GPU-Accelerated Real-Time Tracking of Full-Body Motion With Multi-Layer Search

Optimal power flow based on parallel metaheuristics for graphics processing units

“…A maximum speedup of 7.47× was measured. A second parallel implementation of the PSO for GPU is presented in [25] for motion tracking. The program uses a master-slave approach and runs the algorithm on the CPU while the evaluation of the fitness function is offloaded to the GPU offering a speedup over 30×.…”

Parallelization strategies for markerless human motion capture

“…PSO) can be inserted. In [45,46], two layers of search, with an efficient GPU implementation, support robust and accurate pose recovery: a sampling algorithm with a weak dynamical model introducing a non-parameter niching technique into the particle filter and a hierarchical local optimization to refine the estimation of sampling. Body pose tracking is performed in 3D space using 3D data reconstructed at every frame.…”

“…Finally, the work of Zhang et al [46] proposes an evaluation strategy based on a volumetric reconstruction. The authors design a system that employs GPUs to speed up several steps of the evaluation process.…”

Sparse Pose Regression via Componentwise Clustering Feature Point Representation

“…Human pose estimation, which is referred to as the problem of localization of the centers of body components (e.g., head, palms, feet) and the skeleton joints between adjacent components (e.g., elbows, knees), provides fundamental data for such research work. Although markerless pose estimation from color images has been widely investigated (see [2], [17], [25], [22], [28], [4], [19] and the vast literature cited therein), color images suffer from high ambiguity due to incomplete data caused by illumination and noise, as well as the insufficient description of features of intensity channels. With the development of 3D range imaging devices such as laser scanner, Kinect sensor and Time-of-Flight cameras, range images are widely used for pose estimation.…”