Monday, 16 December 2013

Paper: A practical approach towards accurate dense 3D depth recovery for robotic laparoscopic surgery

For soft tissue in MIS, the surface is generally smooth and continuous. Hence, the problem of occlusion and discontinuity may be negligible. The difficulty of dense depth recovery is usually due to the paucity of identifiable landmarks and view-dependent properties of the wet tissue.

In this paper, multi-resolution free-form registration based on piece-wise bilinear maps (PBM) is used. The lattice of PBM permits non-linear transitions, and can be adopted to a hierarchical implementation.

- What is the PBM?

Normalized cross correlation (NCC) is used as a similarity measure.

Temporal correspondence of surface points was obtained in order to analyse motion characteristics of the deforming surface. As the temporal non-rigid motion does not conform to the epipolar geometry, the search space is not confining to 1D. The image points can be tracked temporally in 3D by traversing the collected PBM lattice transformations from the spatial St (b/w left and right frames) and temporal Tt (b/w consecutive frames) registration

Specular highlight correction in this paper: (i) detecting the highlight using thresholding of the intensity and saturatioin, (ii) identifying the control points in the PBM lattice which have been affected, and (iii) bilinearly interpolating information from the surrounding control points to remove errors.


Paper address: