As cameras and especially 3D cameras got affordable within recent years they can be used by a wider range of people. Human action recognition based on 3D coordinates recorded with such devices including a depth camera made it possible to compute video data easier. Many already published papers ignore the difference of speed within the execution of the same action. This paper suggests and evaluates multiple algorithms handling that problem based on Dynamic Time Warping and compares them with regard to runtime and accuracy. An additional algorithm from Softwerk AB is analyzed, adjusted and compared. An approach for a fast and robust algorithm able to compute a massive amount of stored sequences is proposed within this paper. The new algorithm depends on further research as not all prerequisites are met yet.