Human motion assessment is key for motor-control rehabilitation. Using standardized definitions and spatiotemporal features - usually presented as a movement cycle diagram- specialists can associate kinematic measures to progress in rehabilitation therapy or motor impairment due to trauma or disease. Although devices for capturing human motion today are cheap and widespread, the automatic interpretation of kinematic data for rehabilitation is still poor in terms of quantitative performance evaluation. In this paper we present an automatic approach to extract spatiotemporal features from kinematic data and present it as a cycle diagram. This is done by translating standard definitions from human movement analysis into mathematical elements of a Switching Linear Dynamic System model. The result is a straight-forward procedure to learn a tracking model from a sample execution. This model is robust when used to automatically extract the movement cycle diagram of the same motion (the Sit-Stand-Sit, as an example) executed in different subject-specific manner such as his own motion speed.