Markerless 3D Human Pose Estimation and Tracking based on RGBD Cameras

We present a comparative experimental evaluation of three methods that estimate the 3D position, orientation and articulation of the human body from markerless visual observations obtained by RGBD cameras. The evaluated methods are representatives of three broad 3D human pose estimation/tracking methods. Specifically, the first is the discriminative approach adopted by OpenNI. The second is a hybrid approach that depends on the input of two synchronized and extrinsically calibrated RGBD cameras. Finally, the third one is a recently developed generative method that depends on input provided by a single RGBD camera. The experimental evaluation of these methods has been based on a publicly available data set that is annotated with ground truth. The obtained results expose the characteristics of the three methods and provide evidence that can guide the selection of the most appropriate one depending on the requirements of a certain application domain.