3D Hand Pose Tracking from Depth Images using Deep Reinforcement Learning

Sammanfattning: Low-cost consumer depth cameras have enabled reasonable 3D hand pose trackingfrom single depth images. Such 3D hand pose tracking can be an integralpart of many computer vision applications such as gesture recognition and humanactivity tracking. However, 3D human pose tracking still remains an openresearch problem as tracking of the hand involves nonrigidity due to finger articulationin complex background scenes, and occlusion which makes tracking achallenging task. In this work, we proposed a new approach to track 3D handpose to capture both rigid and non-rigid hand gestures.The common way of hand pose tracking involves dataset of hand imageswith the corresponding ground truth of 3D hand poses and machine learningtechniques like a randomized forest or deep learning with the convolution neuralnetwork are applied to learn the mapping from the appearance of a handin an image and its pose. These methods focus on improving the ability todistinguish the target hand pose and background but overlook the problem ofinefficient search algorithms that explore the region of interest matching withthe tracking model.Recently, with the rapid success of Alpha-Go and Alpha-Zero, there has beenprogress towards using deep neural networks trained by reinforcement learning.The human level performance was achieved by this tracker to pursue the changeof target by repetitive actions controlled by the neural network model. Also,there has been a lot of research to learn the policy from raw video data incomplex RL environment. In this work, we propose to design a new methodologyto model hand pose tracking, where the rigid and nonrigid hand movementwith the state action value pair are estimated and tracked using ReinforcementLearning (RL). The hand pose tracking is done with the bounding box to localizethe gesture location. Similarly, we proposed this model can be extended to estimatethe skeleton to track the nonrigidity of the finger articulation of the hand.In overall, our proposed approach opens a new way to address the handpose tracking problem using Deep RL as a self-learning procedure. To the bestof our knowledge, our tracker is the first neural-network tracker that combinesconvolution neural networks with RL algorithms to track hand gesture.