(1) Make sure the calibration is good. In Tgi3D Calibration Tool you can check the accuracy of the calibration by running calibration with “create calibration uncertainty (or accuracy) information” option. You can then increase the accuracy by going over the calibration points with the help of the "point errors" tab. You can add new points in different photographs to match existing calibration points or adjust the locations of existing points, and re-run the calibration process. In a human scale object the uncertainties of distances between calibration points should be at the order of millimeters for a good reconstruction.
(2) When applying surface fitting to photographs in SketchUP using Tgi3D SU PhotoScan plugin, the initial curves do not have to be exactly correct, they have to be close enough such that color based matching can arrive at the minimum from the given initial condition (a few pixels in the second picture). For that purpose actually, it is enough that one side of the surface is close to the real 3D, the rest of the surface will find its 3D location from that. The idea is to draw/match an easy (and possibly small) part of the object first, and then grow from there. You select the portion of the surface to be optimized, you do not want to spend optimization time on already optimized surface, but you want your surface to touch or include part of an already optimized surface.
(3) Also you can remove and put view locks and any part of the surface as needed. Do not hesitate to erase, repair, lock, unlock parts of surface as you see fit.
(4) If you have enough photographs, you can draw accurate curves using the silhouette of the object. This is sometimes useful. Imagine the silhouette of an arm from the side, you can draw the curve from front and if you have a side view you can position those curves from the side view. In fact if you have enough such silhouettes you can draw the object in 3D by using those silhouettes and simply creating and smoothing the corresponding surfaces without surface fitting to photographs, especially if you do not require a "numerically accurate" model.
(5) To draw curves on an object that does not have edges you may also make use of point-like features on the texture.
(6) One other thing that may be useful is to try to make the initial match with as few faces as possible. Do not increase resolution until you have a good match at lower resolution. Keep in mind though, if your resolution is too low you can never have a good match.
(7) Remember this is also a drawing tool, you can manually help the PhotoScan tool at any point, on any part of the surface with manual adjustments.
(8) For complex shapes some parts of the surface will remain hidden, or just have no texture on them, for those parts you can use smoothing operation. The smoothing operation guesstimates the surface shape from boundary conditions and softness/sharpness of the shapes.
(9) The photographs that you use should not have fake excess resolution. That is, do not increase resolution simply by changing image size in an editor. In fact, for most cameras you can decrease the resolution, decreasing the run time for PhotoScan.
(10) If you are making a really high detail model, you may want to create parts of the model in different SketchUp files, like head in one file, torso in another, legs in other etc, and then join them. SketchUp handles small polygon count files much better, try to remain under 5000-10000 faces per file, for example.
(3) Also you can remove and put view locks and any part of the surface as needed. Do not hesitate to erase, repair, lock, unlock parts of surface as you see fit.
(4) If you have enough photographs, you can draw accurate curves using the silhouette of the object. This is sometimes useful. Imagine the silhouette of an arm from the side, you can draw the curve from front and if you have a side view you can position those curves from the side view. In fact if you have enough such silhouettes you can draw the object in 3D by using those silhouettes and simply creating and smoothing the corresponding surfaces without surface fitting to photographs, especially if you do not require a "numerically accurate" model.
(5) To draw curves on an object that does not have edges you may also make use of point-like features on the texture.
(6) One other thing that may be useful is to try to make the initial match with as few faces as possible. Do not increase resolution until you have a good match at lower resolution. Keep in mind though, if your resolution is too low you can never have a good match.
(7) Remember this is also a drawing tool, you can manually help the PhotoScan tool at any point, on any part of the surface with manual adjustments.
(8) For complex shapes some parts of the surface will remain hidden, or just have no texture on them, for those parts you can use smoothing operation. The smoothing operation guesstimates the surface shape from boundary conditions and softness/sharpness of the shapes.
(9) The photographs that you use should not have fake excess resolution. That is, do not increase resolution simply by changing image size in an editor. In fact, for most cameras you can decrease the resolution, decreasing the run time for PhotoScan.
(10) If you are making a really high detail model, you may want to create parts of the model in different SketchUp files, like head in one file, torso in another, legs in other etc, and then join them. SketchUp handles small polygon count files much better, try to remain under 5000-10000 faces per file, for example.
This comment has been removed by a blog administrator.
ReplyDelete