Kahn, SvenjaSvenjaKahnMalerczyk, CorneliusCorneliusMalerczykGraf, HolgerHolgerGrafBockholt, UlrichUlrichBockholt2022-03-112022-03-112011https://publica.fraunhofer.de/handle/publica/37301010.1051/bioconf/20110100046The capturing of human movements is an important step for the analysis of human skills, e.g. for sports analysis or for learning-by-demonstration tasks. In this paper we introduce a new markerless pose estimation method which estimates human poses from silhouettes. The presented numerical pose estimation algorithm adapts a non-deterministical annealing schedule for silhouette based motion capturing. The pose is estimated by numerically minimizing the differences between the silhouettes of synthesized views of a 3D avatar and the silhouettes of the real person in the camera images. The evaluation results of simulation experiments quantify the trade-off between the accuracy and the execution time of the presented algorithm.enpose estimationmotion capturingoptimization006conference paper