Given the pose of two images, how can I find right correspondences as many as possible?

edit

well, I want to get the ponit cloud of hundards of pictures of a scene. In fact, I can get the point cloud by using the bundler or visualSFM. But I want to use the BRISK descriptors. But since the BRISK is the binary one, it is not as robust as SIFT.I may fail to get the point cloud directly becasue there are less detected correspondences. So my plan is to use the visualSFM to get the pose of every image, and then given the pose, find the right correspondences of every two images, than reconstruct the point cloud descriped by the BRISK descriptors. since given the pose, it will be easy to find correspondences, but how can i find the right correspondences as many as possible.