Grid-ICP: registration for multiple-line-based scan for endoscope

We have developed multiple-line-based scanning technique for endoscope. In those works, a large problem is the degree of reconstruction instability due to image blurring caused by the strong subsurface scattering common to internal tissue. To reduce this instability problem, using a coarse pattern for structured light is an option, however it reduces the resolution of the acquired shape information. In this work, we tackle these shortcomings by developing a new ICP algorithm to increase shape resolution of line-based grid pattern.

As shown in the above figure, we propose to use a multiple shape data registration technique for the grid-structured shapes, which are acquired sequentially by small motions. The key idea is that we only calculate shape differences between the recovered shapes of different line directions, that is vertical and horizontal direction. The sum of the distance is minimized by ICP algorithm. Results of simulation and real objects are shown below. It is confirmed that our grid-ICP has more even and distributed registration results than general ICP method.

Simulation results using Stanford bunny data.

Results of real objects scanned by grid-structured light technique.

Followings are the real registration results using endoscope system. We can also confirm that our grid ICP can produce more smooth and even results.



Publications
Kawasaki Laboratory