Abstract: To effectively track incomplete motion features, a feature-tracking algorithm capable of handling the cases of partially overlapped, self-occluded or totally disappeared features in motion capture is presented. The corresponding remedy measures are using combination of templates, extending the template area, or first forecasting the 2D image of missing feature and then regenerating it from 3D model. Experimental results demonstrate that by applying different tracking algorithms to different kinds of features, our approach can not only track solitary features accurately, but also effectively resolve the problem of tracking incomplete motion features with noises.