Towards Hand-Object Gesture Extraction from Depth Image

Hands can perform numbers of complex tasks such as eating, writing, typing and playing instruments, which include many activities like picking up, touching and a combination of different movements. As an important component, hand recognition can be applied in many areas such as human-computer interaction, sign language recognition, gaming and robot planning. Although much effort on human hand analysis has been made, recognising human hand motions using computer vision is still a challenging issue because of the complexity of kinematics, dynamics of the environment and background noises. The aim of this paper is to recognise hand activities with an object, which needs to fit the hand model accurately while manipulating objects using RGB-D cameras. Only the depth image is used to acquire the hand model effectively. Different features are combined, and random forest algorithms are used to make the proposed method robust in the current challenging dataset.