1-1-9
sequential no.. When doing pair-wise registration, 23 pairs
of views were succeeded (shown in solid lines), while 6
pairs failed (shown in dotted lines). Reasons for failure are
® lack of linear features for registration; e.g. building
surfaces are hidden under large trees in view #23~#27;
© lack of common features due to occlusion, e.g. view
(#6,#7) and view (#27,#28). © Large jump between
views, e.g. view #24~#27. For those failed in automatic
registration, assign an coarse matching by manual
operation is required as the start point for fine adjustment.
It is not a difficult thing when using Z-lmages, since one
can slide or rotate a Z-lmage on another one easily by
keyboard operations. With the help of this half-automatic
registration method, the 29 views of laser range image
were successfully aligned in a common coordinate system
(Figure 11.a). However, an obvious conflicting situation
exists between view#15 and view#16 because of error
accumulation. Minimizing the violation to local matching is
shown in Figure 12 and Figure 13. The final result of
multiple views’ registration is shown in Figure 11.b.
Figure 10. Pair-wise registering 29 views of laser range
image of the buildings in IIS, Univ. of Tokyo. Star marks
stand for the location of viewpoint, characters for the
sequential no. of view. Solid line stands for a successful
pair wise registration, while dotted lines for failed.
(a) Aligning multiple views by pair-wise registration to
the sensor’s coordinate system of view #1
(b) Minimizing the violations to local matching
Figure 11. Result of multiple views’ registration
Figure 12. Convergence in minimizing the violation to local
matching using hill climbing method
Com parng the angfe between two successve vew pars w ith
paiw ise registration result
sequential no.
before adjistment “HK—after adjistm ent]
Figure 13. Violation to local matching before and after
multiple views’ registration