The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
431
candidate for matching. In addition, both candidates for
matching have similar length and orientation related to the
projected LiDAR straight line 5. Figure 4 also shows that the
methodology did not find five (26%) correspondences (false
negatives). All false negative cases are related to either the
absence of candidates (i.e., the nearby LiDAR straight lines 11
and 17) or the presence of invalid candidates (i.e., the nearby
LiDAR straight lines 4, 13, and 15).
Figure 2. Test building
Figure 4. Matching result
Figure 3. Projected LiDAR straight lines
Figure 3 shows the projected LiDAR straight lines obtained
through the projection of the 3D roof contour. The resulting
polygon is relatively close to the building roof edges, as a small
registration error of about 5-pixel maximum is present. This
largest registration error occurs with the LiDAR straight line 5.
However, the LiDAR straight lines 13 and 15 do not aproximate
correctly the details that are nearby them.
Figure 4 shows that twenty-four straight lines are extracted by
the preprocessing steps. The projected straight lines have the
following number of candidate for matching: ten projected
straight lines have only one candidate; seven projected straight
lines have two candidate; and two straight lines have no
candidates. This result shows that the preprocessing steps filter
out irrelevant information properly. Straight lines that are
successfully matched to the projected LiDAR roof contour are
overlaid in white on the image. The remaining straight lines that
are rejected by the matching process appear in black. As shown
in figure 4, the methodology found fourteen (74%)
correspondences, in which thirteen (69%) are correct and one
(5%) is incorrect (false positive). Please note that the incorrect
matching occurred because the matched straight lines is nearer
to the projected LiDAR straight line 5 than the another
Figure 5. Completion result
Figure 5 presents the result of the proposed completion strategy.
Fourteen projected LiDAR straight lines are replaced by the
matched straight lines. These straight lines are potentially better
representations for the corresponding projected LiDAR straight
lines. The projected LiDAR straight lines 4, 11, 13, 15, and 17
are kept because they do not have correspondences among the
straight lines extracted by the preprocessing steps. The refined
image-space roof contour is determined by using the new
straight line grouping constructed through the above rules,
along with the topology of the projected LiDAR roof contour
polygon. Basically, the new image-space roof contour polygon