EVEN, Philippe
6 SEMI-AUTOMATED EDGE SEGMENT SPECIFICATION
The automatic edge segment attraction function has been integrated within a semi-automatic mode that was called segment
magnetisation. It is summarized in the flow chart of figure 5. The operator first draws an initial segment using the mouse.
A press on the mouse button defines the first segment end. As long as the button is pressed, the segment is continuously
displayed so that its second end follows the cursor. At the release of the button, an initial segment is provided, and the
automatic attraction is immediately run. If the segment is attracted towards a spurious edge segment, the operator can
click at one of the segment ends and drag it towards the desired edge. The automatic attraction is run again as soon as
the segment is released. After any automatic attraction, the operator can come back to the manual solution with a simple
undoing action.
Undo -
IN Segment - Automatic » "9 OUT
specification " A attraction
e : Operator's choice Segment .
modification
Figure 5: Flow chart of the semi-automated edge segment specification.
The semi-automated edge segment specification was evaluated using real sites images. Both the performance of the
automatic attraction function and the one of the semi-automatic mode have been evaluated. In the latter case, we recorded
the time spent to model some objects with and without the assistance.
Because of the automatic function robustness, the desired segment edge is often detected at the first go. An initial solution
is very quickly drawn. As most of segments length is less than 300 pixels, the average response time is less than one
second on a Silicon Graphics Indigo workstation. The worst case is observed when the initial segment is one of the
image diagonal. The computation time may then rise to about eight seconds. But such segments are few and the waiting
situations are rather seldom so that their impact on the semi-automatic assistance remains low.
Most frequent failures are due to the presence of a better quality edge in the neighbourhood of the desired segment. When
it happens, a simple modification of the segment to drag it away from the disturbing edge is most of the time sufficient.
Such an example is shown in figure 6. In case of repeated failures, only a manual accurate specification can solve the
problem.
Figure 6: Example of a two steps edge segment specification : a) initial image (the desired edge contour is the left side of
the pipe), b) manual coarse specification of a line segment, c) extracted segments around the initial segment, d) automatic
attraction towards the wrong edge (right side of the screw), e) manual modification to drag the segment towards the desired
edge, f) successful attraction on the correct edge.
The operating mode of the semi-automatic edge segment specification is very intuitive and requires no learning. It is
quite flexible. For instance for the virtual edges of figure (3), the assistance could be deactivated in order to avoid any
disturbance of the operators actions. Up to now, the only images where the assistance had to be deactivated because of
multiple failures are very difficult ones, featuring many piping lines made of highly specular stainless steel which produce
plenty of reflects. Using all the other images from realistic sites, the time spent to the modelling tasks is greatly reduced
thanks to the semi-automatic assistance. The few failures are quickly recovered, so that their influence on the over-all
performance is quite low.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 227