KC registration of real data

The following movie (click on the pictures) shows registering many frames of 2D range data measured by a SICK LMS 221 laser scanner. In our experiments we registered a total of about ten thousand scans without problem.

Green pixels are data from the current view. Red pixels are measured points from the previous 5 scans. That is, warp the world to the current view.

Notice that there are cases that the stationary 2D scene assumption is violated, when the sensor underwent 3D motion, or when there are moving cars in the scene.

Outliers due to sensor side rolling while traveling.

Outliers due to moving car in the scene.

More 3D registration results:

One example when the KC succeeded in the presence of uniformly distributed outlier, while ICP failed. Click on the images for VRML models.

rotated and outlier corrupted

Outlier corrupted reference

KC registration result