Computer Aided Engineering Design

304 COMPUTER AIDED ENGINEERING DESIGN

(b) Number of data points used for segmentation is small, implying that the information from

much of the data is not used to assist in reliable segmentation.

(c) Detection of smooth edges is unreliable since computation of derivatives from the noisy point

cloud is error prone.

The face based approach, on the other hand, attempts to infer connected regions of points with similar
properties as belonging to the same surface (e.g. groups of neighboring points having the same
normal belong to the same plane). This method is more reliable since it works on a larger number of
points, using all the available data. The procedure is illustrated in Figure 10.11 for the same object
as in Figure 10.10. A seed point is chosen and neighboring points are checked to have the same
property as the seed point. If yes, they are added and the region around the seed point grows. Else,
the surface definition is appropriately changed only in the initial stage. After a sufficiently large
number of data points are checked to belong to a surface definition, the latter is retained. In case of
Figure 10.11(b), since the underlying surface is a plane, the definition does not change as the region
grows. The iterative evaluation ends after no more points could be found on the plane having an
upward normal. The region is shown with dark normal needles in Figure 10.11 (b). The segmented
cloud is shown in Figure 10.11(c). Another seed point is chosen from the remaining cloud and the
process continues till all points are classified. The problems associated with the face-based approach are:

(a) Choosing good seed points and surface definition for a sub-region in a cloud is often difficult.

(b) Adaptive change in the surface type has to be performed as the region grows.

(c) Bad points, if accidentally added to the region, may change the surface definition.

(d) Deciding whether or not to add to a region can be difficult since these data are susceptible to

noise.

Figure 10.11 Segmentation by face based methods: (a) point cloud of an object and a selected seed point,
(b) region identified as belonging one base surface (the cloud normal are shown in dark
needles and (c) segmented region.

Seed point Seed point

Segmented cloud

(a) (b)

(c)

In the face-based method, segmentation and surface fitting are simultaneous and that additional
surface fitting techniques may not be required. However, in edge-based methods, surface fitting can