Scene Flow based Hand Tracking using Deep Learning

Abstract

Tracking of human hand is an integral part of many key computer vision systems with diverse applications such as human computer interaction, gesture recognition and augmented reality etc. However this task is very challenging due to the non rigid nature of hands, heavy occlusions due to interaction and it’s smaller size relative to the camera frame. Sceneflow estimation is also one of the key algorithms used in many applications such as image processing, navigation and visual surveillance etc. This task is also very difficult due to the one-to-many correspondence prob lem.In this project, we will develop a deep learning based system for estimation of sceneflow and extend it to solve the hand tracking problem.