The possibility to use depth information with a relatively low-cost sensors has given new perspectives to the hand tracking. Contrariwise, full DoF pose estimation approaches attempt to estimate all the kinematic parameters of the hand skeleton.

Partial means that we do not have all the kinematics parameters of the hand’s skeleton, but only rough information about the hand motion, such as position of the palm or the fingers. (Erol et al., 2007), there are two main approaches to the hand pose estimation problem: partial and full DoF. Fundamentally, the main problems encountered in the design of hand pose estimation systems include: high-dimensionality, self-occlusions, speed processing and hand speed.Īs stated in Erol et al. Despite the significant progress in this field, the problem remains still open and reveals various theoretical and practical challenges due to a large number of issues. Therefore, there is a great interest in developing markerless based solutions because of their potential to provide more natural interaction. Unfortunately, they have several disadvantages in term of casual use as they are invasive, expensive, restrict the naturalness of motion and require complex calibration and setup in order to obtain accurate measurements. They provide a complete set of measurements for all hand’s functionalities. READ FULL TEXT VIEW PDFĬurrently, the best results for capturing hand motion are obtained by means of visual markers, electro-mechanical/magnetic sensing devices, or other specifically designed hardware. Other types of depth sensor, or RGB-D camera. Without any other information such as RGB values or normal vector components.įor this reason, the proposed method can also be applied to data obtained from Proposed framework uses a 'pure' point cloud as provided by the Kinect sensor Transformation that aligns each model with respect to the input data. The pose estimation is obtained by applying a modified version of the IterativeĬlosest Point (ICP) algorithm to synthetic models to obtain the rigid The estimated pose in a frame is used for the initialization of the next one.

TheĪpproach proposed here belongs to the class of partial pose estimation where

The hand-object case isĬlearly the most challenging task having to deal with multiple tracks. Interactions are classified, have been considered. Hand using the point clouds provided by the low-cost Microsoft Kinect sensor.īoth the free-hand case, in which the hand is isolated from the surroundingĮnvironment, and the hand-object case, in which the different types of We present a novel appearance-based approach for pose estimation of a human

Kinect random dot pattern