Tracking a piece of paper as it moves around in a video is a matter of following some points on the video, an image can then be distorted relatively easy to match the movement of the plane.
To track camera movement you need points that are not only following points on the video but that can also be used to define a three dimensional box in front of the camera. When reconstructing camera motion you specify some points that are flat on the ground and also raised up on a wall. This provides a way to re-create the camera movement relative to the given points.
Once you have the camera movement you can then track objects that move relative to these points.
With enough points on an object you can get the same effect as an object moving in front of the camera, but in your example video you have nothing that adds height to the sheet so there is no way of re-creating the rotation of the page, the best you can do is distort a 2d image to match the distortion of the page in front of the camera.
In your screenshot you have a solve error of 18.65. A solve error less than 1.0 is considered a good track with values up to 3.0 considered possibly good enough to work with. Anything higher than 3.0 is considered a bad track.