Software "Volume Carving"

3D Surface Reconstruction of Plant Seeds by Volume Carving: Performance and Accuracies
Roussel, J. ; Geiger, F. ; Fischbach, A. ; Jahnke, S. ; and Scharr, H.

See text here below.

Example provided together with source code as supplemental material. Imaging is done using affordable hardware, i.e., a usual SLR camera (Nikon D7000, AF-S Nikkor 16–85 mm 1:3.5–5.6 GED lens at f = 85 mm) and a motorized turntable (Steinmeyer DT130-360◦-SM01) for rotating the seed. As tool keeping the seed above the turntable’s plane, we use a cut-off ball-pen-tip. A smaller tool, like in our robotic setup, allows for better seed visibility and reconstruction, however it is not easy to build. Even though the SLR is mounted on a sturdy tripod and released by a remote control, the camera center moves from image to image.
In this setup, the base of the tool is visible in each image and can be used as target for correlation-based image registration. From the calculated image shifts, projection matrices Pi are adapted accordingly.

See provided source code for more details. (A) SLR image showing a maize seed, (B,C) cropped 760×564-image with 293×100 target for cross correlation indicated in red, (D,E) mask image with borders of the user-defined reconstruction volume projected back into the image in blue, (F) reconstructed seed and tool, (G) reconstructed seed with tool removed, (H,I), reconstructed seed from a more bottom and more top view.

The Matlab and the Python implementation (zip, 13 MB)