by J. Demantke (IGN)
May 21st at 2pm in Byron blanc
-1- Geometric feature extraction from lidar point clouds.
We aim to provide very generic features that describe the local geometry
around each 3D point. Difficulties in choosing an optimal neighborhood for this task are
discussed, then the proposed method is explained.
Three dimensionality features are calculated on spherical neighborhoods at
various radius sizes. Based on combinations of the eigenvalues of the local structure tensor,
they describe the shape of the neighborhood, indicating whether the local
geometry is more linear (1D), planar (2D) or volumetric (3D).
A radius-selection criterion has been tested to automatically find the
optimal neighborhood radius for each point.
-2- Photorealistic 3D facade model reconstruction from terrestrial lidar
and image data.
The goal is to reconstruct a fine facade model from lidar and image data
provided by the Stereopolis (ign mobile mapping system)
despite the following obstacles: Data are incomplete and complex.
Acquisition is performed from the single road viewpoint.
in addition, in an urban environment, facades are often hidden or
associated with other objects and Parisian facades hold a large variety of
shapes. In these conditions, the proposed 3D model must be robust and generic.
The method attempts to get the best out of the two types of data:
lidar data, inherently 3d, but semi-regular and semi-dense and high
resolution images, that are less reliable and are in 2d.
In the first step, A 3D model from lidar data is calculated.
A streamed RANSAC algorithm extracts vertical rectangles.
Then, for each detected facade, a 2.5d surface is initialized from the
rectangle, and iteratively deformed to approach 3D points.
In the second step, the edges of the surface are refined with the detected
edges in images.
