OpenCV 的 Rational 模型不返回 8 个失真系数

Question

我有以下脚本来校准广角镜头相机的一组图像，但我无法获得所需的所有8个失真系数。

它返回一个由8个零组成的数组，当作为参数传递和不作为参数传递时，我得到了正常的5个失真系数。

import numpy as np
import cv2
import glob
chessboard_flags = cv2.CALIB_CB_ADAPTIVE_THRESH + cv2.CALIB_CB_FAST_CHECK + cv2.CALIB_CB_NORMALIZE_IMAGE
subpix_criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.01)
CHESSBOARD_SIZE = (6, 9)
calibration_flags = cv2.CALIB_RATIONAL_MODEL
term_criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 1e-6)
obj_p = np.zeros((1, CHESSBOARD_SIZE[0] * CHESSBOARD_SIZE[1], 3), np.float32)
obj_p[0, :, :2] = np.mgrid[0:CHESSBOARD_SIZE[0], 0:CHESSBOARD_SIZE[1]].T.reshape(-1, 2)
# Arrays to store object points and image points from all the images.
objpoints = []  # 3d point in real world space
imgpoints = []  # 2d points in image plane.
images = glob.glob(r'Images/*.png')
found = 0
for fname in images:
img = cv2.imread(fname)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, corners = cv2.findChessboardCorners(gray, CHESSBOARD_SIZE, chessboard_flags)
if ret:
objpoints.append(obj_p)
corners2 = cv2.cornerSubPix(gray, corners, (3, 3), (-1, -1), subpix_criteria)
imgpoints.append(corners2)
img = cv2.drawChessboardCorners(img, CHESSBOARD_SIZE, corners2, ret)
found += 1
print("Number of images used for calibration: ", found)
mtx = np.zeros((3, 3))
dist = np.zeros((8, 1))
rvec = [np.zeros((1, 1, 3), dtype=np.float64) for i in range(found)]
tvec = [np.zeros((1, 1, 3), dtype=np.float64) for i in range(found)]
# calibration
ret, _, _, _, _ = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], mtx, dist, rvec, tvec, calibration_flags,
term_criteria)
print("K = np.array(" + str(mtx.tolist()) + ")")
print("D = np.array(" + str(dist.tolist()) + ")")
print("RMS = " + str(ret))

Answer 1

用以下内容替换calibrateCamera行修复了问题。

ret, mtx, dist, _, _ = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], None, None, flags=calibration_flags,
criteria=term_criteria)