Paolo Gamba
ZI] mm, +1 pn; +1
Sj 7 E (1)
where z = m;x+n; is the generic segment equation. We will see in the following subsection that this choice could be improved.
7
6
5
4
3
2
1
0
D 2.5 5 75 10 19.8 15 17.5 02.5 5 75 10 12.8 15 17%
(a) (b)
Oo HWE nds
= hs 0 al N A
= nu H5 (AN A
e
(f)
Figure 1. An example of data analysis by the proposed grouping algorithm: (a) the original data, (b) the
segment extracted, (c) initial plane seed detected, (d) planes after segment and point aggregation,
(e) final planar reconstructed surfaces with (f) their 3D profile.
3. The third step is a region growing algorithm, looking for other segments to be aggregated to the original seeds. After each
aggregation the parameters of the surface are updated, and a new segment search is performed. Planar patches and linear
elements are considered as belonging to the same plane if their orientations in space are sufficiently similar, i.e. if the projection
of the patch on the segment direction is sufficiently large. In this case no additional threshold is needed, since the segment
similarity required for seed detection is used also for this step. Because the value of ot is usually very high, not many segments
are aggregated, at the end of this step, especially in areas where many objects crowds or the data is affected by noise. From the
other end, a lower values of this parameters may provide a larger regularization of the detected surfaces, possibly loosing
smaller details.
Parameter Meaning Suggested value depending on
Qt, max. step between points in the same segment data vertical precision (as large as possible)
Qt, min. segment length data horizontal precision (as small as possible)
ol; min. similarity value data imprecision (between 05 and 0.8)
Oly max. distance point-plane data vertical precision (as large as possible)
Table 1. The parameters of the proposed algorithm.
314 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
r3, - "r1 TA
Pd A uà ui
