Opencv camera calibration (2): fish eye camera calibration

0. explain

Fish eye camera calibration and ordinary camera calibration processes are based on the same ,

The calibration function and parameters will be modified .

1. Code up , Including detailed notes and differences

import cv2
import numpy as np
import glob


def calib(inter_corner_shape, size_per_grid, path):
    #  Get the number of checkerboard corners 
    w, h = inter_corner_shape
    #  The world coordinate dimension is three , Align with corner dimension 
    #  The world coordinate in the calibration of ordinary camera is one dimension less than the corner coordinate , In fish eye camera calibration, less than one dimension will result in an error 
    cp_int = np.zeros((1, w * h, 3), np.float32)
    cp_int[0, :, :2] = np.mgrid[0:w, 0:h].T.reshape(-1, 2).astype(np.float64)
    cp_world = cp_int * size_per_grid
    #  Detect sub-pixel corners 
    subpix_criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.1)
    #  Calibration method 
    calibration_flags = cv2.fisheye.CALIB_RECOMPUTE_EXTRINSIC + cv2.fisheye.CALIB_CHECK_COND + cv2.fisheye.CALIB_FIX_SKEW
    #  Image size 
    img_shape = None
    #  World coordinates 
    obj_points = []
    #  Corner coordinates 
    img_points = []
    #  Get photo 
    images = glob.glob(path + '/*.jpg')
    for fname in images:
        img = cv2.imread(fname)
        #  Use the first picture to get the image size 
        if img_shape == None:
            img_shape = img.shape[:2]
        else:
            assert img_shape == img.shape[:2], "All images must share the same size."

        gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        #  Corner finding 
        ret, corners = cv2.findChessboardCorners(gray_img, (w, h),
            cv2.CALIB_CB_ADAPTIVE_THRESH + cv2.CALIB_CB_FAST_CHECK + cv2.CALIB_CB_NORMALIZE_IMAGE)
        if ret == True:
            obj_points.append(cp_world)
            #  Add sub-pixel corner information 
            cv2.cornerSubPix(gray_img, corners, (3, 3), (-1, -1), subpix_criteria)
            img_points.append(corners)
    #  Actual number of pictures 
    imgs = len(obj_points)
    #  Initializing internal parameters 、 Distortion coefficients and rotation and translation vectors 
    K = np.zeros((3, 3))
    D = np.zeros((4, 1))
    rvecs = [np.zeros((1, 1, 3), dtype=np.float64) for i in range(imgs)]
    tvecs = [np.zeros((1, 1, 3), dtype=np.float64) for i in range(imgs)]
    #  Fish eye camera calibration 
    rms, _, _, _, _ = cv2.fisheye.calibrate(
        obj_points, img_points, gray_img.shape[::-1], K, D, rvecs, tvecs, calibration_flags,
        (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 1e-6))
    #  It is used to adjust the picture size during picture correction 
    DIM = img_shape[::-1]
    print("Found " + str(imgs) + " valid images for calibration")
    print("DIM=" + str(img_shape[::-1]))
    print("K=np.array(" + str(K.tolist()) + ")")
    print("D=np.array(" + str(D.tolist()) + ")")
    return DIM, K, D


    #  Image de distortion 
    # DIM Used to resize the image 
    #  Used to adjust the field of view after distortion removal 
def superundistort(path, K, D, DIM, scale=0.6, imshow=True):
    img = cv2.imread(path)
    dim1 = img.shape[:2][::-1]  # dim1 is the dimension of input image to un-distort
    assert dim1[0] / dim1[1] == DIM[0] / DIM[1], "Image to undistort needs to have same aspect ratio as the ones used in calibration"
    if dim1[0] != DIM[0]:
        img = cv2.resize(img, DIM, interpolation=cv2.INTER_AREA)
    NewK = K.copy()
    if scale:
        NewK[(0, 1), (0, 1)] = scale * NewK[(0, 1), (0, 1)]
    map1, map2 = cv2.fisheye.initUndistortRectifyMap(K, D, np.eye(3), NewK, DIM, cv2.CV_16SC2)
    undistorted_img = cv2.remap(img, map1, map2, interpolation=cv2.INTER_LINEAR, borderMode=cv2.BORDER_CONSTANT)
    if imshow:
        cv2.namedWindow("distorted_img", cv2.WINDOW_NORMAL)
        cv2.namedWindow("undistorted_img", cv2.WINDOW_NORMAL)
        cv2.imshow("distorted_img", img)
        cv2.imshow("undistorted_img", undistorted_img)
        cv2.waitKey(0)
        cv2.destroyAllWindows()
    else:
        return undistorted_img


if __name__ == '__main__':
    inter_corner_shape = (9, 6)
    size_per_grid = 0.02
    DIM, K, D = calib(inter_corner_shape, size_per_grid, 'data/1')
    img = superundistort('data/1/img_1_0074.jpg', K, D, DIM)

2. Code link

360sorround/fisheye at main · liuweixue001/360sorround (github.com)

3. Reference link

HLearning/fisheye: fisheye image calibration (github.com)