The area of computer vision based motion capture and pose estimation has received much research interest. However, most proposed systems are tested in ideal conditions with heavily constricted movement and are thus less than ideal. Furthermore, hardware requirements are often very high with multiple cameras or magnetic sensors reducing the availability of practical solutions. This paper presents a novel stereoscopic approach to estimating the three dimensional pose of an articulated model through the use of non-contact computer vision based motion capture. The system is robust in cluttered and dynamic environments, performing real-time three dimensional model generation. Algorithms are presented for fundamental steps in computer vision based motion capture: tracking, and pose estimation, with accurate results computed.