Nonlinear Body Pose Estimation from Depth Images

Real-Time Human Pose Tracking from Range Data

“…The goal of this experimental evaluation is (a) to verify that our system is able to track a broad range of human motion in real-time using a single depth camera, (b) to show that the system outperforms Articulated ICP [12,16] and existing ray-casting based approaches [19] on their evaluation data set in both accuracy by a large margin and (c) to analyze the practical role of individual elements incorporated in our algorithm including the ability to make use of body part detections. Articulated ICP has been implemented by deactivating the free space constraints and jump proposals in our system.…”

“…A related class of algorithms are variants of the iterative closest point (ICP) algorithm, such as, Articulated ICP (AICP) [12][13][14], Nonrigid ICP [15], or those specifically focused on depth data [16]. To the best of our knowledge, none of the existing algorithms has been shown to be robust enough to track difficult maneuvers like hand-stands, cart-wheels, or even fast full-body movements.…”

Visual Perception for Manipulation and Imitation in Humanoid Robots

“…In [Grest et al, 2005], a system is presented which uses a minimization approach, as presented in Section 3.3.1, on 3D point cloud level. As input, depth maps are calculated by a calibrated stereo system at frame rates up to 20 Hz.…”

“…In [Grest et al, 2006], the approach proposed in [Grest et al, 2005] (see Section 3.3.2) is extended to the application of monocular human motion capture. The presented method consists of two steps.…”

Human pose estimation from a single view point, real-time range sensor

“…8. As a comparison, we show results for the ICP-based algorithm described in [22]. In this ICP implementation we use a cylinder based pose model parameterized by a skeleton with fixed limb lengths and widths.…”