逍遙文工作室

之前在OpenCV 2.4 beta玩的tutorial code是較為完整的教學程式碼，我以為還有其它鏡頭校正 (Camera Calibration)的方法，但試驗該sample code之後，發現其實校正方式一模一樣，還是達不到我的要求，不過在此依然紀錄實驗結果，以供參考。

• 範例程式：C:\OpenCV2.4beta\samples\cpp\calibration.cpp
• 環境設定：Visual Studio 2010 安裝 OpenCV 2.4 beta

我所使用的校正用影像為右邊鏡頭所拍攝的相片，可以看上頭總共有八張影像。棋盤大小為9×6。

程式找到棋盤內角總共8×5個，會標記如下圖：（點圖可放大）

根據以上八章影像所找到的棋盤內角，程式會進行校正運算，再以原本八張影像作為校正對象，其校正結果如下圖：（點圖可放大）

此程式可以使用webcam來取的影像，也可使用圖檔來取得影像。

程式使用範例：

• calibration.exe -w 8 -h 5 -s 0.025 -o camera.yml -op -oe file.xml

程式使用方法：

• -w <board_width> # the number of inner corners per one of board dimension
• -h <board_height> # the number of inner corners per another board dimension
• [-pt <pattern>] # the type of pattern: chessboard or circles’ grid
• [-n <number_of_frames>] # the number of frames to use for calibration
  # (if not specified, it will be set to the number
  # of board views actually available)
• [-d <delay>] # a minimum delay in ms between subsequent attempts to capture a next view
  # (used only for video capturing)
• [-s <squareSize>] # square size in some user-defined units (1 by default)
• [-o <out_camera_params>] # the output filename for intrinsic [and extrinsic] parameters
• [-op] # write detected feature points
• [-oe] # write extrinsic parameters
• [-zt] # assume zero tangential distortion
• [-a <aspectRatio>] # fix aspect ratio (fx/fy)
• [-p] # fix the principal point at the center
• [-v] # flip the captured images around the horizontal axis
• [-V] # use a video file, and not an image list, uses
  # [input_data] string for the video file name
• [-su] # show undistorted images after calibration
• [input_data] # input data, one of the following:
  # – text file with a list of the images of the board
  # the text file can be generated with imagelist_creator
  # – name of video file with a video of the board
  # if input_data not specified, a live view from the camera is used

file.xml全部內容：

<?xml version="1.0″?>
<opencv_storage>
<images>
a1.jpg
a2.jpg
a3.jpg
a4.jpg
a5.jpg
a6.jpg
a7.jpg
a8.jpg
</images>
</opencv_storage>

ax.jpg為想要作為校正用的影像。

camera.yml部份內容：

%YAML:1.0
calibration_time: “05/05/12 14:58:13″
nframes: 8
image_width: 1280
image_height: 1024
board_width: 8
board_height: 5
square_size: 2.5000000372529030e-002
flags: 0
camera_matrix: !!opencv-matrix
rows: 3
cols: 3
dt: d
data: [ 5.1071821593764429e+002, 0., 6.3051054207798882e+002, 0.,
5.0977660977712299e+002, 5.1000061871235062e+002, 0., 0., 1. ]
distortion_coefficients: !!opencv-matrix
rows: 5
cols: 1
dt: d
data: [ -3.1421595155282861e-001, 9.1783527643939114e-002,
-4.4915505204101361e-004, 1.2020915220247500e-004,
-1.1457381174191541e-002 ]
avg_reprojection_error: 9.3127563437128957e-001
per_view_reprojection_errors: !!opencv-matrix
rows: 8
cols: 1
dt: f
data: [ 6.95756018e-001, 5.64228654e-001, 6.18453860e-001,
7.01186538e-001, 1.06855989e+000, 1.11666107e+000,
1.17461216e+000, 1.22194541e+000 ]
# a set of 6-tuples (rotation vector + translation vector) for each view
extrinsic_parameters: !!opencv-matrix
rows: 8
cols: 6
dt: d
data: [ -3.9843277597320970e-001, 2.8861063865948992e-002,
-4.6245901196823688e-002, -1.6914597147771646e-001,

…

image_points: !!opencv-matrix
rows: 8
cols: 40
dt: “2f"
data: [ 3.44058594e+002, 7.63874634e+002, 3.76756073e+002,
7.70987976e+002, 4.13439178e+002, 7.77496582e+002,

…

看來魚眼鏡頭曲率太大，以致校正結果邊緣效果相當差，雖然中間部份彎曲已校正為筆直。

結論：此程式只適用曲率較低的鏡頭校正使用！