Human motion classification is an important area of computer vision with a variety of applications in surveillance, human-computer interfaces and robotics. Many current systems for human motion classification rely on a batch processing scheme to learn their classification model. This excludes these systems from many possible applications where fast response to new classes of stimuli is necessary. This thesis will present two approaches for incremental online classification of human motion, which will allow the system to adapt to new situations on the fly, without the need to go through the whole batch learning process again. The developed algorithms are tested against a current state of the art offline classification system, which already has shown good results on several human motion databases. It will be shown, that the developed online classification systems can archive competitive results while avoiding several limitations of the offline approaches.