I am trying to insert simple 3d objects into a scene to serve as shadow casting objects for a mask of several pieces of furniture. The masked objects and shadows will then be inserted into a 3d environment. The shadow objects however do not stick to the video. The solve error is 0.14. Is the tracker not capable of this or maybe I am doing something wrong?
Most probably your objects are in wrong position relative to solved scene. How did you place them? Did you properly orient the camera and floor and made sure the 3D objects you placed were positioned correctly?
I used the reconstruction panel for the floor , origin, and y axis placement. It seems to fit, but when I create the other objects they only fit for a few frames. I thought since the camera is the only thing that moves that the camera track would be all that is needed. It would seem odd that any object tracking would be needed in this case, but maybe I’m wrong about that?
I find it helps to track certain points of the objects you’re masking, so if it’s a chair, you track where the feet touch the floor, any maybe a few other major corners of it too. Then in the Solve panel, you can select those trackers and create geometry from them. You’ll essentially have verticies of those points and can use them for reconstruction to make sure the object really is right where it needs to be. And if you don’t adjust for lens distortion, often the objects will need to be warped a bit.
If they stay in correct place for only a few frames, the problem is clearly bad placement. Its not about object tracking, it is just that if your object is not where it needs to be, it will not “stick”. For example if you hang a box from ceiling so that it floats in the air, you can orient it in a way that when looking through camera it appears to sit on the floor. But as soon as you move the camera, it will become apparent that it is floating above floor. Your case is the same: your object might appear to be in correct place in one cam position but if it is not placed correctly in 3D space (not just in screen space) it will slide.
Use solved feature positions for properly marking your relevant scene geometry and use that geo to aid placing your objects. Like ElDirector wrote. The part about deforming objects is wrong though, you must undistort your footage before solving and distort your renders for compositing, not warp geometry.
This is my first time using a moving camera track in a commercial project, but in previous work with stills and using blam, once I solved the camera, then I could model by the background image in 3d view and get correct placement or nearly so right off the bat. I got the solve error down to 0.11, select a track for the origin, another for the y axis, and three for the floor. Although the floor looks right at first glance, now I find out that if I extrude it up to make walls and a ceiling, that the angle is way off. So the solve itself is somehow incorrect, though the error is only 0.11?
Stills and moving cam are two different things. The whole era of miniature work in film is based on unableness to distinguish the actual spatial position of element in image, but there is a reason why that work always used still cameras or nodal pans. Illusion will break when stuff is not properly placed.
Solve error in itself does not show anything. It only expresses the average positional difference between 2d features and solved 3d points reprojected through solved camera to screen space. You can have low error but still a very wrong move. Not saying it is the case now, but keep this in mind. First thing to look at is always the solved fov, camera path and layout of solved points in space. If these look good, only then you can take solve error as something meaningful.