In this paper we have proposed a visual tracking strategy for 3D object recognition and tracking in robot vision applications. We adopt a Coarse-to-Fine combination strategy for 3D object tracking. In the coarse step, compared with those edge-based only methods, the convergence range of initial camera pose estimation has been enlarged in our system by a template-based matching between a series of template images rendered from Computer Graphics (CG) and a current image. In the fine step an edge-based object tracking method is used to realize a more accurate visual tracking application with the results from the previous coarse step. The implementation of this strategy is introduced in detail. The experiments results show that this combination strategy is effective for tracking real objects in robot vision applications.