Image Processing Reference
In-Depth Information
(7)
Using the distortion coeicient estimation method, once a chessboard patern is captured,
distortion correction can be executed quickly using only the distortion coefficient. Nonethe-
less, a fine error in the components of the row and column of the chessboard was discovered
through experiment. In the case of an ultra-wide-angle camera, the calibration error becomes
larger near the edge of the image than at the center owing to the characteristics of the lens.
Thus, distortion correction using the FOV model is appropriate ( Table 1 ).
Table 1
Measurement of Distortion Center Estimation Result in the AVM Camera
Cod x Cod y
CAM #1 10
− 9
CAM #2 10
0
CAM #3 4
3
CAM #4 − 4
7
3.4 Distortion Center Estimation Method Using 2D Patterns
The distortion center estimation method using 2D paterns is proposed to overcome the limit-
ation of the distortion correction method using the distortion coefficient ω of the FOV model.
In this chapter, a chessboard patern was shot to estimate the center of distortion followed by
projecting it onto an actual straight line in the image data of corrected distortion, thereby es-
timating the center of distortion using the value of the distance difference between the straight
lines.
In order to estimate the center of distortion, irst the distortion in the 2D paterns was cor-
rected using the distortion coefficient of the FOV model. Then, a certain range surrounding
the center of distortion, which was determined while applying the FOV model, was set to the
detection window of the center of distortion.
Next, straight lines are generated by following the detected points using the method de-
scribed in Ref. [ 10 ] in the detection window of the center of distortion. Figure 3 shows the
straight lines used to compute the distortion error and detection window in which the dis-
tortion center is expected. In the chessboard patern, where a number of crossing points of
straight lines can be found; crossing points of M × N representing every corner point are
found, and then vertical M straight lines and horizontal N straight lines using the main out-
er points are generated based on the outermost points, thereby determining whether crossing
points are present in the center within the straight lines. The smaller the vertical and hori-
zontal distortions are, the closer the point is to the center of distortion, which also means it is
 
Search WWH ::




Custom Search