fast forward), stop (to stop), swipe up (to increase volume), and swipe down (to decrease volume). In this paper, the results are demonstrated using two models - a 3D convolutional neural network (3D CNN), which gave a 97% training and 89.4% testing accuracy, and the other using a 3D convolutional network with a long-short term memory (LSTM) model to process the sequential data information, with which, a 99% training accuracy and a validation accuracy of 96% is achieved. This model is then integrated into VLC media player for controlling the application using gestures. These gestures are captured through the integrated camera, and fed into the model for recognition.