Diplom

Matching of 3D Sensor Data and CAD Model

The subject of this work is to bring the information from a CAD model and a range image to a match. The common representation of both data sources is a surface triangulation. The match is established on the basis of a curved surface segmentation. This work explores the different algorithms for segmentation and the surface types which can be used. Through the segmentation the CAD model and the range image are transformed to a region adjacency graph (RAG). Using graph matching (sub-graph homomorphism) the two RAGs are brought to a match. From this the pose parameters can be derived.

Preprocessing

The first step in the algorithm is data preprocessing. The original range image (a matrix of distance values) is transformed into a surface mesh (triangulation). Because of the quantity of data this mesh is reduced using mesh decimation techniques. Click on the images below to see them at  full resolution.

Region Growing

Starting from a seed region the region is iteratively expanded and a second degree polynomial surface is fit to the data. The expansion is ended when the residuals of the surface fit become too large. The images below show some regions grown on the wire frame model of a Volkswagen Beetle.

Segmentation

Processing the whole of a surface with the region growing algorithm leads o a segmentation of the surface into several surface patches. The result depends on the thresholds for the growing and the selected seed regions. However dominant regions can be segmented rather reliably (see e.g. the roof, the hood and the side in the example below).

Home

Email