In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Voi. XXXVIII, Part 7B 
Width in pixel 
Figure 11: Cropped version of the original image after applying Figure 12: Laser line located on the coil front extracted with an 
a median filter with differing axes scales. adaptive algorithm out of the cropped image. 
beyond the coil. So there are also laser lines projected on back 
ground geometries present. The background geometries results 
in sudden steps of the laser line which can lead to a false detec 
tions of profile defects. This problem can be reduced by cropping 
the original image to a vertical region around the estimated laser 
position. Another problem is the possible presence of edge pro 
tection material on the coil that also results in sudden steps of 
the laser line. Therefore a criterion to only extract reflections of 
the coil front is needed. The following adaptive algorithm is pro 
posed to extract only valid laser lines: 
1. In every row the position of the maximum is detected (un 
der the assumption that only one maximum per column is 
present, which is referring to the laser) and then saved in the 
variable PosMax(i). 
2. The maxima positions are normalized to the height of the 
search region PosMax(i) = PosMax(i) / Height. 
3. Setting of the start point for the adaptive algorithm in the 
middle of the laser line to become independent of variing 
coil sizes. 
polynomial. Subsequently the linear trend of the coil front can 
be calculated using Eqn. 16. The angle 6 is important for quality 
aspects and also helps to eliminate coils with packaging material, 
because the packing material results in nontypical values. For 
example a coil with packaging material is shown in Fig. 13(a) 
and the extracted laser line with the asymptotic line is shown in 
Fig. 13(b). In this case 9 is 5.7° but it should be less than 1 ° for 
unpacked coils. 
Width in pixel 
(a) Post-processed image of a 
coil with packaging material 
and a laser line. 
(b) Extracted laser line (blue) and 
the corresponding fitted line (red). 
Figure 13: Laser line extraction for a coil with packaging material 
visible. 
4. Setting of the start values for the adaptive algorithm with 
DevAv(l) — PosMax(StartPoint). 
5. Run through the maxima positions left and right of the start 
point and save positions which satisfy the following crite 
rion |PosMax(i + 1) — DevAv(i)| < a for a < 1. 
6. If the next maximum position satisfies the criterion from 
above a new comparison value is calculated for each po 
sition with 
DevAv(i + 1) = (DevAv(i) + PosMax(i + l))/2 
otherwise the laser line extraction is finished. 
Under the assumption (that the alteration of the winding displace 
ments occurs continuously) made in Sec. 1, a restrictive crite 
rion to exclude unvalid steps can be found, see point 5. The lo 
cal behaviour of the laser line is estimated according to 6 to get 
an adaptive limitation. The advantages of this algorithm are in 
its simplicity and speed of operation but still delivers good re 
sults. In Fig. 12 the result of the algorithm, the extracted laser 
line is shown. Alternatively a gray value threshold fiter can be 
used to extract the laser line but this has the disadvantage that 
unwanted objects are also included. So it is necessary to use 
post-processing to adapt the laser line by thinning and to exclude 
backgroung geometries and edge protection material. 
By using Eqn. 2 the real coil profile can be calculated out of the 
extracted laser line. For the detection of defects the first derivative 
of the coil profile is calculated. Then the profile subsegments with 
a specified raise related to the deviation and height are marked as 
critical. Additonally critical subsegments in a close neighbour 
hood are joined. The coil profile and the result for the dection of 
defects, that are larger than 2 mm, is shown in Fig. 14. 
Laser position in mm 
Figure 14: Recognized coil profile (black) and detected defects 
(red). 
5 RESULTS AND DISCUSSION 
After the laser line is extracted, the overlayed linear trend repre- To evaluate efficiency of the realized measurement system, the 
sented by k yx can be determined by approximating a first-order maximum systematic error Ah (Schriifer, 2004) is determined by