A real time hand tracking and gesture recognition approach which can deal with dynamic backgrounds is presented. This approach is based on computer vision. It segments hand from dynamic backgrounds using the color-based and appearance-based methods. Then, it locates the hand according to marker on the hand. Last, it recognizes the gesture based on geometry constraint of hand. In comparison with the traditional approaches, it provides a good real time performance, is easy to realize, does not require a stationary camera and is not sensitive with intensity different because its gesture recognition does not depend on the templates. Moreover, an augmented assembly system using the presented approach is described. The experiment result of the augmented assembly system demonstrated the effectiveness and robustness of our approach.