Real-Time Markerless Human Body Tracking with 3-D Voxel Reconstruction

Kinematic-based Markerless Human Tracking in 3D Probabilistic Occupancy Grids

“…It typically needs only a few iterations to generate a satisfying solution. Further details are provided in [36].…”

“…The skeletal kinematic model, adapted by the authors from the model presented by Caillette [36], is shown in Fig. 3.…”

“…Second, which is an inverse kinematic problem, the joints' parameters (joints' angles) must be determined, given the goal positions defined by the blobs in the previous step. A simple approach for the problem is the Cyclic-Coordinate Descent (CCD) algorithm, an iterative local optimization method [36]. Each joint of the skeleton is optimized in an independent way, from leaves to the root.…”

“…Since the blobs' shapes are supposed to remain the same through the tracking, the regeneration of the blobs consists in a series of rotations and translations along the skeleton kinematic tree. According to [36], lets assume that the blob B is attached at an offsetα along a bone of the skeletal model, lets denote by P the global position of this bone obtained after application of the kinematic constraints, and lets denote by R the associated rotation matrix. The corrected mean µ ′ X is computed as a simple conversion from local to global coordinates given by…”

“…The appearance model is composed of a set of associated Gaussian blobs [36]. As the human body is an articulated object composed of rigid parts, each blob models a rigid body part of the tracked person.…”